Add proper transaction handling (#550)

This commit is contained in:
Marko Budiselić 2022-09-21 18:25:51 +02:00 committed by GitHub
parent 817161a915
commit b4d6dc0930
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
62 changed files with 1360 additions and 9012 deletions

1
.gitignore vendored
View File

@ -62,6 +62,7 @@ src/durability/single_node/state_delta.hpp
src/durability/single_node_ha/state_delta.hpp
src/query/frontend/semantic/symbol.hpp
src/query/v2/frontend/semantic/symbol.hpp
src/expr/semantic/symbol.hpp
src/query/distributed/frontend/semantic/symbol_serialization.hpp
src/query/v2/distributed/frontend/semantic/symbol_serialization.hpp
src/query/distributed/plan/ops.hpp

View File

@ -11,6 +11,8 @@
#pragma once
#include <compare>
#include "io/time.hpp"
namespace memgraph::coordinator {
@ -22,7 +24,13 @@ struct Hlc {
uint64_t logical_id;
Time coordinator_wall_clock;
auto operator<=>(const Hlc &other) const { return logical_id <=> other.logical_id; }
bool operator==(const Hlc &other) const = default;
bool operator<(const Hlc &other) const = default;
bool operator==(const uint64_t other) const { return logical_id == other; }
bool operator<(const uint64_t other) const { return logical_id < other; }
bool operator>=(const uint64_t other) const { return logical_id >= other; }
};
} // namespace memgraph::coordinator

View File

@ -263,8 +263,6 @@ class DbAccessor final {
return std::nullopt;
}
void FinalizeTransaction() { accessor_->FinalizeTransaction(); }
VerticesIterable Vertices(storage::v3::View view) { return VerticesIterable(accessor_->Vertices(view)); }
VerticesIterable Vertices(storage::v3::View view, storage::v3::LabelId label) {
@ -376,7 +374,7 @@ class DbAccessor final {
void AdvanceCommand() { accessor_->AdvanceCommand(); }
utils::BasicResult<storage::v3::ConstraintViolation, void> Commit() { return accessor_->Commit(); }
void Commit() { return accessor_->Commit(coordinator::Hlc{}); }
void Abort() { accessor_->Abort(); }
@ -407,8 +405,6 @@ class DbAccessor final {
storage::v3::IndicesInfo ListAllIndices() const { return accessor_->ListAllIndices(); }
storage::v3::ConstraintsInfo ListAllConstraints() const { return accessor_->ListAllConstraints(); }
const storage::v3::SchemaValidator &GetSchemaValidator() const { return accessor_->GetSchemaValidator(); }
storage::v3::SchemasInfo ListAllSchemas() const { return accessor_->ListAllSchemas(); }

View File

@ -269,10 +269,6 @@ PullPlanDump::PullPlanDump(DbAccessor *dba)
CreateLabelIndicesPullChunk(),
// Dump all label property indices
CreateLabelPropertyIndicesPullChunk(),
// Dump all existence constraints
CreateExistenceConstraintsPullChunk(),
// Dump all unique constraints
CreateUniqueConstraintsPullChunk(),
// Create internal index for faster edge creation
CreateInternalIndexPullChunk(),
// Dump all vertices
@ -364,60 +360,6 @@ PullPlanDump::PullChunk PullPlanDump::CreateLabelPropertyIndicesPullChunk() {
};
}
PullPlanDump::PullChunk PullPlanDump::CreateExistenceConstraintsPullChunk() {
return [this, global_index = 0U](AnyStream *stream, std::optional<int> n) mutable -> std::optional<size_t> {
// Delay the construction of constraint vectors
if (!constraints_info_) {
constraints_info_.emplace(dba_->ListAllConstraints());
}
const auto &existence = constraints_info_->existence;
size_t local_counter = 0;
while (global_index < existence.size() && (!n || local_counter < *n)) {
const auto &constraint = existence[global_index];
std::ostringstream os;
DumpExistenceConstraint(&os, dba_, constraint.first, constraint.second);
stream->Result({TypedValue(os.str())});
++global_index;
++local_counter;
}
if (global_index == existence.size()) {
return local_counter;
}
return std::nullopt;
};
}
PullPlanDump::PullChunk PullPlanDump::CreateUniqueConstraintsPullChunk() {
return [this, global_index = 0U](AnyStream *stream, std::optional<int> n) mutable -> std::optional<size_t> {
// Delay the construction of constraint vectors
if (!constraints_info_) {
constraints_info_.emplace(dba_->ListAllConstraints());
}
const auto &unique = constraints_info_->unique;
size_t local_counter = 0;
while (global_index < unique.size() && (!n || local_counter < *n)) {
const auto &constraint = unique[global_index];
std::ostringstream os;
DumpUniqueConstraint(&os, dba_, constraint.first, constraint.second);
stream->Result({TypedValue(os.str())});
++global_index;
++local_counter;
}
if (global_index == unique.size()) {
return local_counter;
}
return std::nullopt;
};
}
PullPlanDump::PullChunk PullPlanDump::CreateInternalIndexPullChunk() {
return [this](AnyStream *stream, std::optional<int>) mutable -> std::optional<size_t> {
if (vertices_iterable_.begin() != vertices_iterable_.end()) {

View File

@ -32,7 +32,6 @@ struct PullPlanDump {
query::v2::DbAccessor *dba_ = nullptr;
std::optional<storage::v3::IndicesInfo> indices_info_ = std::nullopt;
std::optional<storage::v3::ConstraintsInfo> constraints_info_ = std::nullopt;
using VertexAccessorIterable = decltype(std::declval<query::v2::DbAccessor>().Vertices(storage::v3::View::OLD));
using VertexAccessorIterableIterator = decltype(std::declval<VertexAccessorIterable>().begin());
@ -55,8 +54,6 @@ struct PullPlanDump {
PullChunk CreateLabelIndicesPullChunk();
PullChunk CreateLabelPropertyIndicesPullChunk();
PullChunk CreateExistenceConstraintsPullChunk();
PullChunk CreateUniqueConstraintsPullChunk();
PullChunk CreateInternalIndexPullChunk();
PullChunk CreateVertexPullChunk();
PullChunk CreateEdgePullChunk();

View File

@ -129,135 +129,25 @@ std::optional<std::string> GetOptionalStringValue(query::v2::Expression *express
class ReplQueryHandler final : public query::v2::ReplicationQueryHandler {
public:
explicit ReplQueryHandler(storage::v3::Shard *db) : db_(db) {}
explicit ReplQueryHandler(storage::v3::Shard * /*db*/) {}
/// @throw QueryRuntimeException if an error ocurred.
void SetReplicationRole(ReplicationQuery::ReplicationRole replication_role, std::optional<int64_t> port) override {
if (replication_role == ReplicationQuery::ReplicationRole::MAIN) {
if (!db_->SetMainReplicationRole()) {
throw QueryRuntimeException("Couldn't set role to main!");
}
}
if (replication_role == ReplicationQuery::ReplicationRole::REPLICA) {
if (!port || *port < 0 || *port > std::numeric_limits<uint16_t>::max()) {
throw QueryRuntimeException("Port number invalid!");
}
if (!db_->SetReplicaRole(
io::network::Endpoint(query::v2::kDefaultReplicationServerIp, static_cast<uint16_t>(*port)))) {
throw QueryRuntimeException("Couldn't set role to replica!");
}
}
}
void SetReplicationRole(ReplicationQuery::ReplicationRole /*replication_role*/,
std::optional<int64_t> /*port*/) override {}
/// @throw QueryRuntimeException if an error ocurred.
ReplicationQuery::ReplicationRole ShowReplicationRole() const override {
switch (db_->GetReplicationRole()) {
case storage::v3::ReplicationRole::MAIN:
return ReplicationQuery::ReplicationRole::MAIN;
case storage::v3::ReplicationRole::REPLICA:
return ReplicationQuery::ReplicationRole::REPLICA;
}
throw QueryRuntimeException("Couldn't show replication role - invalid role set!");
}
ReplicationQuery::ReplicationRole ShowReplicationRole() const override { return {}; }
/// @throw QueryRuntimeException if an error ocurred.
void RegisterReplica(const std::string &name, const std::string &socket_address,
const ReplicationQuery::SyncMode sync_mode, const std::optional<double> timeout,
const std::chrono::seconds replica_check_frequency) override {
if (db_->GetReplicationRole() == storage::v3::ReplicationRole::REPLICA) {
// replica can't register another replica
throw QueryRuntimeException("Replica can't register another replica!");
}
storage::v3::replication::ReplicationMode repl_mode;
switch (sync_mode) {
case ReplicationQuery::SyncMode::ASYNC: {
repl_mode = storage::v3::replication::ReplicationMode::ASYNC;
break;
}
case ReplicationQuery::SyncMode::SYNC: {
repl_mode = storage::v3::replication::ReplicationMode::SYNC;
break;
}
}
auto maybe_ip_and_port =
io::network::Endpoint::ParseSocketOrIpAddress(socket_address, query::v2::kDefaultReplicationPort);
if (maybe_ip_and_port) {
auto [ip, port] = *maybe_ip_and_port;
auto ret = db_->RegisterReplica(
name, {std::move(ip), port}, repl_mode,
{.timeout = timeout, .replica_check_frequency = replica_check_frequency, .ssl = std::nullopt});
if (ret.HasError()) {
throw QueryRuntimeException(fmt::format("Couldn't register replica '{}'!", name));
}
} else {
throw QueryRuntimeException("Invalid socket address!");
}
}
void RegisterReplica(const std::string & /*name*/, const std::string & /*socket_address*/,
const ReplicationQuery::SyncMode /*sync_mode*/, const std::optional<double> /*timeout*/,
const std::chrono::seconds /*replica_check_frequency*/) override {}
/// @throw QueryRuntimeException if an error ocurred.
void DropReplica(const std::string &replica_name) override {
if (db_->GetReplicationRole() == storage::v3::ReplicationRole::REPLICA) {
// replica can't unregister a replica
throw QueryRuntimeException("Replica can't unregister a replica!");
}
if (!db_->UnregisterReplica(replica_name)) {
throw QueryRuntimeException(fmt::format("Couldn't unregister the replica '{}'", replica_name));
}
}
void DropReplica(const std::string & /*replica_name*/) override {}
using Replica = ReplicationQueryHandler::Replica;
std::vector<Replica> ShowReplicas() const override {
if (db_->GetReplicationRole() == storage::v3::ReplicationRole::REPLICA) {
// replica can't show registered replicas (it shouldn't have any)
throw QueryRuntimeException("Replica can't show registered replicas (it shouldn't have any)!");
}
auto repl_infos = db_->ReplicasInfo();
std::vector<Replica> replicas;
replicas.reserve(repl_infos.size());
const auto from_info = [](const auto &repl_info) -> Replica {
Replica replica;
replica.name = repl_info.name;
replica.socket_address = repl_info.endpoint.SocketAddress();
switch (repl_info.mode) {
case storage::v3::replication::ReplicationMode::SYNC:
replica.sync_mode = ReplicationQuery::SyncMode::SYNC;
break;
case storage::v3::replication::ReplicationMode::ASYNC:
replica.sync_mode = ReplicationQuery::SyncMode::ASYNC;
break;
}
if (repl_info.timeout) {
replica.timeout = *repl_info.timeout;
}
switch (repl_info.state) {
case storage::v3::replication::ReplicaState::READY:
replica.state = ReplicationQuery::ReplicaState::READY;
break;
case storage::v3::replication::ReplicaState::REPLICATING:
replica.state = ReplicationQuery::ReplicaState::REPLICATING;
break;
case storage::v3::replication::ReplicaState::RECOVERY:
replica.state = ReplicationQuery::ReplicaState::RECOVERY;
break;
case storage::v3::replication::ReplicaState::INVALID:
replica.state = ReplicationQuery::ReplicaState::INVALID;
break;
}
return replica;
};
std::transform(repl_infos.begin(), repl_infos.end(), std::back_inserter(replicas), from_info);
return replicas;
}
private:
storage::v3::Shard *db_;
std::vector<Replica> ShowReplicas() const override { return {}; }
};
/// returns false if the replication role can't be set
/// @throw QueryRuntimeException if an error ocurred.
@ -1158,8 +1048,8 @@ PreparedQuery Interpreter::PrepareTransactionQuery(std::string_view query_upper)
in_explicit_transaction_ = true;
expect_rollback_ = false;
db_accessor_ =
std::make_unique<storage::v3::Shard::Accessor>(interpreter_context_->db->Access(GetIsolationLevelOverride()));
db_accessor_ = std::make_unique<storage::v3::Shard::Accessor>(
interpreter_context_->db->Access(coordinator::Hlc{}, GetIsolationLevelOverride()));
execution_db_accessor_.emplace(db_accessor_.get());
if (interpreter_context_->trigger_store.HasTriggers()) {
@ -1581,48 +1471,22 @@ PreparedQuery PrepareReplicationQuery(ParsedQuery parsed_query, const bool in_ex
PreparedQuery PrepareLockPathQuery(ParsedQuery parsed_query, const bool in_explicit_transaction,
InterpreterContext *interpreter_context, DbAccessor *dba) {
if (in_explicit_transaction) {
throw LockPathModificationInMulticommandTxException();
}
auto *lock_path_query = utils::Downcast<LockPathQuery>(parsed_query.query);
return PreparedQuery{{},
std::move(parsed_query.required_privileges),
[interpreter_context, action = lock_path_query->action_](
AnyStream *stream, std::optional<int> n) -> std::optional<QueryHandlerResult> {
switch (action) {
case LockPathQuery::Action::LOCK_PATH:
if (!interpreter_context->db->LockPath()) {
throw QueryRuntimeException("Failed to lock the data directory");
}
break;
case LockPathQuery::Action::UNLOCK_PATH:
if (!interpreter_context->db->UnlockPath()) {
throw QueryRuntimeException("Failed to unlock the data directory");
}
break;
}
return QueryHandlerResult::COMMIT;
},
RWType::NONE};
throw SemanticException("LockPath query is not supported!");
}
PreparedQuery PrepareFreeMemoryQuery(ParsedQuery parsed_query, const bool in_explicit_transaction,
InterpreterContext *interpreter_context) {
InterpreterContext * /*interpreter_context*/) {
if (in_explicit_transaction) {
throw FreeMemoryModificationInMulticommandTxException();
}
return PreparedQuery{
{},
std::move(parsed_query.required_privileges),
[interpreter_context](AnyStream *stream, std::optional<int> n) -> std::optional<QueryHandlerResult> {
interpreter_context->db->FreeMemory();
memory::PurgeUnusedMemory();
return QueryHandlerResult::COMMIT;
},
RWType::NONE};
return PreparedQuery{{},
std::move(parsed_query.required_privileges),
[](AnyStream * /*stream*/, std::optional<int> /*n*/) -> std::optional<QueryHandlerResult> {
memory::PurgeUnusedMemory();
return QueryHandlerResult::COMMIT;
},
RWType::NONE};
}
TriggerEventType ToTriggerEventType(const TriggerQuery::EventType event_type) {
@ -1835,24 +1699,7 @@ PreparedQuery PrepareIsolationLevelQuery(ParsedQuery parsed_query, const bool in
PreparedQuery PrepareCreateSnapshotQuery(ParsedQuery parsed_query, bool in_explicit_transaction,
InterpreterContext *interpreter_context) {
if (in_explicit_transaction) {
throw CreateSnapshotInMulticommandTxException();
}
return PreparedQuery{
{},
std::move(parsed_query.required_privileges),
[interpreter_context](AnyStream *stream, std::optional<int> n) -> std::optional<QueryHandlerResult> {
if (auto maybe_error = interpreter_context->db->CreateSnapshot(); maybe_error.HasError()) {
switch (maybe_error.GetError()) {
case storage::v3::Shard::CreateSnapshotError::DisabledForReplica:
throw utils::BasicException(
"Failed to create a snapshot. Replica instances are not allowed to create them.");
}
}
return QueryHandlerResult::COMMIT;
},
RWType::NONE};
throw SemanticException("CreateSnapshot query is not supported!");
}
PreparedQuery PrepareSettingQuery(ParsedQuery parsed_query, const bool in_explicit_transaction, DbAccessor *dba) {
@ -1920,7 +1767,6 @@ PreparedQuery PrepareInfoQuery(ParsedQuery parsed_query, bool in_explicit_transa
{TypedValue("edge_count"), TypedValue(static_cast<int64_t>(info.edge_count))},
{TypedValue("average_degree"), TypedValue(info.average_degree)},
{TypedValue("memory_usage"), TypedValue(static_cast<int64_t>(info.memory_usage))},
{TypedValue("disk_usage"), TypedValue(static_cast<int64_t>(info.disk_usage))},
{TypedValue("memory_allocated"), TypedValue(static_cast<int64_t>(utils::total_memory_tracker.Amount()))},
{TypedValue("allocation_limit"),
TypedValue(static_cast<int64_t>(utils::total_memory_tracker.HardLimit()))}};
@ -1945,28 +1791,7 @@ PreparedQuery PrepareInfoQuery(ParsedQuery parsed_query, bool in_explicit_transa
};
break;
case InfoQuery::InfoType::CONSTRAINT:
header = {"constraint type", "label", "properties"};
handler = [interpreter_context] {
auto *db = interpreter_context->db;
auto info = db->ListAllConstraints();
std::vector<std::vector<TypedValue>> results;
results.reserve(info.existence.size() + info.unique.size());
for (const auto &item : info.existence) {
results.push_back({TypedValue("exists"), TypedValue(db->LabelToName(item.first)),
TypedValue(db->PropertyToName(item.second))});
}
for (const auto &item : info.unique) {
std::vector<TypedValue> properties;
properties.reserve(item.second.size());
for (const auto &property : item.second) {
properties.emplace_back(db->PropertyToName(property));
}
results.push_back(
{TypedValue("unique"), TypedValue(db->LabelToName(item.first)), TypedValue(std::move(properties))});
}
return std::pair{results, QueryHandlerResult::NOTHING};
};
break;
throw SemanticException("Constraints are not yet supported!");
}
return PreparedQuery{std::move(header), std::move(parsed_query.required_privileges),
@ -1990,185 +1815,7 @@ PreparedQuery PrepareInfoQuery(ParsedQuery parsed_query, bool in_explicit_transa
PreparedQuery PrepareConstraintQuery(ParsedQuery parsed_query, bool in_explicit_transaction,
std::vector<Notification> *notifications,
InterpreterContext *interpreter_context) {
if (in_explicit_transaction) {
throw ConstraintInMulticommandTxException();
}
auto *constraint_query = utils::Downcast<ConstraintQuery>(parsed_query.query);
std::function<void(Notification &)> handler;
const auto label = interpreter_context->NameToLabelId(constraint_query->constraint_.label.name);
std::vector<storage::v3::PropertyId> properties;
std::vector<std::string> properties_string;
properties.reserve(constraint_query->constraint_.properties.size());
properties_string.reserve(constraint_query->constraint_.properties.size());
for (const auto &prop : constraint_query->constraint_.properties) {
properties.push_back(interpreter_context->NameToPropertyId(prop.name));
properties_string.push_back(prop.name);
}
auto properties_stringified = utils::Join(properties_string, ", ");
Notification constraint_notification(SeverityLevel::INFO);
switch (constraint_query->action_type_) {
case ConstraintQuery::ActionType::CREATE: {
constraint_notification.code = NotificationCode::CREATE_CONSTRAINT;
switch (constraint_query->constraint_.type) {
case Constraint::Type::NODE_KEY:
throw utils::NotYetImplemented("Node key constraints");
case Constraint::Type::EXISTS:
if (properties.empty() || properties.size() > 1) {
throw SyntaxException("Exactly one property must be used for existence constraints.");
}
constraint_notification.title = fmt::format("Created EXISTS constraint on label {} on properties {}.",
constraint_query->constraint_.label.name, properties_stringified);
handler = [interpreter_context, label, label_name = constraint_query->constraint_.label.name,
properties_stringified = std::move(properties_stringified),
properties = std::move(properties)](Notification &constraint_notification) {
auto res = interpreter_context->db->CreateExistenceConstraint(label, properties[0]);
if (res.HasError()) {
auto violation = res.GetError();
auto label_name = interpreter_context->db->LabelToName(violation.label);
MG_ASSERT(violation.properties.size() == 1U);
auto property_name = interpreter_context->db->PropertyToName(*violation.properties.begin());
throw QueryRuntimeException(
"Unable to create existence constraint :{}({}), because an "
"existing node violates it.",
label_name, property_name);
}
if (res.HasValue() && !res.GetValue()) {
constraint_notification.code = NotificationCode::EXISTANT_CONSTRAINT;
constraint_notification.title = fmt::format(
"Constraint EXISTS on label {} on properties {} already exists.", label_name, properties_stringified);
}
};
break;
case Constraint::Type::UNIQUE:
std::set<storage::v3::PropertyId> property_set;
for (const auto &property : properties) {
property_set.insert(property);
}
if (property_set.size() != properties.size()) {
throw SyntaxException("The given set of properties contains duplicates.");
}
constraint_notification.title =
fmt::format("Created UNIQUE constraint on label {} on properties {}.",
constraint_query->constraint_.label.name, utils::Join(properties_string, ", "));
handler = [interpreter_context, label, label_name = constraint_query->constraint_.label.name,
properties_stringified = std::move(properties_stringified),
property_set = std::move(property_set)](Notification &constraint_notification) {
auto res = interpreter_context->db->CreateUniqueConstraint(label, property_set);
if (res.HasError()) {
auto violation = res.GetError();
auto label_name = interpreter_context->db->LabelToName(violation.label);
std::stringstream property_names_stream;
utils::PrintIterable(property_names_stream, violation.properties, ", ",
[&interpreter_context](auto &stream, const auto &prop) {
stream << interpreter_context->db->PropertyToName(prop);
});
throw QueryRuntimeException(
"Unable to create unique constraint :{}({}), because an "
"existing node violates it.",
label_name, property_names_stream.str());
}
switch (res.GetValue()) {
case storage::v3::UniqueConstraints::CreationStatus::EMPTY_PROPERTIES:
throw SyntaxException(
"At least one property must be used for unique "
"constraints.");
case storage::v3::UniqueConstraints::CreationStatus::PROPERTIES_SIZE_LIMIT_EXCEEDED:
throw SyntaxException(
"Too many properties specified. Limit of {} properties "
"for unique constraints is exceeded.",
storage::v3::kUniqueConstraintsMaxProperties);
case storage::v3::UniqueConstraints::CreationStatus::ALREADY_EXISTS:
constraint_notification.code = NotificationCode::EXISTANT_CONSTRAINT;
constraint_notification.title =
fmt::format("Constraint UNIQUE on label {} on properties {} already exists.", label_name,
properties_stringified);
break;
case storage::v3::UniqueConstraints::CreationStatus::SUCCESS:
break;
}
};
break;
}
} break;
case ConstraintQuery::ActionType::DROP: {
constraint_notification.code = NotificationCode::DROP_CONSTRAINT;
switch (constraint_query->constraint_.type) {
case Constraint::Type::NODE_KEY:
throw utils::NotYetImplemented("Node key constraints");
case Constraint::Type::EXISTS:
if (properties.empty() || properties.size() > 1) {
throw SyntaxException("Exactly one property must be used for existence constraints.");
}
constraint_notification.title =
fmt::format("Dropped EXISTS constraint on label {} on properties {}.",
constraint_query->constraint_.label.name, utils::Join(properties_string, ", "));
handler = [interpreter_context, label, label_name = constraint_query->constraint_.label.name,
properties_stringified = std::move(properties_stringified),
properties = std::move(properties)](Notification &constraint_notification) {
if (!interpreter_context->db->DropExistenceConstraint(label, properties[0])) {
constraint_notification.code = NotificationCode::NONEXISTANT_CONSTRAINT;
constraint_notification.title = fmt::format(
"Constraint EXISTS on label {} on properties {} doesn't exist.", label_name, properties_stringified);
}
return std::vector<std::vector<TypedValue>>();
};
break;
case Constraint::Type::UNIQUE:
std::set<storage::v3::PropertyId> property_set;
for (const auto &property : properties) {
property_set.insert(property);
}
if (property_set.size() != properties.size()) {
throw SyntaxException("The given set of properties contains duplicates.");
}
constraint_notification.title =
fmt::format("Dropped UNIQUE constraint on label {} on properties {}.",
constraint_query->constraint_.label.name, utils::Join(properties_string, ", "));
handler = [interpreter_context, label, label_name = constraint_query->constraint_.label.name,
properties_stringified = std::move(properties_stringified),
property_set = std::move(property_set)](Notification &constraint_notification) {
auto res = interpreter_context->db->DropUniqueConstraint(label, property_set);
switch (res) {
case storage::v3::UniqueConstraints::DeletionStatus::EMPTY_PROPERTIES:
throw SyntaxException(
"At least one property must be used for unique "
"constraints.");
break;
case storage::v3::UniqueConstraints::DeletionStatus::PROPERTIES_SIZE_LIMIT_EXCEEDED:
throw SyntaxException(
"Too many properties specified. Limit of {} properties for "
"unique constraints is exceeded.",
storage::v3::kUniqueConstraintsMaxProperties);
break;
case storage::v3::UniqueConstraints::DeletionStatus::NOT_FOUND:
constraint_notification.code = NotificationCode::NONEXISTANT_CONSTRAINT;
constraint_notification.title =
fmt::format("Constraint UNIQUE on label {} on properties {} doesn't exist.", label_name,
properties_stringified);
break;
case storage::v3::UniqueConstraints::DeletionStatus::SUCCESS:
break;
}
return std::vector<std::vector<TypedValue>>();
};
}
} break;
}
return PreparedQuery{{},
std::move(parsed_query.required_privileges),
[handler = std::move(handler), constraint_notification = std::move(constraint_notification),
notifications](AnyStream * /*stream*/, std::optional<int> /*n*/) mutable {
handler(constraint_notification);
notifications->push_back(constraint_notification);
return QueryHandlerResult::COMMIT;
},
RWType::NONE};
throw SemanticException("Constraint query is not supported!");
}
PreparedQuery PrepareSchemaQuery(ParsedQuery parsed_query, bool in_explicit_transaction,
@ -2262,8 +1909,8 @@ Interpreter::PrepareResult Interpreter::Prepare(const std::string &query_string,
(utils::Downcast<CypherQuery>(parsed_query.query) || utils::Downcast<ExplainQuery>(parsed_query.query) ||
utils::Downcast<ProfileQuery>(parsed_query.query) || utils::Downcast<DumpQuery>(parsed_query.query) ||
utils::Downcast<TriggerQuery>(parsed_query.query))) {
db_accessor_ =
std::make_unique<storage::v3::Shard::Accessor>(interpreter_context_->db->Access(GetIsolationLevelOverride()));
db_accessor_ = std::make_unique<storage::v3::Shard::Accessor>(
interpreter_context_->db->Access(coordinator::Hlc{}, GetIsolationLevelOverride()));
execution_db_accessor_.emplace(db_accessor_.get());
if (utils::Downcast<CypherQuery>(parsed_query.query) && interpreter_context_->trigger_store.HasTriggers()) {
@ -2345,13 +1992,6 @@ Interpreter::PrepareResult Interpreter::Prepare(const std::string &query_string,
UpdateTypeCount(rw_type);
if (const auto query_type = query_execution->prepared_query->rw_type;
interpreter_context_->db->GetReplicationRole() == storage::v3::ReplicationRole::REPLICA &&
(query_type == RWType::W || query_type == RWType::RW)) {
query_execution = nullptr;
throw QueryException("Write query forbidden on the replica!");
}
return {query_execution->prepared_query->header, query_execution->prepared_query->privileges, qid};
} catch (const utils::BasicException &) {
EventCounter::IncrementCounter(EventCounter::FailedQuery);
@ -2378,7 +2018,7 @@ void RunTriggersIndividually(const utils::SkipList<Trigger> &triggers, Interpret
utils::MonotonicBufferResource execution_memory{kExecutionMemoryBlockSize};
// create a new transaction for each trigger
auto storage_acc = interpreter_context->db->Access();
auto storage_acc = interpreter_context->db->Access(coordinator::Hlc{});
DbAccessor db_accessor{&storage_acc};
trigger_context.AdaptForAccessor(&db_accessor);
@ -2391,29 +2031,7 @@ void RunTriggersIndividually(const utils::SkipList<Trigger> &triggers, Interpret
continue;
}
auto maybe_constraint_violation = db_accessor.Commit();
if (maybe_constraint_violation.HasError()) {
const auto &constraint_violation = maybe_constraint_violation.GetError();
switch (constraint_violation.type) {
case storage::v3::ConstraintViolation::Type::EXISTENCE: {
const auto &label_name = db_accessor.LabelToName(constraint_violation.label);
MG_ASSERT(constraint_violation.properties.size() == 1U);
const auto &property_name = db_accessor.PropertyToName(*constraint_violation.properties.begin());
spdlog::warn("Trigger '{}' failed to commit due to existence constraint violation on :{}({})", trigger.Name(),
label_name, property_name);
break;
}
case storage::v3::ConstraintViolation::Type::UNIQUE: {
const auto &label_name = db_accessor.LabelToName(constraint_violation.label);
std::stringstream property_names_stream;
utils::PrintIterable(property_names_stream, constraint_violation.properties, ", ",
[&](auto &stream, const auto &prop) { stream << db_accessor.PropertyToName(prop); });
spdlog::warn("Trigger '{}' failed to commit due to unique constraint violation on :{}({})", trigger.Name(),
label_name, property_names_stream.str());
break;
}
}
}
db_accessor.Commit();
}
}
} // namespace
@ -2454,33 +2072,7 @@ void Interpreter::Commit() {
trigger_context_collector_.reset();
};
auto maybe_constraint_violation = db_accessor_->Commit();
if (maybe_constraint_violation.HasError()) {
const auto &constraint_violation = maybe_constraint_violation.GetError();
switch (constraint_violation.type) {
case storage::v3::ConstraintViolation::Type::EXISTENCE: {
auto label_name = execution_db_accessor_->LabelToName(constraint_violation.label);
MG_ASSERT(constraint_violation.properties.size() == 1U);
auto property_name = execution_db_accessor_->PropertyToName(*constraint_violation.properties.begin());
reset_necessary_members();
throw QueryException("Unable to commit due to existence constraint violation on :{}({})", label_name,
property_name);
break;
}
case storage::v3::ConstraintViolation::Type::UNIQUE: {
auto label_name = execution_db_accessor_->LabelToName(constraint_violation.label);
std::stringstream property_names_stream;
utils::PrintIterable(
property_names_stream, constraint_violation.properties, ", ",
[this](auto &stream, const auto &prop) { stream << execution_db_accessor_->PropertyToName(prop); });
reset_necessary_members();
throw QueryException("Unable to commit due to unique constraint violation on :{}({})", label_name,
property_names_stream.str());
break;
}
}
}
db_accessor_->Commit(coordinator::Hlc{});
// The ordered execution of after commit triggers is heavily depending on the exclusiveness of db_accessor_->Commit():
// only one of the transactions can be commiting at the same time, so when the commit is finished, that transaction
// probably will schedule its after commit triggers, because the other transactions that want to commit are still
@ -2492,7 +2084,6 @@ void Interpreter::Commit() {
user_transaction = std::shared_ptr(std::move(db_accessor_))]() mutable {
RunTriggersIndividually(interpreter_context->trigger_store.AfterCommitTriggers(), interpreter_context,
std::move(trigger_context));
user_transaction->FinalizeTransaction();
SPDLOG_DEBUG("Finished executing after commit triggers"); // NOLINT(bugprone-lambda-function-name)
});
}

View File

@ -47,6 +47,7 @@
#include "utils/likely.hpp"
#include "utils/logging.hpp"
#include "utils/memory.hpp"
#include "utils/message.hpp"
#include "utils/pmr/unordered_map.hpp"
#include "utils/pmr/unordered_set.hpp"
#include "utils/pmr/vector.hpp"

View File

@ -496,12 +496,21 @@ struct UpdateEdgesResponse {
bool success;
};
struct CommitRequest {
Hlc transaction_id;
Hlc commit_timestamp;
};
struct CommitResponse {
bool success;
};
using ReadRequests = std::variant<ExpandOneRequest, GetPropertiesRequest, ScanVerticesRequest>;
using ReadResponses = std::variant<ExpandOneResponse, GetPropertiesResponse, ScanVerticesResponse>;
using WriteRequests = std::variant<CreateVerticesRequest, DeleteVerticesRequest, UpdateVerticesRequest,
CreateEdgesRequest, DeleteEdgesRequest, UpdateEdgesRequest>;
CreateEdgesRequest, DeleteEdgesRequest, UpdateEdgesRequest, CommitRequest>;
using WriteResponses = std::variant<CreateVerticesResponse, DeleteVerticesResponse, UpdateVerticesResponse,
CreateEdgesResponse, DeleteEdgesResponse, UpdateEdgesResponse>;
CreateEdgesResponse, DeleteEdgesResponse, UpdateEdgesResponse, CommitResponse>;
} // namespace memgraph::msgs

View File

@ -490,7 +490,7 @@ Streams::StreamsMap::iterator Streams::CreateConsumer(StreamsMap &map, const std
total_retries = interpreter_context_->config.stream_transaction_conflict_retries,
retry_interval = interpreter_context_->config.stream_transaction_retry_interval](
const std::vector<typename TStream::Message> &messages) mutable {
auto accessor = interpreter_context->db->Access();
auto accessor = interpreter_context->db->Access(coordinator::Hlc{});
EventCounter::IncrementCounter(EventCounter::MessagesConsumed, messages.size());
CallCustomTransformation(transformation_name, messages, result, accessor, *memory_resource, stream_name);
@ -738,7 +738,7 @@ TransformationResult Streams::Check(const std::string &stream_name, std::optiona
auto consumer_function = [interpreter_context = interpreter_context_, memory_resource, &stream_name,
&transformation_name = transformation_name, &result,
&test_result]<typename T>(const std::vector<T> &messages) mutable {
auto accessor = interpreter_context->db->Access();
auto accessor = interpreter_context->db->Access(coordinator::Hlc{});
CallCustomTransformation(transformation_name, messages, result, accessor, *memory_resource, stream_name);
auto result_row = std::vector<TypedValue>();

View File

@ -6,13 +6,7 @@ add_custom_target(generate_lcp_ast_storage_v3 DEPENDS ${generated_lcp_storage_v3
set(storage_v3_src_files
${lcp_storage_v3_cpp_files}
commit_log.cpp
constraints.cpp
temporal.cpp
durability/durability.cpp
durability/serialization.cpp
durability/snapshot.cpp
durability/wal.cpp
edge_accessor.cpp
indices.cpp
key_store.cpp
@ -26,21 +20,6 @@ set(storage_v3_src_files
shard_rsm.cpp
storage.cpp)
# #### Replication #####
define_add_lcp(add_lcp_storage lcp_storage_cpp_files generated_lcp_storage_files)
add_lcp_storage(replication/rpc.lcp SLK_SERIALIZE)
add_custom_target(generate_lcp_storage_v3 DEPENDS ${generated_lcp_storage_files})
set(storage_v3_src_files
${storage_v3_src_files}
replication/replication_client.cpp
replication/replication_server.cpp
replication/serialization.cpp
replication/slk.cpp
${lcp_storage_cpp_files})
# ######################
find_package(gflags REQUIRED)
find_package(Threads REQUIRED)
@ -51,4 +30,4 @@ target_link_libraries(mg-storage-v3 Threads::Threads mg-utils gflags)
target_include_directories(mg-storage-v3 PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/bindings)
add_dependencies(mg-storage-v3 generate_lcp_storage)
target_link_libraries(mg-storage-v3 mg-rpc mg-slk mg-expr)
target_link_libraries(mg-storage-v3 mg-slk mg-expr mg-io)

View File

@ -60,8 +60,6 @@ class DbAccessor final {
return std::nullopt;
}
void FinalizeTransaction() { accessor_->FinalizeTransaction(); }
VerticesIterable Vertices(storage::v3::View view) { return VerticesIterable(accessor_->Vertices(view)); }
VerticesIterable Vertices(storage::v3::View view, storage::v3::LabelId label) {
@ -169,8 +167,6 @@ class DbAccessor final {
void AdvanceCommand() { accessor_->AdvanceCommand(); }
utils::BasicResult<storage::v3::ConstraintViolation, void> Commit() { return accessor_->Commit(); }
void Abort() { accessor_->Abort(); }
bool LabelIndexExists(storage::v3::LabelId label) const { return accessor_->LabelIndexExists(label); }
@ -200,8 +196,6 @@ class DbAccessor final {
storage::v3::IndicesInfo ListAllIndices() const { return accessor_->ListAllIndices(); }
storage::v3::ConstraintsInfo ListAllConstraints() const { return accessor_->ListAllConstraints(); }
const storage::v3::SchemaValidator &GetSchemaValidator() const { return accessor_->GetSchemaValidator(); }
storage::v3::SchemasInfo ListAllSchemas() const { return accessor_->ListAllSchemas(); }

View File

@ -1,108 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include "storage/v3/commit_log.hpp"
#include "utils/memory.hpp"
namespace memgraph::storage::v3 {
CommitLog::CommitLog() : allocator_(utils::NewDeleteResource()) {}
CommitLog::CommitLog(uint64_t oldest_active)
: head_start_{oldest_active / kIdsInBlock * kIdsInBlock},
next_start_{head_start_ + kIdsInBlock},
allocator_{utils::NewDeleteResource()} {
head_ = allocator_.allocate(1);
allocator_.construct(head_);
// set all the previous ids
const auto field_idx = (oldest_active % kIdsInBlock) / kIdsInField;
for (size_t i = 0; i < field_idx; ++i) {
head_->field[i] = std::numeric_limits<uint64_t>::max();
}
const auto idx_in_field = oldest_active % kIdsInField;
if (idx_in_field != 0) {
head_->field[field_idx] = std::numeric_limits<uint64_t>::max();
head_->field[field_idx] >>= kIdsInField - idx_in_field;
}
oldest_active_ = oldest_active;
}
CommitLog::~CommitLog() {
while (head_) {
Block *tmp = head_->next;
head_->~Block();
allocator_.deallocate(head_, 1);
head_ = tmp;
}
}
void CommitLog::MarkFinished(uint64_t id) {
Block *block = FindOrCreateBlock(id);
block->field[(id % kIdsInBlock) / kIdsInField] |= 1ULL << (id % kIdsInField);
if (id == oldest_active_) {
UpdateOldestActive();
}
}
uint64_t CommitLog::OldestActive() const noexcept { return oldest_active_; }
void CommitLog::UpdateOldestActive() {
while (head_) {
// This is necessary for amortized constant complexity. If we always start
// from the 0th field, the amount of steps we make through each block is
// quadratic in kBlockSize.
uint64_t start_field = oldest_active_ >= head_start_ ? (oldest_active_ - head_start_) / kIdsInField : 0;
for (uint64_t i = start_field; i < kBlockSize; ++i) {
if (head_->field[i] != std::numeric_limits<uint64_t>::max()) {
// NOLINTNEXTLINE(cppcoreguidelines-narrowing-conversions)
oldest_active_ = head_start_ + i * kIdsInField + __builtin_ffsl(static_cast<int64_t>(~head_->field[i])) - 1;
return;
}
}
// All IDs in this block are marked, we can delete it now.
Block *tmp = head_->next;
head_->~Block();
allocator_.deallocate(head_, 1);
head_ = tmp;
head_start_ += kIdsInBlock;
}
oldest_active_ = next_start_;
}
CommitLog::Block *CommitLog::FindOrCreateBlock(const uint64_t id) {
if (!head_) {
head_ = allocator_.allocate(1);
allocator_.construct(head_);
head_start_ = next_start_;
next_start_ += kIdsInBlock;
}
Block *current = head_;
uint64_t current_start = head_start_;
while (id >= current_start + kIdsInBlock) {
if (!current->next) {
current->next = allocator_.allocate(1);
allocator_.construct(current->next);
next_start_ += kIdsInBlock;
}
current = current->next;
current_start += kIdsInBlock;
}
return current;
}
} // namespace memgraph::storage::v3

View File

@ -1,78 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
/// @file commit_log.hpp
#pragma once
#include <cstdint>
#include <mutex>
#include "utils/memory.hpp"
#include "utils/spin_lock.hpp"
namespace memgraph::storage::v3 {
/// This class keeps track of finalized transactions to provide info on the
/// oldest active transaction (minimal transaction ID which could still be
/// active).
///
/// Basically, it is a set which, at the beginning, contains all transaction
/// IDs and supports two operations: remove an ID from the set (\ref
/// SetFinished) and retrieve the minimal ID still in the set (\ref
/// OldestActive).
///
/// This class is thread-safe.
class CommitLog final {
public:
// TODO(mtomic): use pool allocator for blocks
CommitLog();
/// Create a commit log which has the oldest active id set to
/// oldest_active
/// @param oldest_active the oldest active id
explicit CommitLog(uint64_t oldest_active);
CommitLog(const CommitLog &) = delete;
CommitLog &operator=(const CommitLog &) = delete;
CommitLog(CommitLog &&) = delete;
CommitLog &operator=(CommitLog &&) = delete;
~CommitLog();
/// Mark a transaction as finished.
/// @throw std::bad_alloc
void MarkFinished(uint64_t id);
/// Retrieve the oldest transaction still not marked as finished.
uint64_t OldestActive() const noexcept;
private:
static constexpr uint64_t kBlockSize = 8192;
static constexpr uint64_t kIdsInField = sizeof(uint64_t) * 8;
static constexpr uint64_t kIdsInBlock = kBlockSize * kIdsInField;
struct Block {
Block *next{nullptr};
uint64_t field[kBlockSize]{};
};
void UpdateOldestActive();
/// @throw std::bad_alloc
Block *FindOrCreateBlock(uint64_t id);
Block *head_{nullptr};
uint64_t head_start_{0};
uint64_t next_start_{0};
uint64_t oldest_active_{0};
utils::Allocator<Block> allocator_;
};
} // namespace memgraph::storage::v3

View File

@ -14,6 +14,8 @@
#include <chrono>
#include <cstdint>
#include <filesystem>
#include "io/time.hpp"
#include "storage/v3/id_types.hpp"
#include "storage/v3/isolation_level.hpp"
#include "storage/v3/transaction.hpp"
@ -24,36 +26,14 @@ namespace memgraph::storage::v3 {
/// the storage. This class also defines the default behavior.
struct Config {
struct Gc {
// TODO(antaljanosbenjamin): How to handle garbage collection?
enum class Type { NONE };
Type type{Type::NONE};
std::chrono::milliseconds interval{std::chrono::milliseconds(1000)};
// Interval after which the committed deltas are cleaned up
io::Duration reclamation_interval{};
} gc;
struct Items {
bool properties_on_edges{true};
} items;
struct Durability {
enum class SnapshotWalMode { DISABLED, PERIODIC_SNAPSHOT, PERIODIC_SNAPSHOT_WITH_WAL };
std::filesystem::path storage_directory{"storage"};
bool recover_on_startup{false};
SnapshotWalMode snapshot_wal_mode{SnapshotWalMode::DISABLED};
std::chrono::milliseconds snapshot_interval{std::chrono::minutes(2)};
uint64_t snapshot_retention_count{3};
uint64_t wal_file_size_kibibytes{static_cast<uint64_t>(20 * 1024)};
uint64_t wal_file_flush_every_n_tx{100000};
bool snapshot_on_exit{false};
} durability;
struct Transaction {
IsolationLevel isolation_level{IsolationLevel::SNAPSHOT_ISOLATION};
} transaction;

View File

@ -1,414 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include "storage/v3/constraints.hpp"
#include <algorithm>
#include <cstring>
#include <map>
#include "storage/v3/mvcc.hpp"
#include "storage/v3/vertex.hpp"
#include "utils/logging.hpp"
namespace memgraph::storage::v3 {
namespace {
/// Helper function that determines position of the given `property` in the
/// sorted `property_array` using binary search. In the case that `property`
/// cannot be found, `std::nullopt` is returned.
std::optional<size_t> FindPropertyPosition(const PropertyIdArray &property_array, PropertyId property) {
const auto *it = std::lower_bound(property_array.values, property_array.values + property_array.size, property);
if (it == property_array.values + property_array.size || *it != property) {
return std::nullopt;
}
return it - property_array.values;
}
/// Helper function for validating unique constraints on commit. Returns true if
/// the last committed version of the given vertex contains the given label and
/// set of property values. This function should be called when commit lock is
/// active.
bool LastCommittedVersionHasLabelProperty(const Vertex &vertex, LabelId label, const std::set<PropertyId> &properties,
const std::vector<PropertyValue> &value_array, const Transaction &transaction,
uint64_t commit_timestamp) {
MG_ASSERT(properties.size() == value_array.size(), "Invalid database state!");
PropertyIdArray property_array(properties.size());
bool current_value_equal_to_value[kUniqueConstraintsMaxProperties];
memset(current_value_equal_to_value, 0, sizeof(current_value_equal_to_value));
// Since the commit lock is active, any transaction that tries to write to
// a vertex which is part of the given `transaction` will result in a
// serialization error. But, note that the given `vertex`'s data does not have
// to be modified in the current `transaction`, meaning that a guard lock to
// access vertex's data is still necessary because another active transaction
// could modify it in the meantime.
Delta *delta{nullptr};
bool deleted{false};
bool has_label{false};
{
delta = vertex.delta;
deleted = vertex.deleted;
has_label = VertexHasLabel(vertex, label);
size_t i = 0;
for (const auto &property : properties) {
current_value_equal_to_value[i] = vertex.properties.IsPropertyEqual(property, value_array[i]);
property_array.values[i] = property;
i++;
}
}
while (delta != nullptr) {
auto ts = delta->timestamp->load(std::memory_order_acquire);
if (ts < commit_timestamp || ts == transaction.transaction_id) {
break;
}
switch (delta->action) {
case Delta::Action::SET_PROPERTY: {
auto pos = FindPropertyPosition(property_array, delta->property.key);
if (pos) {
current_value_equal_to_value[*pos] = delta->property.value == value_array[*pos];
}
break;
}
case Delta::Action::DELETE_OBJECT: {
MG_ASSERT(!deleted, "Invalid database state!");
deleted = true;
break;
}
case Delta::Action::RECREATE_OBJECT: {
MG_ASSERT(deleted, "Invalid database state!");
deleted = false;
break;
}
case Delta::Action::ADD_LABEL: {
if (delta->label == label) {
MG_ASSERT(!has_label, "Invalid database state!");
has_label = true;
break;
}
}
case Delta::Action::REMOVE_LABEL: {
if (delta->label == label) {
MG_ASSERT(has_label, "Invalid database state!");
has_label = false;
break;
}
}
case Delta::Action::ADD_IN_EDGE:
case Delta::Action::ADD_OUT_EDGE:
case Delta::Action::REMOVE_IN_EDGE:
case Delta::Action::REMOVE_OUT_EDGE:
break;
}
delta = delta->next.load(std::memory_order_acquire);
}
for (size_t i = 0; i < properties.size(); ++i) {
if (!current_value_equal_to_value[i]) {
return false;
}
}
return !deleted && has_label;
}
/// Helper function for unique constraint garbage collection. Returns true if
/// there's a reachable version of the vertex that has the given label and
/// property values.
bool AnyVersionHasLabelProperty(const Vertex &vertex, LabelId label, const std::set<PropertyId> &properties,
const std::vector<PropertyValue> &values, uint64_t timestamp) {
MG_ASSERT(properties.size() == values.size(), "Invalid database state!");
PropertyIdArray property_array(properties.size());
bool current_value_equal_to_value[kUniqueConstraintsMaxProperties];
memset(current_value_equal_to_value, 0, sizeof(current_value_equal_to_value));
bool has_label{false};
bool deleted{false};
Delta *delta{nullptr};
{
has_label = VertexHasLabel(vertex, label);
deleted = vertex.deleted;
delta = vertex.delta;
size_t i = 0;
for (const auto &property : properties) {
current_value_equal_to_value[i] = vertex.properties.IsPropertyEqual(property, values[i]);
property_array.values[i] = property;
i++;
}
}
{
bool all_values_match = true;
for (size_t i = 0; i < values.size(); ++i) {
if (!current_value_equal_to_value[i]) {
all_values_match = false;
break;
}
}
if (!deleted && has_label && all_values_match) {
return true;
}
}
while (delta != nullptr) {
auto ts = delta->timestamp->load(std::memory_order_acquire);
if (ts < timestamp) {
break;
}
switch (delta->action) {
case Delta::Action::ADD_LABEL:
if (delta->label == label) {
MG_ASSERT(!has_label, "Invalid database state!");
has_label = true;
}
break;
case Delta::Action::REMOVE_LABEL:
if (delta->label == label) {
MG_ASSERT(has_label, "Invalid database state!");
has_label = false;
}
break;
case Delta::Action::SET_PROPERTY: {
auto pos = FindPropertyPosition(property_array, delta->property.key);
if (pos) {
current_value_equal_to_value[*pos] = delta->property.value == values[*pos];
}
break;
}
case Delta::Action::RECREATE_OBJECT: {
MG_ASSERT(deleted, "Invalid database state!");
deleted = false;
break;
}
case Delta::Action::DELETE_OBJECT: {
MG_ASSERT(!deleted, "Invalid database state!");
deleted = true;
break;
}
case Delta::Action::ADD_IN_EDGE:
case Delta::Action::ADD_OUT_EDGE:
case Delta::Action::REMOVE_IN_EDGE:
case Delta::Action::REMOVE_OUT_EDGE:
break;
}
bool all_values_match = true;
for (size_t i = 0; i < values.size(); ++i) {
if (!current_value_equal_to_value[i]) {
all_values_match = false;
break;
}
}
if (!deleted && has_label && all_values_match) {
return true;
}
delta = delta->next.load(std::memory_order_acquire);
}
return false;
}
/// Helper function that, given the set of `properties`, extracts corresponding
/// property values from the `vertex`.
/// @throw std::bad_alloc
std::optional<std::vector<PropertyValue>> ExtractPropertyValues(const Vertex &vertex,
const std::set<PropertyId> &properties) {
std::vector<PropertyValue> value_array;
value_array.reserve(properties.size());
for (const auto &prop : properties) {
auto value = vertex.properties.GetProperty(prop);
if (value.IsNull()) {
return std::nullopt;
}
value_array.emplace_back(std::move(value));
}
return std::move(value_array);
}
} // namespace
bool operator==(const ConstraintViolation &lhs, const ConstraintViolation &rhs) {
return lhs.type == rhs.type && lhs.label == rhs.label && lhs.properties == rhs.properties;
}
bool UniqueConstraints::Entry::operator<(const Entry &rhs) const {
if (values < rhs.values) {
return true;
}
if (rhs.values < values) {
return false;
}
return std::make_tuple(vertex, timestamp) < std::make_tuple(rhs.vertex, rhs.timestamp);
}
bool UniqueConstraints::Entry::operator==(const Entry &rhs) const {
return values == rhs.values && vertex == rhs.vertex && timestamp == rhs.timestamp;
}
bool UniqueConstraints::Entry::operator<(const std::vector<PropertyValue> &rhs) const { return values < rhs; }
bool UniqueConstraints::Entry::operator==(const std::vector<PropertyValue> &rhs) const { return values == rhs; }
void UniqueConstraints::UpdateBeforeCommit(const Vertex *vertex, const Transaction &tx) {
for (auto &[label_props, storage] : constraints_) {
if (!VertexHasLabel(*vertex, label_props.first)) {
continue;
}
auto values = ExtractPropertyValues(*vertex, label_props.second);
if (values) {
auto acc = storage.access();
acc.insert(Entry{std::move(*values), vertex, tx.start_timestamp});
}
}
}
utils::BasicResult<ConstraintViolation, UniqueConstraints::CreationStatus> UniqueConstraints::CreateConstraint(
LabelId label, const std::set<PropertyId> &properties, VerticesSkipList::Accessor vertices) {
if (properties.empty()) {
return CreationStatus::EMPTY_PROPERTIES;
}
if (properties.size() > kUniqueConstraintsMaxProperties) {
return CreationStatus::PROPERTIES_SIZE_LIMIT_EXCEEDED;
}
auto [constraint, emplaced] =
constraints_.emplace(std::piecewise_construct, std::forward_as_tuple(label, properties), std::forward_as_tuple());
if (!emplaced) {
// Constraint already exists.
return CreationStatus::ALREADY_EXISTS;
}
bool violation_found = false;
{
auto acc = constraint->second.access();
for (const auto &lo_vertex : vertices) {
const auto &vertex = lo_vertex.vertex;
if (vertex.deleted || !VertexHasLabel(vertex, label)) {
continue;
}
auto values = ExtractPropertyValues(vertex, properties);
if (!values) {
continue;
}
// Check whether there already is a vertex with the same values for the
// given label and property.
auto it = acc.find_equal_or_greater(*values);
if (it != acc.end() && it->values == *values) {
violation_found = true;
break;
}
acc.insert(Entry{std::move(*values), &vertex, 0});
}
}
if (violation_found) {
// In the case of the violation, storage for the current constraint has to
// be removed.
constraints_.erase(constraint);
return ConstraintViolation{ConstraintViolation::Type::UNIQUE, label, properties};
}
return CreationStatus::SUCCESS;
}
UniqueConstraints::DeletionStatus UniqueConstraints::DropConstraint(LabelId label,
const std::set<PropertyId> &properties) {
if (properties.empty()) {
return UniqueConstraints::DeletionStatus::EMPTY_PROPERTIES;
}
if (properties.size() > kUniqueConstraintsMaxProperties) {
return UniqueConstraints::DeletionStatus::PROPERTIES_SIZE_LIMIT_EXCEEDED;
}
if (constraints_.erase({label, properties}) > 0) {
return UniqueConstraints::DeletionStatus::SUCCESS;
}
return UniqueConstraints::DeletionStatus::NOT_FOUND;
}
std::optional<ConstraintViolation> UniqueConstraints::Validate(const Vertex &vertex, const Transaction &tx,
uint64_t commit_timestamp) const {
if (vertex.deleted) {
return std::nullopt;
}
for (const auto &[label_props, storage] : constraints_) {
const auto &label = label_props.first;
const auto &properties = label_props.second;
if (!VertexHasLabel(vertex, label)) {
continue;
}
auto value_array = ExtractPropertyValues(vertex, properties);
if (!value_array) {
continue;
}
auto acc = storage.access();
auto it = acc.find_equal_or_greater(*value_array);
for (; it != acc.end(); ++it) {
if (*value_array < it->values) {
break;
}
// The `vertex` that is going to be committed violates a unique constraint
// if it's different than a vertex indexed in the list of constraints and
// has the same label and property value as the last committed version of
// the vertex from the list.
if (&vertex != it->vertex &&
LastCommittedVersionHasLabelProperty(*it->vertex, label, properties, *value_array, tx, commit_timestamp)) {
return ConstraintViolation{ConstraintViolation::Type::UNIQUE, label, properties};
}
}
}
return std::nullopt;
}
std::vector<std::pair<LabelId, std::set<PropertyId>>> UniqueConstraints::ListConstraints() const {
std::vector<std::pair<LabelId, std::set<PropertyId>>> ret;
ret.reserve(constraints_.size());
for (const auto &[label_props, _] : constraints_) {
ret.push_back(label_props);
}
return ret;
}
void UniqueConstraints::RemoveObsoleteEntries(uint64_t oldest_active_start_timestamp) {
for (auto &[label_props, storage] : constraints_) {
auto acc = storage.access();
for (auto it = acc.begin(); it != acc.end();) {
auto next_it = it;
++next_it;
if (it->timestamp >= oldest_active_start_timestamp) {
it = next_it;
continue;
}
if ((next_it != acc.end() && it->vertex == next_it->vertex && it->values == next_it->values) ||
!AnyVersionHasLabelProperty(*it->vertex, label_props.first, label_props.second, it->values,
oldest_active_start_timestamp)) {
acc.remove(*it);
}
it = next_it;
}
}
}
} // namespace memgraph::storage::v3

View File

@ -1,202 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <optional>
#include <set>
#include <vector>
#include "storage/v3/id_types.hpp"
#include "storage/v3/transaction.hpp"
#include "storage/v3/vertex.hpp"
#include "storage/v3/vertices_skip_list.hpp"
#include "utils/logging.hpp"
#include "utils/result.hpp"
#include "utils/skip_list.hpp"
namespace memgraph::storage::v3 {
// NOLINTNEXTLINE(misc-definitions-in-headers)
const size_t kUniqueConstraintsMaxProperties = 32;
/// Utility class to store data in a fixed size array. The array is used
/// instead of `std::vector` to avoid `std::bad_alloc` exception where not
/// necessary.
template <class T>
struct FixedCapacityArray {
size_t size;
T values[kUniqueConstraintsMaxProperties];
explicit FixedCapacityArray(size_t array_size) : size(array_size) {
MG_ASSERT(size <= kUniqueConstraintsMaxProperties, "Invalid array size!");
}
};
using PropertyIdArray = FixedCapacityArray<PropertyId>;
struct ConstraintViolation {
enum class Type {
EXISTENCE,
UNIQUE,
};
Type type;
LabelId label;
// While multiple properties are supported by unique constraints, the
// `properties` set will always have exactly one element in the case of
// existence constraint violation.
std::set<PropertyId> properties;
};
bool operator==(const ConstraintViolation &lhs, const ConstraintViolation &rhs);
class UniqueConstraints {
private:
struct Entry {
std::vector<PropertyValue> values;
const Vertex *vertex;
uint64_t timestamp;
bool operator<(const Entry &rhs) const;
bool operator==(const Entry &rhs) const;
bool operator<(const std::vector<PropertyValue> &rhs) const;
bool operator==(const std::vector<PropertyValue> &rhs) const;
};
public:
/// Status for creation of unique constraints.
/// Note that this does not cover the case when the constraint is violated.
enum class CreationStatus {
SUCCESS,
ALREADY_EXISTS,
EMPTY_PROPERTIES,
PROPERTIES_SIZE_LIMIT_EXCEEDED,
};
/// Status for deletion of unique constraints.
enum class DeletionStatus {
SUCCESS,
NOT_FOUND,
EMPTY_PROPERTIES,
PROPERTIES_SIZE_LIMIT_EXCEEDED,
};
/// Indexes the given vertex for relevant labels and properties.
/// This method should be called before committing and validating vertices
/// against unique constraints.
/// @throw std::bad_alloc
void UpdateBeforeCommit(const Vertex *vertex, const Transaction &tx);
/// Creates unique constraint on the given `label` and a list of `properties`.
/// Returns constraint violation if there are multiple vertices with the same
/// label and property values. Returns `CreationStatus::ALREADY_EXISTS` if
/// constraint already existed, `CreationStatus::EMPTY_PROPERTIES` if the
/// given list of properties is empty,
/// `CreationStatus::PROPERTIES_SIZE_LIMIT_EXCEEDED` if the list of properties
/// exceeds the maximum allowed number of properties, and
/// `CreationStatus::SUCCESS` on success.
/// @throw std::bad_alloc
utils::BasicResult<ConstraintViolation, CreationStatus> CreateConstraint(LabelId label,
const std::set<PropertyId> &properties,
VerticesSkipList::Accessor vertices);
/// Deletes the specified constraint. Returns `DeletionStatus::NOT_FOUND` if
/// there is not such constraint in the storage,
/// `DeletionStatus::EMPTY_PROPERTIES` if the given set of `properties` is
/// empty, `DeletionStatus::PROPERTIES_SIZE_LIMIT_EXCEEDED` if the given set
/// of `properties` exceeds the maximum allowed number of properties, and
/// `DeletionStatus::SUCCESS` on success.
DeletionStatus DropConstraint(LabelId label, const std::set<PropertyId> &properties);
bool ConstraintExists(LabelId label, const std::set<PropertyId> &properties) {
return constraints_.find({label, properties}) != constraints_.end();
}
/// Validates the given vertex against unique constraints before committing.
/// This method should be called while commit lock is active with
/// `commit_timestamp` being a potential commit timestamp of the transaction.
/// @throw std::bad_alloc
std::optional<ConstraintViolation> Validate(const Vertex &vertex, const Transaction &tx,
uint64_t commit_timestamp) const;
std::vector<std::pair<LabelId, std::set<PropertyId>>> ListConstraints() const;
/// GC method that removes outdated entries from constraints' storages.
void RemoveObsoleteEntries(uint64_t oldest_active_start_timestamp);
void Clear() { constraints_.clear(); }
private:
std::map<std::pair<LabelId, std::set<PropertyId>>, utils::SkipList<Entry>> constraints_;
};
struct Constraints {
std::vector<std::pair<LabelId, PropertyId>> existence_constraints;
UniqueConstraints unique_constraints;
};
/// Adds a unique constraint to `constraints`. Returns true if the constraint
/// was successfully added, false if it already exists and a
/// `ConstraintViolation` if there is an existing vertex violating the
/// constraint.
///
/// @throw std::bad_alloc
/// @throw std::length_error
inline utils::BasicResult<ConstraintViolation, bool> CreateExistenceConstraint(Constraints *constraints, LabelId label,
PropertyId property,
VerticesSkipList::Accessor vertices) {
if (utils::Contains(constraints->existence_constraints, std::make_pair(label, property))) {
return false;
}
for (const auto &lgo_vertex : vertices) {
const auto &vertex = lgo_vertex.vertex;
if (!vertex.deleted && VertexHasLabel(vertex, label) && !vertex.properties.HasProperty(property)) {
return ConstraintViolation{ConstraintViolation::Type::EXISTENCE, label, std::set<PropertyId>{property}};
}
}
constraints->existence_constraints.emplace_back(label, property);
return true;
}
/// Removes a unique constraint from `constraints`. Returns true if the
/// constraint was removed, and false if it doesn't exist.
inline bool DropExistenceConstraint(Constraints *constraints, LabelId label, PropertyId property) {
auto it = std::find(constraints->existence_constraints.begin(), constraints->existence_constraints.end(),
std::make_pair(label, property));
if (it == constraints->existence_constraints.end()) {
return false;
}
constraints->existence_constraints.erase(it);
return true;
}
/// Verifies that the given vertex satisfies all existence constraints. Returns
/// `std::nullopt` if all checks pass, and `ConstraintViolation` describing the
/// violated constraint otherwise.
[[nodiscard]] inline std::optional<ConstraintViolation> ValidateExistenceConstraints(const Vertex &vertex,
const Constraints &constraints) {
for (const auto &[label, property] : constraints.existence_constraints) {
if (!vertex.deleted && VertexHasLabel(vertex, label) && !vertex.properties.HasProperty(property)) {
return ConstraintViolation{ConstraintViolation::Type::EXISTENCE, label, std::set<PropertyId>{property}};
}
}
return std::nullopt;
}
/// Returns a list of all created existence constraints.
inline std::vector<std::pair<LabelId, PropertyId>> ListExistenceConstraints(const Constraints &constraints) {
return constraints.existence_constraints;
}
} // namespace memgraph::storage::v3

View File

@ -11,9 +11,7 @@
#pragma once
#include <atomic>
#include <memory>
#include "storage/v3/edge_ref.hpp"
#include "storage/v3/id_types.hpp"
#include "storage/v3/property_value.hpp"
@ -26,6 +24,7 @@ namespace memgraph::storage::v3 {
struct Vertex;
struct Edge;
struct Delta;
struct CommitInfo;
// This class stores one of three pointers (`Delta`, `Vertex` and `Edge`)
// without using additional memory for storing the type. The type is stored in
@ -63,31 +62,30 @@ class PreviousPtr {
Edge *edge{nullptr};
};
PreviousPtr() : storage_(0) {}
PreviousPtr() {}
PreviousPtr(const PreviousPtr &other) noexcept : storage_(other.storage_.load(std::memory_order_acquire)) {}
PreviousPtr(const PreviousPtr &other) noexcept : storage_(other.storage_) {}
PreviousPtr(PreviousPtr &&) = delete;
PreviousPtr &operator=(const PreviousPtr &) = delete;
PreviousPtr &operator=(PreviousPtr &&) = delete;
~PreviousPtr() = default;
Pointer Get() const {
uintptr_t value = storage_.load(std::memory_order_acquire);
if (value == 0) {
if (storage_ == 0) {
return {};
}
uintptr_t type = value & kMask;
uintptr_t type = storage_ & kMask;
if (type == kDelta) {
// NOLINTNEXTLINE(performance-no-int-to-ptr)
return Pointer{reinterpret_cast<Delta *>(value & ~kMask)};
return Pointer{reinterpret_cast<Delta *>(storage_ & ~kMask)};
}
if (type == kVertex) {
// NOLINTNEXTLINE(performance-no-int-to-ptr)
return Pointer{reinterpret_cast<Vertex *>(value & ~kMask)};
return Pointer{reinterpret_cast<Vertex *>(storage_ & ~kMask)};
}
if (type == kEdge) {
// NOLINTNEXTLINE(performance-no-int-to-ptr)
return Pointer{reinterpret_cast<Edge *>(value & ~kMask)};
return Pointer{reinterpret_cast<Edge *>(storage_ & ~kMask)};
}
LOG_FATAL("Invalid pointer type!");
}
@ -95,23 +93,23 @@ class PreviousPtr {
void Set(Delta *delta) {
auto value = reinterpret_cast<uintptr_t>(delta);
MG_ASSERT((value & kMask) == 0, "Invalid pointer!");
storage_.store(value | kDelta, std::memory_order_release);
storage_ = value | kDelta;
}
void Set(Vertex *vertex) {
auto value = reinterpret_cast<uintptr_t>(vertex);
MG_ASSERT((value & kMask) == 0, "Invalid pointer!");
storage_.store(value | kVertex, std::memory_order_release);
storage_ = value | kVertex;
}
void Set(Edge *edge) {
auto value = reinterpret_cast<uintptr_t>(edge);
MG_ASSERT((value & kMask) == 0, "Invalid pointer!");
storage_.store(value | kEdge, std::memory_order_release);
storage_ = value | kEdge;
}
private:
std::atomic<uintptr_t> storage_;
uintptr_t storage_{0};
};
inline bool operator==(const PreviousPtr::Pointer &a, const PreviousPtr::Pointer &b) {
@ -159,47 +157,47 @@ struct Delta {
struct RemoveInEdgeTag {};
struct RemoveOutEdgeTag {};
Delta(DeleteObjectTag /*unused*/, std::atomic<uint64_t> *timestamp, uint64_t command_id)
: action(Action::DELETE_OBJECT), timestamp(timestamp), command_id(command_id) {}
Delta(DeleteObjectTag /*unused*/, CommitInfo *commit_info, uint64_t command_id)
: action(Action::DELETE_OBJECT), commit_info(commit_info), command_id(command_id) {}
Delta(RecreateObjectTag /*unused*/, std::atomic<uint64_t> *timestamp, uint64_t command_id)
: action(Action::RECREATE_OBJECT), timestamp(timestamp), command_id(command_id) {}
Delta(RecreateObjectTag /*unused*/, CommitInfo *commit_info, uint64_t command_id)
: action(Action::RECREATE_OBJECT), commit_info(commit_info), command_id(command_id) {}
Delta(AddLabelTag /*unused*/, LabelId label, std::atomic<uint64_t> *timestamp, uint64_t command_id)
: action(Action::ADD_LABEL), timestamp(timestamp), command_id(command_id), label(label) {}
Delta(AddLabelTag /*unused*/, LabelId label, CommitInfo *commit_info, uint64_t command_id)
: action(Action::ADD_LABEL), commit_info(commit_info), command_id(command_id), label(label) {}
Delta(RemoveLabelTag /*unused*/, LabelId label, std::atomic<uint64_t> *timestamp, uint64_t command_id)
: action(Action::REMOVE_LABEL), timestamp(timestamp), command_id(command_id), label(label) {}
Delta(RemoveLabelTag /*unused*/, LabelId label, CommitInfo *commit_info, uint64_t command_id)
: action(Action::REMOVE_LABEL), commit_info(commit_info), command_id(command_id), label(label) {}
Delta(SetPropertyTag /*unused*/, PropertyId key, const PropertyValue &value, std::atomic<uint64_t> *timestamp,
Delta(SetPropertyTag /*unused*/, PropertyId key, const PropertyValue &value, CommitInfo *commit_info,
uint64_t command_id)
: action(Action::SET_PROPERTY), timestamp(timestamp), command_id(command_id), property({key, value}) {}
: action(Action::SET_PROPERTY), commit_info(commit_info), command_id(command_id), property({key, value}) {}
Delta(AddInEdgeTag /*unused*/, EdgeTypeId edge_type, VertexId vertex_id, EdgeRef edge,
std::atomic<uint64_t> *timestamp, uint64_t command_id)
Delta(AddInEdgeTag /*unused*/, EdgeTypeId edge_type, VertexId vertex_id, EdgeRef edge, CommitInfo *commit_info,
uint64_t command_id)
: action(Action::ADD_IN_EDGE),
timestamp(timestamp),
commit_info(commit_info),
command_id(command_id),
vertex_edge({edge_type, std::move(vertex_id), edge}) {}
Delta(AddOutEdgeTag /*unused*/, EdgeTypeId edge_type, VertexId vertex_id, EdgeRef edge,
std::atomic<uint64_t> *timestamp, uint64_t command_id)
Delta(AddOutEdgeTag /*unused*/, EdgeTypeId edge_type, VertexId vertex_id, EdgeRef edge, CommitInfo *commit_info,
uint64_t command_id)
: action(Action::ADD_OUT_EDGE),
timestamp(timestamp),
commit_info(commit_info),
command_id(command_id),
vertex_edge({edge_type, std::move(vertex_id), edge}) {}
Delta(RemoveInEdgeTag /*unused*/, EdgeTypeId edge_type, VertexId vertex_id, EdgeRef edge,
std::atomic<uint64_t> *timestamp, uint64_t command_id)
Delta(RemoveInEdgeTag /*unused*/, EdgeTypeId edge_type, VertexId vertex_id, EdgeRef edge, CommitInfo *commit_info,
uint64_t command_id)
: action(Action::REMOVE_IN_EDGE),
timestamp(timestamp),
commit_info(commit_info),
command_id(command_id),
vertex_edge({edge_type, std::move(vertex_id), edge}) {}
Delta(RemoveOutEdgeTag /*unused*/, EdgeTypeId edge_type, VertexId vertex_id, EdgeRef edge,
std::atomic<uint64_t> *timestamp, uint64_t command_id)
Delta(RemoveOutEdgeTag /*unused*/, EdgeTypeId edge_type, VertexId vertex_id, EdgeRef edge, CommitInfo *commit_info,
uint64_t command_id)
: action(Action::REMOVE_OUT_EDGE),
timestamp(timestamp),
commit_info(commit_info),
command_id(command_id),
vertex_edge({edge_type, std::move(vertex_id), edge}) {}
@ -230,10 +228,10 @@ struct Delta {
Action action;
// TODO: optimize with in-place copy
std::atomic<uint64_t> *timestamp;
CommitInfo *commit_info;
uint64_t command_id;
PreviousPtr prev;
std::atomic<Delta *> next{nullptr};
Delta *next{nullptr};
union {
LabelId label;

View File

@ -1,346 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include "storage/v3/durability/durability.hpp"
#include <pwd.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <unistd.h>
#include <cerrno>
#include <cstring>
#include <algorithm>
#include <tuple>
#include <utility>
#include <vector>
#include "storage/v3/durability/paths.hpp"
#include "storage/v3/durability/snapshot.hpp"
#include "storage/v3/durability/wal.hpp"
#include "utils/logging.hpp"
#include "utils/memory_tracker.hpp"
#include "utils/message.hpp"
namespace memgraph::storage::v3::durability {
void VerifyStorageDirectoryOwnerAndProcessUserOrDie(const std::filesystem::path &storage_directory) {
// Get the process user ID.
auto process_euid = geteuid();
// Get the data directory owner ID.
struct stat statbuf;
auto ret = stat(storage_directory.c_str(), &statbuf);
if (ret != 0 && errno == ENOENT) {
// The directory doesn't currently exist.
return;
}
MG_ASSERT(ret == 0, "Couldn't get stat for '{}' because of: {} ({})", storage_directory, strerror(errno), errno);
auto directory_owner = statbuf.st_uid;
auto get_username = [](auto uid) {
auto info = getpwuid(uid);
if (!info) return std::to_string(uid);
return std::string(info->pw_name);
};
auto user_process = get_username(process_euid);
auto user_directory = get_username(directory_owner);
MG_ASSERT(process_euid == directory_owner,
"The process is running as user {}, but the data directory is "
"owned by user {}. Please start the process as user {}!",
user_process, user_directory, user_directory);
}
std::vector<SnapshotDurabilityInfo> GetSnapshotFiles(const std::filesystem::path &snapshot_directory,
const std::string_view uuid) {
std::vector<SnapshotDurabilityInfo> snapshot_files;
std::error_code error_code;
if (utils::DirExists(snapshot_directory)) {
for (const auto &item : std::filesystem::directory_iterator(snapshot_directory, error_code)) {
if (!item.is_regular_file()) continue;
try {
auto info = ReadSnapshotInfo(item.path());
if (uuid.empty() || info.uuid == uuid) {
snapshot_files.emplace_back(item.path(), std::move(info.uuid), info.start_timestamp);
}
} catch (const RecoveryFailure &) {
continue;
}
}
MG_ASSERT(!error_code, "Couldn't recover data because an error occurred: {}!", error_code.message());
}
return snapshot_files;
}
std::optional<std::vector<WalDurabilityInfo>> GetWalFiles(const std::filesystem::path &wal_directory,
const std::string_view uuid,
const std::optional<size_t> current_seq_num) {
if (!utils::DirExists(wal_directory)) return std::nullopt;
std::vector<WalDurabilityInfo> wal_files;
std::error_code error_code;
for (const auto &item : std::filesystem::directory_iterator(wal_directory, error_code)) {
if (!item.is_regular_file()) continue;
try {
auto info = ReadWalInfo(item.path());
if ((uuid.empty() || info.uuid == uuid) && (!current_seq_num || info.seq_num < *current_seq_num))
wal_files.emplace_back(info.seq_num, info.from_timestamp, info.to_timestamp, std::move(info.uuid),
std::move(info.epoch_id), item.path());
} catch (const RecoveryFailure &e) {
spdlog::warn("Failed to read {}", item.path());
continue;
}
}
MG_ASSERT(!error_code, "Couldn't recover data because an error occurred: {}!", error_code.message());
std::sort(wal_files.begin(), wal_files.end());
return std::move(wal_files);
}
// Function used to recover all discovered indices and constraints. The
// indices and constraints must be recovered after the data recovery is done
// to ensure that the indices and constraints are consistent at the end of the
// recovery process.
void RecoverIndicesAndConstraints(const RecoveredIndicesAndConstraints &indices_constraints, Indices *indices,
Constraints *constraints, VerticesSkipList *vertices) {
spdlog::info("Recreating indices from metadata.");
// Recover label indices.
spdlog::info("Recreating {} label indices from metadata.", indices_constraints.indices.label.size());
for (const auto &item : indices_constraints.indices.label) {
if (!indices->label_index.CreateIndex(item, vertices->access()))
throw RecoveryFailure("The label index must be created here!");
spdlog::info("A label index is recreated from metadata.");
}
spdlog::info("Label indices are recreated.");
// Recover label+property indices.
spdlog::info("Recreating {} label+property indices from metadata.",
indices_constraints.indices.label_property.size());
for (const auto &item : indices_constraints.indices.label_property) {
if (!indices->label_property_index.CreateIndex(item.first, item.second, vertices->access()))
throw RecoveryFailure("The label+property index must be created here!");
spdlog::info("A label+property index is recreated from metadata.");
}
spdlog::info("Label+property indices are recreated.");
spdlog::info("Indices are recreated.");
spdlog::info("Recreating constraints from metadata.");
// Recover existence constraints.
spdlog::info("Recreating {} existence constraints from metadata.", indices_constraints.constraints.existence.size());
for (const auto &item : indices_constraints.constraints.existence) {
auto ret = CreateExistenceConstraint(constraints, item.first, item.second, vertices->access());
if (ret.HasError() || !ret.GetValue()) throw RecoveryFailure("The existence constraint must be created here!");
spdlog::info("A existence constraint is recreated from metadata.");
}
spdlog::info("Existence constraints are recreated from metadata.");
// Recover unique constraints.
spdlog::info("Recreating {} unique constraints from metadata.", indices_constraints.constraints.unique.size());
for (const auto &item : indices_constraints.constraints.unique) {
auto ret = constraints->unique_constraints.CreateConstraint(item.first, item.second, vertices->access());
if (ret.HasError() || ret.GetValue() != UniqueConstraints::CreationStatus::SUCCESS)
throw RecoveryFailure("The unique constraint must be created here!");
spdlog::info("A unique constraint is recreated from metadata.");
}
spdlog::info("Unique constraints are recreated from metadata.");
spdlog::info("Constraints are recreated from metadata.");
}
std::optional<RecoveryInfo> RecoverData(const std::filesystem::path &snapshot_directory,
const std::filesystem::path &wal_directory, std::string *uuid,
std::string *epoch_id,
std::deque<std::pair<std::string, uint64_t>> *epoch_history,
VerticesSkipList *vertices, utils::SkipList<Edge> *edges, uint64_t *edge_count,
NameIdMapper *name_id_mapper, Indices *indices, Constraints *constraints,
Config::Items items, uint64_t *wal_seq_num) {
utils::MemoryTracker::OutOfMemoryExceptionEnabler oom_exception;
spdlog::info("Recovering persisted data using snapshot ({}) and WAL directory ({}).", snapshot_directory,
wal_directory);
if (!utils::DirExists(snapshot_directory) && !utils::DirExists(wal_directory)) {
spdlog::warn(utils::MessageWithLink("Snapshot or WAL directory don't exist, there is nothing to recover.",
"https://memgr.ph/durability"));
return std::nullopt;
}
auto snapshot_files = GetSnapshotFiles(snapshot_directory);
RecoveryInfo recovery_info;
RecoveredIndicesAndConstraints indices_constraints;
std::optional<uint64_t> snapshot_timestamp;
if (!snapshot_files.empty()) {
spdlog::info("Try recovering from snapshot directory {}.", snapshot_directory);
// Order the files by name
std::sort(snapshot_files.begin(), snapshot_files.end());
// UUID used for durability is the UUID of the last snapshot file.
*uuid = snapshot_files.back().uuid;
std::optional<RecoveredSnapshot> recovered_snapshot;
for (auto it = snapshot_files.rbegin(); it != snapshot_files.rend(); ++it) {
const auto &[path, file_uuid, _] = *it;
if (file_uuid != *uuid) {
spdlog::warn("The snapshot file {} isn't related to the latest snapshot file!", path);
continue;
}
spdlog::info("Starting snapshot recovery from {}.", path);
try {
recovered_snapshot = LoadSnapshot(path, vertices, edges, epoch_history, name_id_mapper, edge_count, items);
spdlog::info("Snapshot recovery successful!");
break;
} catch (const RecoveryFailure &e) {
spdlog::warn("Couldn't recover snapshot from {} because of: {}.", path, e.what());
continue;
}
}
MG_ASSERT(recovered_snapshot,
"The database is configured to recover on startup, but couldn't "
"recover using any of the specified snapshots! Please inspect them "
"and restart the database.");
recovery_info = recovered_snapshot->recovery_info;
indices_constraints = std::move(recovered_snapshot->indices_constraints);
snapshot_timestamp = recovered_snapshot->snapshot_info.start_timestamp;
*epoch_id = std::move(recovered_snapshot->snapshot_info.epoch_id);
if (!utils::DirExists(wal_directory)) {
RecoverIndicesAndConstraints(indices_constraints, indices, constraints, vertices);
return recovered_snapshot->recovery_info;
}
} else {
spdlog::info("No snapshot file was found, collecting information from WAL directory {}.", wal_directory);
std::error_code error_code;
if (!utils::DirExists(wal_directory)) return std::nullopt;
// We use this smaller struct that contains only a subset of information
// necessary for the rest of the recovery function.
// Also, the struct is sorted primarily on the path it contains.
struct WalFileInfo {
explicit WalFileInfo(std::filesystem::path path, std::string uuid, std::string epoch_id)
: path(std::move(path)), uuid(std::move(uuid)), epoch_id(std::move(epoch_id)) {}
std::filesystem::path path;
std::string uuid;
std::string epoch_id;
// NOLINTNEXTLINE(modernize-use-nullptr): bug in clang-tidy
auto operator<=>(const WalFileInfo &) const = default;
};
std::vector<WalFileInfo> wal_files;
for (const auto &item : std::filesystem::directory_iterator(wal_directory, error_code)) {
if (!item.is_regular_file()) continue;
try {
auto info = ReadWalInfo(item.path());
wal_files.emplace_back(item.path(), std::move(info.uuid), std::move(info.epoch_id));
} catch (const RecoveryFailure &e) {
continue;
}
}
MG_ASSERT(!error_code, "Couldn't recover data because an error occurred: {}!", error_code.message());
if (wal_files.empty()) {
spdlog::warn(utils::MessageWithLink("No snapshot or WAL file found.", "https://memgr.ph/durability"));
return std::nullopt;
}
std::sort(wal_files.begin(), wal_files.end());
// UUID used for durability is the UUID of the last WAL file.
// Same for the epoch id.
*uuid = std::move(wal_files.back().uuid);
*epoch_id = std::move(wal_files.back().epoch_id);
}
auto maybe_wal_files = GetWalFiles(wal_directory, *uuid);
if (!maybe_wal_files) {
spdlog::warn(
utils::MessageWithLink("Couldn't get WAL file info from the WAL directory.", "https://memgr.ph/durability"));
return std::nullopt;
}
// Array of all discovered WAL files, ordered by sequence number.
auto &wal_files = *maybe_wal_files;
// By this point we should have recovered from a snapshot, or we should have
// found some WAL files to recover from in the above `else`. This is just a
// sanity check to circumvent the following case: The database didn't recover
// from a snapshot, the above `else` triggered to find the recovery UUID from
// a WAL file. The above `else` has an early exit in case there are no WAL
// files. Because we reached this point there must have been some WAL files
// and we must have some WAL files after this second WAL directory iteration.
MG_ASSERT(snapshot_timestamp || !wal_files.empty(),
"The database didn't recover from a snapshot and didn't find any WAL "
"files that match the last WAL file!");
if (!wal_files.empty()) {
spdlog::info("Checking WAL files.");
{
const auto &first_wal = wal_files[0];
if (first_wal.seq_num != 0) {
// We don't have all WAL files. We need to see whether we need them all.
if (!snapshot_timestamp) {
// We didn't recover from a snapshot and we must have all WAL files
// starting from the first one (seq_num == 0) to be able to recover
// data from them.
LOG_FATAL(
"There are missing prefix WAL files and data can't be "
"recovered without them!");
} else if (first_wal.from_timestamp >= *snapshot_timestamp) {
// We recovered from a snapshot and we must have at least one WAL file
// that has at least one delta that was created before the snapshot in order to
// verify that nothing is missing from the beginning of the WAL chain.
LOG_FATAL(
"You must have at least one WAL file that contains at least one "
"delta that was created before the snapshot file!");
}
}
}
std::optional<uint64_t> previous_seq_num;
auto last_loaded_timestamp = snapshot_timestamp;
spdlog::info("Trying to load WAL files.");
for (auto &wal_file : wal_files) {
if (previous_seq_num && (wal_file.seq_num - *previous_seq_num) > 1) {
LOG_FATAL("You are missing a WAL file with the sequence number {}!", *previous_seq_num + 1);
}
previous_seq_num = wal_file.seq_num;
if (wal_file.epoch_id != *epoch_id) {
// This way we skip WALs finalized only because of role change.
// We can also set the last timestamp to 0 if last loaded timestamp
// is nullopt as this can only happen if the WAL file with seq = 0
// does not contain any deltas and we didn't find any snapshots.
if (last_loaded_timestamp) {
epoch_history->emplace_back(wal_file.epoch_id, *last_loaded_timestamp);
}
*epoch_id = std::move(wal_file.epoch_id);
}
try {
auto info = LoadWal(wal_file.path, &indices_constraints, last_loaded_timestamp, vertices, edges, name_id_mapper,
edge_count, items);
recovery_info.next_vertex_id = std::max(recovery_info.next_vertex_id, info.next_vertex_id);
recovery_info.next_edge_id = std::max(recovery_info.next_edge_id, info.next_edge_id);
recovery_info.next_timestamp = std::max(recovery_info.next_timestamp, info.next_timestamp);
recovery_info.last_commit_timestamp = info.last_commit_timestamp;
} catch (const RecoveryFailure &e) {
LOG_FATAL("Couldn't recover WAL deltas from {} because of: {}", wal_file.path, e.what());
}
if (recovery_info.next_timestamp != 0) {
last_loaded_timestamp.emplace(recovery_info.next_timestamp - 1);
}
}
// The sequence number needs to be recovered even though `LoadWal` didn't
// load any deltas from that file.
*wal_seq_num = *previous_seq_num + 1;
spdlog::info("All necessary WAL files are loaded successfully.");
}
RecoverIndicesAndConstraints(indices_constraints, indices, constraints, vertices);
return recovery_info;
}
} // namespace memgraph::storage::v3::durability

View File

@ -1,112 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <cstdint>
#include <filesystem>
#include <optional>
#include <string>
#include <variant>
#include "storage/v3/config.hpp"
#include "storage/v3/constraints.hpp"
#include "storage/v3/durability/metadata.hpp"
#include "storage/v3/durability/wal.hpp"
#include "storage/v3/edge.hpp"
#include "storage/v3/indices.hpp"
#include "storage/v3/name_id_mapper.hpp"
#include "storage/v3/vertex.hpp"
#include "utils/skip_list.hpp"
namespace memgraph::storage::v3::durability {
/// Verifies that the owner of the storage directory is the same user that
/// started the current process. If the verification fails, the process is
/// killed (`CHECK` failure).
void VerifyStorageDirectoryOwnerAndProcessUserOrDie(const std::filesystem::path &storage_directory);
// Used to capture the snapshot's data related to durability
struct SnapshotDurabilityInfo {
explicit SnapshotDurabilityInfo(std::filesystem::path path, std::string uuid, const uint64_t start_timestamp)
: path(std::move(path)), uuid(std::move(uuid)), start_timestamp(start_timestamp) {}
std::filesystem::path path;
std::string uuid;
uint64_t start_timestamp;
auto operator<=>(const SnapshotDurabilityInfo &) const = default;
};
/// Get list of snapshot files with their UUID.
/// @param snapshot_directory Directory containing the Snapshot files.
/// @param uuid UUID of the Snapshot files. If not empty, fetch only Snapshot
/// file with the specified UUID. Otherwise, fetch only Snapshot files in the
/// snapshot_directory.
/// @return List of snapshot files defined with its path and UUID.
std::vector<SnapshotDurabilityInfo> GetSnapshotFiles(const std::filesystem::path &snapshot_directory,
std::string_view uuid = "");
/// Used to capture a WAL's data related to durability
struct WalDurabilityInfo {
explicit WalDurabilityInfo(const uint64_t seq_num, const uint64_t from_timestamp, const uint64_t to_timestamp,
std::string uuid, std::string epoch_id, std::filesystem::path path)
: seq_num(seq_num),
from_timestamp(from_timestamp),
to_timestamp(to_timestamp),
uuid(std::move(uuid)),
epoch_id(std::move(epoch_id)),
path(std::move(path)) {}
uint64_t seq_num;
uint64_t from_timestamp;
uint64_t to_timestamp;
std::string uuid;
std::string epoch_id;
std::filesystem::path path;
auto operator<=>(const WalDurabilityInfo &) const = default;
};
/// Get list of WAL files ordered by the sequence number
/// @param wal_directory Directory containing the WAL files.
/// @param uuid UUID of the WAL files. If not empty, fetch only WAL files
/// with the specified UUID. Otherwise, fetch all WAL files in the
/// wal_directory.
/// @param current_seq_num Sequence number of the WAL file which is currently
/// being written. If specified, load only finalized WAL files, i.e. WAL files
/// with seq_num < current_seq_num.
/// @return List of WAL files. Each WAL file is defined with its sequence
/// number, from timestamp, to timestamp and path.
std::optional<std::vector<WalDurabilityInfo>> GetWalFiles(const std::filesystem::path &wal_directory,
std::string_view uuid = "",
std::optional<size_t> current_seq_num = {});
// Helper function used to recover all discovered indices and constraints. The
// indices and constraints must be recovered after the data recovery is done
// to ensure that the indices and constraints are consistent at the end of the
// recovery process.
/// @throw RecoveryFailure
void RecoverIndicesAndConstraints(const RecoveredIndicesAndConstraints &indices_constraints, Indices *indices,
Constraints *constraints, VerticesSkipList *vertices);
/// Recovers data either from a snapshot and/or WAL files.
/// @throw RecoveryFailure
/// @throw std::bad_alloc
std::optional<RecoveryInfo> RecoverData(const std::filesystem::path &snapshot_directory,
const std::filesystem::path &wal_directory, std::string *uuid,
std::string *epoch_id,
std::deque<std::pair<std::string, uint64_t>> *epoch_history,
VerticesSkipList *vertices, utils::SkipList<Edge> *edges, uint64_t *edge_count,
NameIdMapper *name_id_mapper, Indices *indices, Constraints *constraints,
Config::Items items, uint64_t *wal_seq_num);
} // namespace memgraph::storage::v3::durability

View File

@ -1,23 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include "utils/exceptions.hpp"
namespace memgraph::storage::v3::durability {
/// Exception used to handle errors during recovery.
class RecoveryFailure : public utils::BasicException {
using utils::BasicException::BasicException;
};
} // namespace memgraph::storage::v3::durability

View File

@ -1,106 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <cstdint>
namespace memgraph::storage::v3::durability {
/// Markers that are used to indicate crucial parts of the snapshot/WAL.
/// IMPORTANT: Don't forget to update the list of all markers `kMarkersAll` when
/// you add a new Marker.
enum class Marker : uint8_t {
TYPE_NULL = 0x10,
TYPE_BOOL = 0x11,
TYPE_INT = 0x12,
TYPE_DOUBLE = 0x13,
TYPE_STRING = 0x14,
TYPE_LIST = 0x15,
TYPE_MAP = 0x16,
TYPE_PROPERTY_VALUE = 0x17,
TYPE_TEMPORAL_DATA = 0x18,
SECTION_VERTEX = 0x20,
SECTION_EDGE = 0x21,
SECTION_MAPPER = 0x22,
SECTION_METADATA = 0x23,
SECTION_INDICES = 0x24,
SECTION_CONSTRAINTS = 0x25,
SECTION_DELTA = 0x26,
SECTION_EPOCH_HISTORY = 0x27,
SECTION_OFFSETS = 0x42,
DELTA_VERTEX_CREATE = 0x50,
DELTA_VERTEX_DELETE = 0x51,
DELTA_VERTEX_ADD_LABEL = 0x52,
DELTA_VERTEX_REMOVE_LABEL = 0x53,
DELTA_VERTEX_SET_PROPERTY = 0x54,
DELTA_EDGE_CREATE = 0x55,
DELTA_EDGE_DELETE = 0x56,
DELTA_EDGE_SET_PROPERTY = 0x57,
DELTA_TRANSACTION_END = 0x58,
DELTA_LABEL_INDEX_CREATE = 0x59,
DELTA_LABEL_INDEX_DROP = 0x5a,
DELTA_LABEL_PROPERTY_INDEX_CREATE = 0x5b,
DELTA_LABEL_PROPERTY_INDEX_DROP = 0x5c,
DELTA_EXISTENCE_CONSTRAINT_CREATE = 0x5d,
DELTA_EXISTENCE_CONSTRAINT_DROP = 0x5e,
DELTA_UNIQUE_CONSTRAINT_CREATE = 0x5f,
DELTA_UNIQUE_CONSTRAINT_DROP = 0x60,
VALUE_FALSE = 0x00,
VALUE_TRUE = 0xff,
};
/// List of all available markers.
/// IMPORTANT: Don't forget to update this list when you add a new Marker.
static const Marker kMarkersAll[] = {
Marker::TYPE_NULL,
Marker::TYPE_BOOL,
Marker::TYPE_INT,
Marker::TYPE_DOUBLE,
Marker::TYPE_STRING,
Marker::TYPE_LIST,
Marker::TYPE_MAP,
Marker::TYPE_TEMPORAL_DATA,
Marker::TYPE_PROPERTY_VALUE,
Marker::SECTION_VERTEX,
Marker::SECTION_EDGE,
Marker::SECTION_MAPPER,
Marker::SECTION_METADATA,
Marker::SECTION_INDICES,
Marker::SECTION_CONSTRAINTS,
Marker::SECTION_DELTA,
Marker::SECTION_EPOCH_HISTORY,
Marker::SECTION_OFFSETS,
Marker::DELTA_VERTEX_CREATE,
Marker::DELTA_VERTEX_DELETE,
Marker::DELTA_VERTEX_ADD_LABEL,
Marker::DELTA_VERTEX_REMOVE_LABEL,
Marker::DELTA_VERTEX_SET_PROPERTY,
Marker::DELTA_EDGE_CREATE,
Marker::DELTA_EDGE_DELETE,
Marker::DELTA_EDGE_SET_PROPERTY,
Marker::DELTA_TRANSACTION_END,
Marker::DELTA_LABEL_INDEX_CREATE,
Marker::DELTA_LABEL_INDEX_DROP,
Marker::DELTA_LABEL_PROPERTY_INDEX_CREATE,
Marker::DELTA_LABEL_PROPERTY_INDEX_DROP,
Marker::DELTA_EXISTENCE_CONSTRAINT_CREATE,
Marker::DELTA_EXISTENCE_CONSTRAINT_DROP,
Marker::DELTA_UNIQUE_CONSTRAINT_CREATE,
Marker::DELTA_UNIQUE_CONSTRAINT_DROP,
Marker::VALUE_FALSE,
Marker::VALUE_TRUE,
};
} // namespace memgraph::storage::v3::durability

View File

@ -1,75 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <algorithm>
#include <optional>
#include <set>
#include <utility>
#include <vector>
#include "storage/v3/durability/exceptions.hpp"
#include "storage/v3/id_types.hpp"
namespace memgraph::storage::v3::durability {
/// Structure used to hold metadata about the recovered snapshot/WAL.
struct RecoveryInfo {
uint64_t next_vertex_id{0};
uint64_t next_edge_id{0};
uint64_t next_timestamp{0};
// last timestamp read from a WAL file
std::optional<uint64_t> last_commit_timestamp;
};
/// Structure used to track indices and constraints during recovery.
struct RecoveredIndicesAndConstraints {
struct {
std::vector<LabelId> label;
std::vector<std::pair<LabelId, PropertyId>> label_property;
} indices;
struct {
std::vector<std::pair<LabelId, PropertyId>> existence;
std::vector<std::pair<LabelId, std::set<PropertyId>>> unique;
} constraints;
};
// Helper function used to insert indices/constraints into the recovered
// indices/constraints object.
// @throw RecoveryFailure
template <typename TObj>
void AddRecoveredIndexConstraint(std::vector<TObj> *list, TObj obj, const char *error_message) {
auto it = std::find(list->begin(), list->end(), obj);
if (it == list->end()) {
list->push_back(obj);
} else {
throw RecoveryFailure(error_message);
}
}
// Helper function used to remove indices/constraints from the recovered
// indices/constraints object.
// @throw RecoveryFailure
template <typename TObj>
void RemoveRecoveredIndexConstraint(std::vector<TObj> *list, TObj obj, const char *error_message) {
auto it = std::find(list->begin(), list->end(), obj);
if (it != list->end()) {
std::swap(*it, list->back());
list->pop_back();
} else {
throw RecoveryFailure(error_message);
}
}
} // namespace memgraph::storage::v3::durability

View File

@ -1,50 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <cstdint>
#include <string>
#include "utils/timestamp.hpp"
namespace memgraph::storage::v3::durability {
static const std::string kSnapshotDirectory{"snapshots"};
static const std::string kWalDirectory{"wal"};
static const std::string kBackupDirectory{".backup"};
static const std::string kLockFile{".lock"};
// This is the prefix used for Snapshot and WAL filenames. It is a timestamp
// format that equals to: YYYYmmddHHMMSSffffff
const std::string kTimestampFormat = "{:04d}{:02d}{:02d}{:02d}{:02d}{:02d}{:06d}";
// Generates the name for a snapshot in a well-defined sortable format with the
// start timestamp appended to the file name.
inline std::string MakeSnapshotName(uint64_t start_timestamp) {
std::string date_str = utils::Timestamp::Now().ToString(kTimestampFormat);
return date_str + "_timestamp_" + std::to_string(start_timestamp);
}
// Generates the name for a WAL file in a well-defined sortable format.
inline std::string MakeWalName() {
std::string date_str = utils::Timestamp::Now().ToString(kTimestampFormat);
return date_str + "_current";
}
// Generates the name for a WAL file in a well-defined sortable format with the
// range of timestamps contained [from, to] appended to the name.
inline std::string RemakeWalName(const std::string &current_name, uint64_t from_timestamp, uint64_t to_timestamp) {
return current_name.substr(0, current_name.size() - 8) + "_from_" + std::to_string(from_timestamp) + "_to_" +
std::to_string(to_timestamp);
}
} // namespace memgraph::storage::v3::durability

View File

@ -1,468 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include "storage/v3/durability/serialization.hpp"
#include "storage/v3/temporal.hpp"
#include "utils/endian.hpp"
namespace memgraph::storage::v3::durability {
//////////////////////////
// Encoder implementation.
//////////////////////////
namespace {
void WriteSize(Encoder *encoder, uint64_t size) {
size = utils::HostToLittleEndian(size);
encoder->Write(reinterpret_cast<const uint8_t *>(&size), sizeof(size));
}
} // namespace
void Encoder::Initialize(const std::filesystem::path &path, const std::string_view &magic, uint64_t version) {
file_.Open(path, utils::OutputFile::Mode::OVERWRITE_EXISTING);
Write(reinterpret_cast<const uint8_t *>(magic.data()), magic.size());
auto version_encoded = utils::HostToLittleEndian(version);
Write(reinterpret_cast<const uint8_t *>(&version_encoded), sizeof(version_encoded));
}
void Encoder::OpenExisting(const std::filesystem::path &path) {
file_.Open(path, utils::OutputFile::Mode::APPEND_TO_EXISTING);
}
void Encoder::Close() {
if (file_.IsOpen()) {
file_.Close();
}
}
void Encoder::Write(const uint8_t *data, uint64_t size) { file_.Write(data, size); }
void Encoder::WriteMarker(Marker marker) {
auto value = static_cast<uint8_t>(marker);
Write(&value, sizeof(value));
}
void Encoder::WriteBool(bool value) {
WriteMarker(Marker::TYPE_BOOL);
if (value) {
WriteMarker(Marker::VALUE_TRUE);
} else {
WriteMarker(Marker::VALUE_FALSE);
}
}
void Encoder::WriteUint(uint64_t value) {
value = utils::HostToLittleEndian(value);
WriteMarker(Marker::TYPE_INT);
Write(reinterpret_cast<const uint8_t *>(&value), sizeof(value));
}
void Encoder::WriteDouble(double value) {
auto value_uint = utils::MemcpyCast<uint64_t>(value);
value_uint = utils::HostToLittleEndian(value_uint);
WriteMarker(Marker::TYPE_DOUBLE);
Write(reinterpret_cast<const uint8_t *>(&value_uint), sizeof(value_uint));
}
void Encoder::WriteString(const std::string_view &value) {
WriteMarker(Marker::TYPE_STRING);
WriteSize(this, value.size());
Write(reinterpret_cast<const uint8_t *>(value.data()), value.size());
}
void Encoder::WritePropertyValue(const PropertyValue &value) {
WriteMarker(Marker::TYPE_PROPERTY_VALUE);
switch (value.type()) {
case PropertyValue::Type::Null: {
WriteMarker(Marker::TYPE_NULL);
break;
}
case PropertyValue::Type::Bool: {
WriteBool(value.ValueBool());
break;
}
case PropertyValue::Type::Int: {
WriteUint(utils::MemcpyCast<uint64_t>(value.ValueInt()));
break;
}
case PropertyValue::Type::Double: {
WriteDouble(value.ValueDouble());
break;
}
case PropertyValue::Type::String: {
WriteString(value.ValueString());
break;
}
case PropertyValue::Type::List: {
const auto &list = value.ValueList();
WriteMarker(Marker::TYPE_LIST);
WriteSize(this, list.size());
for (const auto &item : list) {
WritePropertyValue(item);
}
break;
}
case PropertyValue::Type::Map: {
const auto &map = value.ValueMap();
WriteMarker(Marker::TYPE_MAP);
WriteSize(this, map.size());
for (const auto &item : map) {
WriteString(item.first);
WritePropertyValue(item.second);
}
break;
}
case PropertyValue::Type::TemporalData: {
const auto temporal_data = value.ValueTemporalData();
WriteMarker(Marker::TYPE_TEMPORAL_DATA);
WriteUint(static_cast<uint64_t>(temporal_data.type));
WriteUint(utils::MemcpyCast<uint64_t>(temporal_data.microseconds));
break;
}
}
}
uint64_t Encoder::GetPosition() { return file_.GetPosition(); }
void Encoder::SetPosition(uint64_t position) {
file_.SetPosition(utils::OutputFile::Position::SET, static_cast<ssize_t>(position));
}
void Encoder::Sync() { file_.Sync(); }
void Encoder::Finalize() {
file_.Sync();
file_.Close();
}
void Encoder::DisableFlushing() { file_.DisableFlushing(); }
void Encoder::EnableFlushing() { file_.EnableFlushing(); }
void Encoder::TryFlushing() { file_.TryFlushing(); }
std::pair<const uint8_t *, size_t> Encoder::CurrentFileBuffer() const { return file_.CurrentBuffer(); }
size_t Encoder::GetSize() { return file_.GetSize(); }
//////////////////////////
// Decoder implementation.
//////////////////////////
namespace {
std::optional<Marker> CastToMarker(uint8_t value) {
for (auto marker : kMarkersAll) {
if (static_cast<uint8_t>(marker) == value) {
return marker;
}
}
return std::nullopt;
}
std::optional<uint64_t> ReadSize(Decoder *decoder) {
uint64_t size{0};
if (!decoder->Read(reinterpret_cast<uint8_t *>(&size), sizeof(size))) return std::nullopt;
size = utils::LittleEndianToHost(size);
return size;
}
} // namespace
std::optional<uint64_t> Decoder::Initialize(const std::filesystem::path &path, const std::string &magic) {
if (!file_.Open(path)) return std::nullopt;
std::string file_magic(magic.size(), '\0');
if (!Read(reinterpret_cast<uint8_t *>(file_magic.data()), file_magic.size())) return std::nullopt;
if (file_magic != magic) return std::nullopt;
uint64_t version_encoded{0};
if (!Read(reinterpret_cast<uint8_t *>(&version_encoded), sizeof(version_encoded))) return std::nullopt;
return utils::LittleEndianToHost(version_encoded);
}
bool Decoder::Read(uint8_t *data, size_t size) { return file_.Read(data, size); }
bool Decoder::Peek(uint8_t *data, size_t size) { return file_.Peek(data, size); }
std::optional<Marker> Decoder::PeekMarker() {
uint8_t value{0};
if (!Peek(&value, sizeof(value))) return std::nullopt;
auto marker = CastToMarker(value);
if (!marker) return std::nullopt;
return *marker;
}
std::optional<Marker> Decoder::ReadMarker() {
uint8_t value{0};
if (!Read(&value, sizeof(value))) return std::nullopt;
auto marker = CastToMarker(value);
if (!marker) return std::nullopt;
return *marker;
}
std::optional<bool> Decoder::ReadBool() {
auto marker = ReadMarker();
if (!marker || *marker != Marker::TYPE_BOOL) return std::nullopt;
auto value = ReadMarker();
if (!value || (*value != Marker::VALUE_FALSE && *value != Marker::VALUE_TRUE)) return std::nullopt;
return *value == Marker::VALUE_TRUE;
}
std::optional<uint64_t> Decoder::ReadUint() {
auto marker = ReadMarker();
if (!marker || *marker != Marker::TYPE_INT) return std::nullopt;
uint64_t value{0};
if (!Read(reinterpret_cast<uint8_t *>(&value), sizeof(value))) return std::nullopt;
value = utils::LittleEndianToHost(value);
return value;
}
std::optional<double> Decoder::ReadDouble() {
auto marker = ReadMarker();
if (!marker || *marker != Marker::TYPE_DOUBLE) return std::nullopt;
uint64_t value_int{0};
if (!Read(reinterpret_cast<uint8_t *>(&value_int), sizeof(value_int))) return std::nullopt;
value_int = utils::LittleEndianToHost(value_int);
auto value = utils::MemcpyCast<double>(value_int);
return value;
}
std::optional<std::string> Decoder::ReadString() {
auto marker = ReadMarker();
if (!marker || *marker != Marker::TYPE_STRING) return std::nullopt;
auto size = ReadSize(this);
if (!size) return std::nullopt;
std::string value(*size, '\0');
if (!Read(reinterpret_cast<uint8_t *>(value.data()), *size)) return std::nullopt;
return value;
}
namespace {
std::optional<TemporalData> ReadTemporalData(Decoder &decoder) {
const auto inner_marker = decoder.ReadMarker();
if (!inner_marker || *inner_marker != Marker::TYPE_TEMPORAL_DATA) return std::nullopt;
const auto type = decoder.ReadUint();
if (!type) return std::nullopt;
const auto microseconds = decoder.ReadUint();
if (!microseconds) return std::nullopt;
return TemporalData{static_cast<TemporalType>(*type), utils::MemcpyCast<int64_t>(*microseconds)};
}
} // namespace
std::optional<PropertyValue> Decoder::ReadPropertyValue() {
auto pv_marker = ReadMarker();
if (!pv_marker || *pv_marker != Marker::TYPE_PROPERTY_VALUE) return std::nullopt;
auto marker = PeekMarker();
if (!marker) return std::nullopt;
switch (*marker) {
case Marker::TYPE_NULL: {
auto inner_marker = ReadMarker();
if (!inner_marker || *inner_marker != Marker::TYPE_NULL) return std::nullopt;
return PropertyValue();
}
case Marker::TYPE_BOOL: {
auto value = ReadBool();
if (!value) return std::nullopt;
return PropertyValue(*value);
}
case Marker::TYPE_INT: {
auto value = ReadUint();
if (!value) return std::nullopt;
return PropertyValue(utils::MemcpyCast<int64_t>(*value));
}
case Marker::TYPE_DOUBLE: {
auto value = ReadDouble();
if (!value) return std::nullopt;
return PropertyValue(*value);
}
case Marker::TYPE_STRING: {
auto value = ReadString();
if (!value) return std::nullopt;
return PropertyValue(std::move(*value));
}
case Marker::TYPE_LIST: {
auto inner_marker = ReadMarker();
if (!inner_marker || *inner_marker != Marker::TYPE_LIST) return std::nullopt;
auto size = ReadSize(this);
if (!size) return std::nullopt;
std::vector<PropertyValue> value;
value.reserve(*size);
for (uint64_t i = 0; i < *size; ++i) {
auto item = ReadPropertyValue();
if (!item) return std::nullopt;
value.emplace_back(std::move(*item));
}
return PropertyValue(std::move(value));
}
case Marker::TYPE_MAP: {
auto inner_marker = ReadMarker();
if (!inner_marker || *inner_marker != Marker::TYPE_MAP) return std::nullopt;
auto size = ReadSize(this);
if (!size) return std::nullopt;
std::map<std::string, PropertyValue> value;
for (uint64_t i = 0; i < *size; ++i) {
auto key = ReadString();
if (!key) return std::nullopt;
auto item = ReadPropertyValue();
if (!item) return std::nullopt;
value.emplace(std::move(*key), std::move(*item));
}
return PropertyValue(std::move(value));
}
case Marker::TYPE_TEMPORAL_DATA: {
const auto maybe_temporal_data = ReadTemporalData(*this);
if (!maybe_temporal_data) return std::nullopt;
return PropertyValue(*maybe_temporal_data);
}
case Marker::TYPE_PROPERTY_VALUE:
case Marker::SECTION_VERTEX:
case Marker::SECTION_EDGE:
case Marker::SECTION_MAPPER:
case Marker::SECTION_METADATA:
case Marker::SECTION_INDICES:
case Marker::SECTION_CONSTRAINTS:
case Marker::SECTION_DELTA:
case Marker::SECTION_EPOCH_HISTORY:
case Marker::SECTION_OFFSETS:
case Marker::DELTA_VERTEX_CREATE:
case Marker::DELTA_VERTEX_DELETE:
case Marker::DELTA_VERTEX_ADD_LABEL:
case Marker::DELTA_VERTEX_REMOVE_LABEL:
case Marker::DELTA_VERTEX_SET_PROPERTY:
case Marker::DELTA_EDGE_CREATE:
case Marker::DELTA_EDGE_DELETE:
case Marker::DELTA_EDGE_SET_PROPERTY:
case Marker::DELTA_TRANSACTION_END:
case Marker::DELTA_LABEL_INDEX_CREATE:
case Marker::DELTA_LABEL_INDEX_DROP:
case Marker::DELTA_LABEL_PROPERTY_INDEX_CREATE:
case Marker::DELTA_LABEL_PROPERTY_INDEX_DROP:
case Marker::DELTA_EXISTENCE_CONSTRAINT_CREATE:
case Marker::DELTA_EXISTENCE_CONSTRAINT_DROP:
case Marker::DELTA_UNIQUE_CONSTRAINT_CREATE:
case Marker::DELTA_UNIQUE_CONSTRAINT_DROP:
case Marker::VALUE_FALSE:
case Marker::VALUE_TRUE:
return std::nullopt;
}
}
bool Decoder::SkipString() {
auto marker = ReadMarker();
if (!marker || *marker != Marker::TYPE_STRING) return false;
auto maybe_size = ReadSize(this);
if (!maybe_size) return false;
const uint64_t kBufferSize = 262144;
uint8_t buffer[kBufferSize];
uint64_t size = *maybe_size;
while (size > 0) {
uint64_t to_read = size < kBufferSize ? size : kBufferSize;
if (!Read(reinterpret_cast<uint8_t *>(&buffer), to_read)) return false;
size -= to_read;
}
return true;
}
bool Decoder::SkipPropertyValue() {
auto pv_marker = ReadMarker();
if (!pv_marker || *pv_marker != Marker::TYPE_PROPERTY_VALUE) return false;
auto marker = PeekMarker();
if (!marker) return false;
switch (*marker) {
case Marker::TYPE_NULL: {
auto inner_marker = ReadMarker();
return inner_marker && *inner_marker == Marker::TYPE_NULL;
}
case Marker::TYPE_BOOL: {
return !!ReadBool();
}
case Marker::TYPE_INT: {
return !!ReadUint();
}
case Marker::TYPE_DOUBLE: {
return !!ReadDouble();
}
case Marker::TYPE_STRING: {
return SkipString();
}
case Marker::TYPE_LIST: {
auto inner_marker = ReadMarker();
if (!inner_marker || *inner_marker != Marker::TYPE_LIST) return false;
auto size = ReadSize(this);
if (!size) return false;
for (uint64_t i = 0; i < *size; ++i) {
if (!SkipPropertyValue()) return false;
}
return true;
}
case Marker::TYPE_MAP: {
auto inner_marker = ReadMarker();
if (!inner_marker || *inner_marker != Marker::TYPE_MAP) return false;
auto size = ReadSize(this);
if (!size) return false;
for (uint64_t i = 0; i < *size; ++i) {
if (!SkipString()) return false;
if (!SkipPropertyValue()) return false;
}
return true;
}
case Marker::TYPE_TEMPORAL_DATA: {
return !!ReadTemporalData(*this);
}
case Marker::TYPE_PROPERTY_VALUE:
case Marker::SECTION_VERTEX:
case Marker::SECTION_EDGE:
case Marker::SECTION_MAPPER:
case Marker::SECTION_METADATA:
case Marker::SECTION_INDICES:
case Marker::SECTION_CONSTRAINTS:
case Marker::SECTION_DELTA:
case Marker::SECTION_EPOCH_HISTORY:
case Marker::SECTION_OFFSETS:
case Marker::DELTA_VERTEX_CREATE:
case Marker::DELTA_VERTEX_DELETE:
case Marker::DELTA_VERTEX_ADD_LABEL:
case Marker::DELTA_VERTEX_REMOVE_LABEL:
case Marker::DELTA_VERTEX_SET_PROPERTY:
case Marker::DELTA_EDGE_CREATE:
case Marker::DELTA_EDGE_DELETE:
case Marker::DELTA_EDGE_SET_PROPERTY:
case Marker::DELTA_TRANSACTION_END:
case Marker::DELTA_LABEL_INDEX_CREATE:
case Marker::DELTA_LABEL_INDEX_DROP:
case Marker::DELTA_LABEL_PROPERTY_INDEX_CREATE:
case Marker::DELTA_LABEL_PROPERTY_INDEX_DROP:
case Marker::DELTA_EXISTENCE_CONSTRAINT_CREATE:
case Marker::DELTA_EXISTENCE_CONSTRAINT_DROP:
case Marker::DELTA_UNIQUE_CONSTRAINT_CREATE:
case Marker::DELTA_UNIQUE_CONSTRAINT_DROP:
case Marker::VALUE_FALSE:
case Marker::VALUE_TRUE:
return false;
}
}
std::optional<uint64_t> Decoder::GetSize() { return file_.GetSize(); }
std::optional<uint64_t> Decoder::GetPosition() { return file_.GetPosition(); }
bool Decoder::SetPosition(uint64_t position) {
return !!file_.SetPosition(utils::InputFile::Position::SET, static_cast<ssize_t>(position));
}
} // namespace memgraph::storage::v3::durability

View File

@ -1,143 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <cstdint>
#include <filesystem>
#include <string_view>
#include "storage/v3/config.hpp"
#include "storage/v3/durability/marker.hpp"
#include "storage/v3/name_id_mapper.hpp"
#include "storage/v3/property_value.hpp"
#include "utils/file.hpp"
namespace memgraph::storage::v3::durability {
/// Encoder interface class. Used to implement streams to different targets
/// (e.g. file and network).
class BaseEncoder {
protected:
BaseEncoder() = default;
~BaseEncoder() = default;
public:
BaseEncoder(const BaseEncoder &) = delete;
BaseEncoder(BaseEncoder &&) = delete;
BaseEncoder &operator=(const BaseEncoder &) = delete;
BaseEncoder &operator=(BaseEncoder &&) = delete;
virtual void WriteMarker(Marker marker) = 0;
virtual void WriteBool(bool value) = 0;
virtual void WriteUint(uint64_t value) = 0;
virtual void WriteDouble(double value) = 0;
virtual void WriteString(const std::string_view &value) = 0;
virtual void WritePropertyValue(const PropertyValue &value) = 0;
};
/// Encoder that is used to generate a snapshot/WAL.
class Encoder final : public BaseEncoder {
public:
void Initialize(const std::filesystem::path &path, const std::string_view &magic, uint64_t version);
void OpenExisting(const std::filesystem::path &path);
void Close();
// Main write function, the only one that is allowed to write to the `file_`
// directly.
void Write(const uint8_t *data, uint64_t size);
void WriteMarker(Marker marker) override;
void WriteBool(bool value) override;
void WriteUint(uint64_t value) override;
void WriteDouble(double value) override;
void WriteString(const std::string_view &value) override;
void WritePropertyValue(const PropertyValue &value) override;
uint64_t GetPosition();
void SetPosition(uint64_t position);
void Sync();
void Finalize();
// Disable flushing of the internal buffer.
void DisableFlushing();
// Enable flushing of the internal buffer.
void EnableFlushing();
// Try flushing the internal buffer.
void TryFlushing();
// Get the current internal buffer with its size.
std::pair<const uint8_t *, size_t> CurrentFileBuffer() const;
// Get the total size of the current file.
size_t GetSize();
private:
utils::OutputFile file_;
};
/// Decoder interface class. Used to implement streams from different sources
/// (e.g. file and network).
class BaseDecoder {
protected:
BaseDecoder() = default;
~BaseDecoder() = default;
public:
BaseDecoder(const BaseDecoder &) = delete;
BaseDecoder(BaseDecoder &&) = delete;
BaseDecoder &operator=(const BaseDecoder &) = delete;
BaseDecoder &operator=(BaseDecoder &&) = delete;
virtual std::optional<Marker> ReadMarker() = 0;
virtual std::optional<bool> ReadBool() = 0;
virtual std::optional<uint64_t> ReadUint() = 0;
virtual std::optional<double> ReadDouble() = 0;
virtual std::optional<std::string> ReadString() = 0;
virtual std::optional<PropertyValue> ReadPropertyValue() = 0;
virtual bool SkipString() = 0;
virtual bool SkipPropertyValue() = 0;
};
/// Decoder that is used to read a generated snapshot/WAL.
class Decoder final : public BaseDecoder {
public:
std::optional<uint64_t> Initialize(const std::filesystem::path &path, const std::string &magic);
// Main read functions, the only one that are allowed to read from the `file_`
// directly.
bool Read(uint8_t *data, size_t size);
bool Peek(uint8_t *data, size_t size);
std::optional<Marker> PeekMarker();
std::optional<Marker> ReadMarker() override;
std::optional<bool> ReadBool() override;
std::optional<uint64_t> ReadUint() override;
std::optional<double> ReadDouble() override;
std::optional<std::string> ReadString() override;
std::optional<PropertyValue> ReadPropertyValue() override;
bool SkipString() override;
bool SkipPropertyValue() override;
std::optional<uint64_t> GetSize();
std::optional<uint64_t> GetPosition();
bool SetPosition(uint64_t position);
private:
utils::InputFile file_;
};
} // namespace memgraph::storage::v3::durability

File diff suppressed because it is too large Load Diff

View File

@ -1,77 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <cstdint>
#include <filesystem>
#include <string>
#include "storage/v3/config.hpp"
#include "storage/v3/constraints.hpp"
#include "storage/v3/durability/metadata.hpp"
#include "storage/v3/edge.hpp"
#include "storage/v3/indices.hpp"
#include "storage/v3/name_id_mapper.hpp"
#include "storage/v3/schema_validator.hpp"
#include "storage/v3/transaction.hpp"
#include "storage/v3/vertex.hpp"
#include "utils/file_locker.hpp"
#include "utils/skip_list.hpp"
namespace memgraph::storage::v3::durability {
/// Structure used to hold information about a snapshot.
struct SnapshotInfo {
uint64_t offset_edges;
uint64_t offset_vertices;
uint64_t offset_indices;
uint64_t offset_constraints;
uint64_t offset_mapper;
uint64_t offset_epoch_history;
uint64_t offset_metadata;
std::string uuid;
std::string epoch_id;
uint64_t start_timestamp;
uint64_t edges_count;
uint64_t vertices_count;
};
/// Structure used to hold information about the snapshot that has been
/// recovered.
struct RecoveredSnapshot {
SnapshotInfo snapshot_info;
RecoveryInfo recovery_info;
RecoveredIndicesAndConstraints indices_constraints;
};
/// Function used to read information about the snapshot file.
/// @throw RecoveryFailure
SnapshotInfo ReadSnapshotInfo(const std::filesystem::path &path);
/// Function used to load the snapshot data into the storage.
/// @throw RecoveryFailure
RecoveredSnapshot LoadSnapshot(const std::filesystem::path &path, VerticesSkipList *vertices,
utils::SkipList<Edge> *edges,
std::deque<std::pair<std::string, uint64_t>> *epoch_history,
NameIdMapper *name_id_mapper, uint64_t *edge_count, Config::Items items);
/// Function used to create a snapshot using the given transaction.
void CreateSnapshot(Transaction *transaction, const std::filesystem::path &snapshot_directory,
const std::filesystem::path &wal_directory, uint64_t snapshot_retention_count,
VerticesSkipList *vertices, utils::SkipList<Edge> *edges, NameIdMapper *name_id_mapper,
Indices *indices, Constraints *constraints, Config::Items items,
const SchemaValidator &schema_validator, const std::string &uuid, std::string_view epoch_id,
const std::deque<std::pair<std::string, uint64_t>> &epoch_history,
utils::FileRetainer *file_retainer);
} // namespace memgraph::storage::v3::durability

View File

@ -1,37 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <cstdint>
#include <string>
#include <type_traits>
namespace memgraph::storage::v3::durability {
// The current version of snapshot and WAL encoding / decoding.
// IMPORTANT: Please bump this version for every snapshot and/or WAL format
// change!!!
const uint64_t kVersion{14};
const uint64_t kOldestSupportedVersion{14};
const uint64_t kUniqueConstraintVersion{13};
// Magic values written to the start of a snapshot/WAL file to identify it.
const std::string kSnapshotMagic{"MGsn"};
const std::string kWalMagic{"MGwl"};
static_assert(std::is_same_v<uint8_t, unsigned char>);
// Checks whether the loaded snapshot/WAL version is supported.
inline bool IsVersionSupported(uint64_t version) { return version >= kOldestSupportedVersion && version <= kVersion; }
} // namespace memgraph::storage::v3::durability

File diff suppressed because it is too large Load Diff

View File

@ -1,262 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <cstdint>
#include <filesystem>
#include <set>
#include <string>
#include "storage/v3/config.hpp"
#include "storage/v3/delta.hpp"
#include "storage/v3/durability/metadata.hpp"
#include "storage/v3/durability/serialization.hpp"
#include "storage/v3/edge.hpp"
#include "storage/v3/id_types.hpp"
#include "storage/v3/name_id_mapper.hpp"
#include "storage/v3/property_value.hpp"
#include "storage/v3/vertex.hpp"
#include "storage/v3/vertices_skip_list.hpp"
#include "utils/file_locker.hpp"
#include "utils/skip_list.hpp"
namespace memgraph::storage::v3::durability {
/// Structure used to hold information about a WAL.
struct WalInfo {
uint64_t offset_metadata;
uint64_t offset_deltas;
std::string uuid;
std::string epoch_id;
uint64_t seq_num;
uint64_t from_timestamp;
uint64_t to_timestamp;
uint64_t num_deltas;
};
/// Structure used to return loaded WAL delta data.
struct WalDeltaData {
enum class Type {
VERTEX_CREATE,
VERTEX_DELETE,
VERTEX_ADD_LABEL,
VERTEX_REMOVE_LABEL,
VERTEX_SET_PROPERTY,
EDGE_CREATE,
EDGE_DELETE,
EDGE_SET_PROPERTY,
TRANSACTION_END,
LABEL_INDEX_CREATE,
LABEL_INDEX_DROP,
LABEL_PROPERTY_INDEX_CREATE,
LABEL_PROPERTY_INDEX_DROP,
EXISTENCE_CONSTRAINT_CREATE,
EXISTENCE_CONSTRAINT_DROP,
UNIQUE_CONSTRAINT_CREATE,
UNIQUE_CONSTRAINT_DROP,
};
Type type{Type::TRANSACTION_END};
struct {
Gid gid;
} vertex_create_delete;
struct {
Gid gid;
std::string label;
} vertex_add_remove_label;
struct {
Gid gid;
std::string property;
PropertyValue value;
} vertex_edge_set_property;
struct {
Gid gid;
std::string edge_type;
Gid from_vertex;
Gid to_vertex;
} edge_create_delete;
struct {
std::string label;
} operation_label;
struct {
std::string label;
std::string property;
} operation_label_property;
struct {
std::string label;
std::set<std::string> properties;
} operation_label_properties;
};
bool operator==(const WalDeltaData &a, const WalDeltaData &b);
bool operator!=(const WalDeltaData &a, const WalDeltaData &b);
/// Enum used to indicate a global database operation that isn't transactional.
enum class StorageGlobalOperation {
LABEL_INDEX_CREATE,
LABEL_INDEX_DROP,
LABEL_PROPERTY_INDEX_CREATE,
LABEL_PROPERTY_INDEX_DROP,
EXISTENCE_CONSTRAINT_CREATE,
EXISTENCE_CONSTRAINT_DROP,
UNIQUE_CONSTRAINT_CREATE,
UNIQUE_CONSTRAINT_DROP,
};
constexpr bool IsWalDeltaDataTypeTransactionEnd(const WalDeltaData::Type type) {
switch (type) {
// These delta actions are all found inside transactions so they don't
// indicate a transaction end.
case WalDeltaData::Type::VERTEX_CREATE:
case WalDeltaData::Type::VERTEX_DELETE:
case WalDeltaData::Type::VERTEX_ADD_LABEL:
case WalDeltaData::Type::VERTEX_REMOVE_LABEL:
case WalDeltaData::Type::EDGE_CREATE:
case WalDeltaData::Type::EDGE_DELETE:
case WalDeltaData::Type::VERTEX_SET_PROPERTY:
case WalDeltaData::Type::EDGE_SET_PROPERTY:
return false;
// This delta explicitly indicates that a transaction is done.
case WalDeltaData::Type::TRANSACTION_END:
// These operations aren't transactional and they are encoded only using
// a single delta, so they each individually mark the end of their
// 'transaction'.
case WalDeltaData::Type::LABEL_INDEX_CREATE:
case WalDeltaData::Type::LABEL_INDEX_DROP:
case WalDeltaData::Type::LABEL_PROPERTY_INDEX_CREATE:
case WalDeltaData::Type::LABEL_PROPERTY_INDEX_DROP:
case WalDeltaData::Type::EXISTENCE_CONSTRAINT_CREATE:
case WalDeltaData::Type::EXISTENCE_CONSTRAINT_DROP:
case WalDeltaData::Type::UNIQUE_CONSTRAINT_CREATE:
case WalDeltaData::Type::UNIQUE_CONSTRAINT_DROP:
return true;
}
}
/// Function used to read information about the WAL file.
/// @throw RecoveryFailure
WalInfo ReadWalInfo(const std::filesystem::path &path);
/// Function used to read the WAL delta header. The function returns the delta
/// timestamp.
/// @throw RecoveryFailure
uint64_t ReadWalDeltaHeader(BaseDecoder *decoder);
/// Function used to read the current WAL delta data. The function returns the
/// read delta data. The WAL delta header must be read before calling this
/// function.
/// @throw RecoveryFailure
WalDeltaData ReadWalDeltaData(BaseDecoder *decoder);
/// Function used to skip the current WAL delta data. The function returns the
/// skipped delta type. The WAL delta header must be read before calling this
/// function.
/// @throw RecoveryFailure
WalDeltaData::Type SkipWalDeltaData(BaseDecoder *decoder);
/// Function used to encode a `Delta` that originated from a `Vertex`.
void EncodeDelta(BaseEncoder *encoder, NameIdMapper *name_id_mapper, Config::Items items, const Delta &delta,
const Vertex &vertex, uint64_t timestamp);
/// Function used to encode a `Delta` that originated from an `Edge`.
void EncodeDelta(BaseEncoder *encoder, NameIdMapper *name_id_mapper, const Delta &delta, const Edge &edge,
uint64_t timestamp);
/// Function used to encode the transaction end.
void EncodeTransactionEnd(BaseEncoder *encoder, uint64_t timestamp);
/// Function used to encode non-transactional operation.
void EncodeOperation(BaseEncoder *encoder, NameIdMapper *name_id_mapper, StorageGlobalOperation operation,
LabelId label, const std::set<PropertyId> &properties, uint64_t timestamp);
/// Function used to load the WAL data into the storage.
/// @throw RecoveryFailure
RecoveryInfo LoadWal(const std::filesystem::path &path, RecoveredIndicesAndConstraints *indices_constraints,
std::optional<uint64_t> last_loaded_timestamp, VerticesSkipList *vertices,
utils::SkipList<Edge> *edges, NameIdMapper *name_id_mapper, uint64_t *edge_count,
Config::Items items);
/// WalFile class used to append deltas and operations to the WAL file.
class WalFile {
public:
WalFile(const std::filesystem::path &wal_directory, std::string_view uuid, std::string_view epoch_id,
Config::Items items, NameIdMapper *name_id_mapper, uint64_t seq_num, utils::FileRetainer *file_retainer);
WalFile(std::filesystem::path current_wal_path, Config::Items items, NameIdMapper *name_id_mapper, uint64_t seq_num,
uint64_t from_timestamp, uint64_t to_timestamp, uint64_t count, utils::FileRetainer *file_retainer);
WalFile(const WalFile &) = delete;
WalFile(WalFile &&) = delete;
WalFile &operator=(const WalFile &) = delete;
WalFile &operator=(WalFile &&) = delete;
~WalFile();
void AppendDelta(const Delta &delta, const Vertex &vertex, uint64_t timestamp);
void AppendDelta(const Delta &delta, const Edge &edge, uint64_t timestamp);
void AppendTransactionEnd(uint64_t timestamp);
void AppendOperation(StorageGlobalOperation operation, LabelId label, const std::set<PropertyId> &properties,
uint64_t timestamp);
void Sync();
uint64_t GetSize();
uint64_t SequenceNumber() const;
auto FromTimestamp() const { return from_timestamp_; }
auto ToTimestamp() const { return to_timestamp_; }
auto Count() const { return count_; }
// Disable flushing of the internal buffer.
void DisableFlushing();
// Enable flushing of the internal buffer.
void EnableFlushing();
// Try flushing the internal buffer.
void TryFlushing();
// Get the internal buffer with its size.
std::pair<const uint8_t *, size_t> CurrentFileBuffer() const;
// Get the path of the current WAL file.
const auto &Path() const { return path_; }
void FinalizeWal();
void DeleteWal();
private:
void UpdateStats(uint64_t timestamp);
Config::Items items_;
NameIdMapper *name_id_mapper_;
Encoder wal_;
std::filesystem::path path_;
uint64_t from_timestamp_;
uint64_t to_timestamp_;
uint64_t count_;
uint64_t seq_num_;
utils::FileRetainer *file_retainer_;
};
} // namespace memgraph::storage::v3::durability

View File

@ -28,7 +28,6 @@ namespace memgraph::storage::v3 {
struct Vertex;
class VertexAccessor;
struct Indices;
struct Constraints;
class EdgeAccessor final {
private:
@ -36,14 +35,13 @@ class EdgeAccessor final {
public:
EdgeAccessor(EdgeRef edge, EdgeTypeId edge_type, VertexId from_vertex, VertexId to_vertex, Transaction *transaction,
Indices *indices, Constraints *constraints, Config::Items config, bool for_deleted = false)
Indices *indices, Config::Items config, bool for_deleted = false)
: edge_(edge),
edge_type_(edge_type),
from_vertex_(std::move(from_vertex)),
to_vertex_(std::move(to_vertex)),
transaction_(transaction),
indices_(indices),
constraints_(constraints),
config_(config),
for_deleted_(for_deleted) {}
@ -93,7 +91,6 @@ class EdgeAccessor final {
VertexId to_vertex_;
Transaction *transaction_;
Indices *indices_;
Constraints *constraints_;
Config::Items config_;
// if the accessor was created for a deleted edge.

View File

@ -32,16 +32,16 @@ namespace {
template <typename TCallback>
bool AnyVersionSatisfiesPredicate(uint64_t timestamp, const Delta *delta, const TCallback &predicate) {
while (delta != nullptr) {
auto ts = delta->timestamp->load(std::memory_order_acquire);
const auto commit_info = *delta->commit_info;
// This is a committed change that we see so we shouldn't undo it.
if (ts < timestamp) {
if (commit_info.is_locally_committed && commit_info.start_or_commit_timestamp.logical_id < timestamp) {
break;
}
if (predicate(*delta)) {
return true;
}
// Move to the next delta.
delta = delta->next.load(std::memory_order_acquire);
delta = delta->next;
}
return false;
}
@ -266,7 +266,7 @@ void LabelIndex::UpdateOnAddLabel(LabelId label, Vertex *vertex, const Transacti
auto it = index_.find(label);
if (it == index_.end()) return;
auto acc = it->second.access();
acc.insert(Entry{vertex, tx.start_timestamp});
acc.insert(Entry{vertex, tx.start_timestamp.logical_id});
}
bool LabelIndex::CreateIndex(LabelId label, VerticesSkipList::Accessor vertices) {
@ -302,20 +302,20 @@ std::vector<LabelId> LabelIndex::ListIndices() const {
return ret;
}
void LabelIndex::RemoveObsoleteEntries(uint64_t oldest_active_start_timestamp) {
void LabelIndex::RemoveObsoleteEntries(const uint64_t clean_up_before_timestamp) {
for (auto &label_storage : index_) {
auto vertices_acc = label_storage.second.access();
for (auto it = vertices_acc.begin(); it != vertices_acc.end();) {
auto next_it = it;
++next_it;
if (it->timestamp >= oldest_active_start_timestamp) {
if (it->timestamp >= clean_up_before_timestamp) {
it = next_it;
continue;
}
if ((next_it != vertices_acc.end() && it->vertex == next_it->vertex) ||
!AnyVersionHasLabel(*it->vertex, label_storage.first, oldest_active_start_timestamp)) {
!AnyVersionHasLabel(*it->vertex, label_storage.first, clean_up_before_timestamp)) {
vertices_acc.remove(*it);
}
@ -327,7 +327,7 @@ void LabelIndex::RemoveObsoleteEntries(uint64_t oldest_active_start_timestamp) {
LabelIndex::Iterable::Iterator::Iterator(Iterable *self, utils::SkipList<Entry>::Iterator index_iterator)
: self_(self),
index_iterator_(index_iterator),
current_vertex_accessor_(nullptr, nullptr, nullptr, nullptr, self_->config_, *self_->vertex_validator_),
current_vertex_accessor_(nullptr, nullptr, nullptr, self_->config_, *self_->vertex_validator_),
current_vertex_(nullptr) {
AdvanceUntilValid();
}
@ -345,22 +345,21 @@ void LabelIndex::Iterable::Iterator::AdvanceUntilValid() {
}
if (CurrentVersionHasLabel(*index_iterator_->vertex, self_->label_, self_->transaction_, self_->view_)) {
current_vertex_ = index_iterator_->vertex;
current_vertex_accessor_ = VertexAccessor{current_vertex_, self_->transaction_, self_->indices_,
self_->constraints_, self_->config_, *self_->vertex_validator_};
current_vertex_accessor_ = VertexAccessor{current_vertex_, self_->transaction_, self_->indices_, self_->config_,
*self_->vertex_validator_};
break;
}
}
}
LabelIndex::Iterable::Iterable(utils::SkipList<Entry>::Accessor index_accessor, LabelId label, View view,
Transaction *transaction, Indices *indices, Constraints *constraints,
Config::Items config, const VertexValidator &vertex_validator)
Transaction *transaction, Indices *indices, Config::Items config,
const VertexValidator &vertex_validator)
: index_accessor_(std::move(index_accessor)),
label_(label),
view_(view),
transaction_(transaction),
indices_(indices),
constraints_(constraints),
config_(config),
vertex_validator_(&vertex_validator) {}
@ -396,7 +395,7 @@ void LabelPropertyIndex::UpdateOnAddLabel(LabelId label, Vertex *vertex, const T
auto prop_value = vertex->properties.GetProperty(label_prop.second);
if (!prop_value.IsNull()) {
auto acc = storage.access();
acc.insert(Entry{std::move(prop_value), vertex, tx.start_timestamp});
acc.insert(Entry{std::move(prop_value), vertex, tx.start_timestamp.logical_id});
}
}
}
@ -412,7 +411,7 @@ void LabelPropertyIndex::UpdateOnSetProperty(PropertyId property, const Property
}
if (utils::Contains(vertex->labels, label_prop.first)) {
auto acc = storage.access();
acc.insert(Entry{value, vertex, tx.start_timestamp});
acc.insert(Entry{value, vertex, tx.start_timestamp.logical_id});
}
}
}
@ -455,21 +454,21 @@ std::vector<std::pair<LabelId, PropertyId>> LabelPropertyIndex::ListIndices() co
return ret;
}
void LabelPropertyIndex::RemoveObsoleteEntries(uint64_t oldest_active_start_timestamp) {
void LabelPropertyIndex::RemoveObsoleteEntries(const uint64_t clean_up_before_timestamp) {
for (auto &[label_property, index] : index_) {
auto index_acc = index.access();
for (auto it = index_acc.begin(); it != index_acc.end();) {
auto next_it = it;
++next_it;
if (it->timestamp >= oldest_active_start_timestamp) {
if (it->timestamp >= clean_up_before_timestamp) {
it = next_it;
continue;
}
if ((next_it != index_acc.end() && it->vertex == next_it->vertex && it->value == next_it->value) ||
!AnyVersionHasLabelProperty(*it->vertex, label_property.first, label_property.second, it->value,
oldest_active_start_timestamp)) {
clean_up_before_timestamp)) {
index_acc.remove(*it);
}
it = next_it;
@ -480,7 +479,7 @@ void LabelPropertyIndex::RemoveObsoleteEntries(uint64_t oldest_active_start_time
LabelPropertyIndex::Iterable::Iterator::Iterator(Iterable *self, utils::SkipList<Entry>::Iterator index_iterator)
: self_(self),
index_iterator_(index_iterator),
current_vertex_accessor_(nullptr, nullptr, nullptr, nullptr, self_->config_, *self_->vertex_validator_),
current_vertex_accessor_(nullptr, nullptr, nullptr, self_->config_, *self_->vertex_validator_),
current_vertex_(nullptr) {
AdvanceUntilValid();
}
@ -519,8 +518,8 @@ void LabelPropertyIndex::Iterable::Iterator::AdvanceUntilValid() {
if (CurrentVersionHasLabelProperty(*index_iterator_->vertex, self_->label_, self_->property_,
index_iterator_->value, self_->transaction_, self_->view_)) {
current_vertex_ = index_iterator_->vertex;
current_vertex_accessor_ = VertexAccessor(current_vertex_, self_->transaction_, self_->indices_,
self_->constraints_, self_->config_, *self_->vertex_validator_);
current_vertex_accessor_ = VertexAccessor(current_vertex_, self_->transaction_, self_->indices_, self_->config_,
*self_->vertex_validator_);
break;
}
}
@ -542,8 +541,8 @@ LabelPropertyIndex::Iterable::Iterable(utils::SkipList<Entry>::Accessor index_ac
PropertyId property,
const std::optional<utils::Bound<PropertyValue>> &lower_bound,
const std::optional<utils::Bound<PropertyValue>> &upper_bound, View view,
Transaction *transaction, Indices *indices, Constraints *constraints,
Config::Items config, const VertexValidator &vertex_validator)
Transaction *transaction, Indices *indices, Config::Items config,
const VertexValidator &vertex_validator)
: index_accessor_(std::move(index_accessor)),
label_(label),
property_(property),
@ -552,7 +551,6 @@ LabelPropertyIndex::Iterable::Iterable(utils::SkipList<Entry>::Accessor index_ac
view_(view),
transaction_(transaction),
indices_(indices),
constraints_(constraints),
config_(config),
vertex_validator_(&vertex_validator) {
// We have to fix the bounds that the user provided to us. If the user
@ -698,9 +696,9 @@ void LabelPropertyIndex::RunGC() {
}
}
void RemoveObsoleteEntries(Indices *indices, uint64_t oldest_active_start_timestamp) {
indices->label_index.RemoveObsoleteEntries(oldest_active_start_timestamp);
indices->label_property_index.RemoveObsoleteEntries(oldest_active_start_timestamp);
void RemoveObsoleteEntries(Indices *indices, const uint64_t clean_up_before_timestamp) {
indices->label_index.RemoveObsoleteEntries(clean_up_before_timestamp);
indices->label_property_index.RemoveObsoleteEntries(clean_up_before_timestamp);
}
void UpdateOnAddLabel(Indices *indices, LabelId label, Vertex *vertex, const Transaction &tx) {

View File

@ -28,7 +28,6 @@
namespace memgraph::storage::v3 {
struct Indices;
struct Constraints;
class LabelIndex {
private:
@ -53,8 +52,8 @@ class LabelIndex {
};
public:
LabelIndex(Indices *indices, Constraints *constraints, Config::Items config, const VertexValidator &vertex_validator)
: indices_(indices), constraints_(constraints), config_(config), vertex_validator_{&vertex_validator} {}
LabelIndex(Indices *indices, Config::Items config, const VertexValidator &vertex_validator)
: indices_(indices), config_(config), vertex_validator_{&vertex_validator} {}
/// @throw std::bad_alloc
void UpdateOnAddLabel(LabelId label, Vertex *vertex, const Transaction &tx);
@ -69,12 +68,12 @@ class LabelIndex {
std::vector<LabelId> ListIndices() const;
void RemoveObsoleteEntries(uint64_t oldest_active_start_timestamp);
void RemoveObsoleteEntries(uint64_t clean_up_before_timestamp);
class Iterable {
public:
Iterable(utils::SkipList<Entry>::Accessor index_accessor, LabelId label, View view, Transaction *transaction,
Indices *indices, Constraints *constraints, Config::Items config, const VertexValidator &vertex_validator);
Indices *indices, Config::Items config, const VertexValidator &vertex_validator);
class Iterator {
public:
@ -105,7 +104,6 @@ class LabelIndex {
View view_;
Transaction *transaction_;
Indices *indices_;
Constraints *constraints_;
Config::Items config_;
const VertexValidator *vertex_validator_;
};
@ -114,7 +112,7 @@ class LabelIndex {
Iterable Vertices(LabelId label, View view, Transaction *transaction) {
auto it = index_.find(label);
MG_ASSERT(it != index_.end(), "Index for label {} doesn't exist", label.AsUint());
return {it->second.access(), label, view, transaction, indices_, constraints_, config_, *vertex_validator_};
return {it->second.access(), label, view, transaction, indices_, config_, *vertex_validator_};
}
int64_t ApproximateVertexCount(LabelId label) {
@ -130,7 +128,6 @@ class LabelIndex {
private:
std::map<LabelId, utils::SkipList<Entry>> index_;
Indices *indices_;
Constraints *constraints_;
Config::Items config_;
const VertexValidator *vertex_validator_;
};
@ -150,9 +147,8 @@ class LabelPropertyIndex {
};
public:
LabelPropertyIndex(Indices *indices, Constraints *constraints, Config::Items config,
const VertexValidator &vertex_validator)
: indices_(indices), constraints_(constraints), config_(config), vertex_validator_{&vertex_validator} {}
LabelPropertyIndex(Indices *indices, Config::Items config, const VertexValidator &vertex_validator)
: indices_(indices), config_(config), vertex_validator_{&vertex_validator} {}
/// @throw std::bad_alloc
void UpdateOnAddLabel(LabelId label, Vertex *vertex, const Transaction &tx);
@ -169,14 +165,14 @@ class LabelPropertyIndex {
std::vector<std::pair<LabelId, PropertyId>> ListIndices() const;
void RemoveObsoleteEntries(uint64_t oldest_active_start_timestamp);
void RemoveObsoleteEntries(uint64_t clean_up_before_timestamp);
class Iterable {
public:
Iterable(utils::SkipList<Entry>::Accessor index_accessor, LabelId label, PropertyId property,
const std::optional<utils::Bound<PropertyValue>> &lower_bound,
const std::optional<utils::Bound<PropertyValue>> &upper_bound, View view, Transaction *transaction,
Indices *indices, Constraints *constraints, Config::Items config, const VertexValidator &vertex_validator);
Indices *indices, Config::Items config, const VertexValidator &vertex_validator);
class Iterator {
public:
@ -211,7 +207,6 @@ class LabelPropertyIndex {
View view_;
Transaction *transaction_;
Indices *indices_;
Constraints *constraints_;
Config::Items config_;
const VertexValidator *vertex_validator_;
};
@ -222,8 +217,8 @@ class LabelPropertyIndex {
auto it = index_.find({label, property});
MG_ASSERT(it != index_.end(), "Index for label {} and property {} doesn't exist", label.AsUint(),
property.AsUint());
return {it->second.access(), label, property, lower_bound, upper_bound, view,
transaction, indices_, constraints_, config_, *vertex_validator_};
return {it->second.access(), label, property, lower_bound, upper_bound, view,
transaction, indices_, config_, *vertex_validator_};
}
int64_t ApproximateVertexCount(LabelId label, PropertyId property) const {
@ -250,15 +245,13 @@ class LabelPropertyIndex {
private:
std::map<std::pair<LabelId, PropertyId>, utils::SkipList<Entry>> index_;
Indices *indices_;
Constraints *constraints_;
Config::Items config_;
const VertexValidator *vertex_validator_;
};
struct Indices {
Indices(Constraints *constraints, Config::Items config, const VertexValidator &vertex_validator)
: label_index(this, constraints, config, vertex_validator),
label_property_index(this, constraints, config, vertex_validator) {}
Indices(Config::Items config, const VertexValidator &vertex_validator)
: label_index(this, config, vertex_validator), label_property_index(this, config, vertex_validator) {}
// Disable copy and move because members hold pointer to `this`.
Indices(const Indices &) = delete;
@ -273,7 +266,7 @@ struct Indices {
/// This function should be called from garbage collection to clean-up the
/// index.
void RemoveObsoleteEntries(Indices *indices, uint64_t oldest_active_start_timestamp);
void RemoveObsoleteEntries(Indices *indices, uint64_t clean_up_before_timestamp);
// Indices are updated whenever an update occurs, instead of only on commit or
// advance command. This is necessary because we want indices to support `NEW`

View File

@ -28,11 +28,9 @@ inline void ApplyDeltasForRead(Transaction *transaction, const Delta *delta, Vie
// if the transaction is not committed, then its deltas have transaction_id for the timestamp, otherwise they have
// its commit timestamp set.
// This allows the transaction to see its changes even though it's committed.
const auto commit_timestamp = transaction->commit_timestamp
? transaction->commit_timestamp->load(std::memory_order_acquire)
: transaction->transaction_id;
const auto &commit_info = *transaction->commit_info;
while (delta != nullptr) {
auto ts = delta->timestamp->load(std::memory_order_acquire);
const auto &delta_commit_info = *delta->commit_info;
auto cid = delta->command_id;
// For SNAPSHOT ISOLATION -> we can only see the changes which were committed before the start of the current
@ -44,21 +42,24 @@ inline void ApplyDeltasForRead(Transaction *transaction, const Delta *delta, Vie
// id value, that the change is committed.
//
// For READ UNCOMMITTED -> we accept any change.
if ((transaction->isolation_level == IsolationLevel::SNAPSHOT_ISOLATION && ts < transaction->start_timestamp) ||
(transaction->isolation_level == IsolationLevel::READ_COMMITTED && ts < kTransactionInitialId) ||
if ((transaction->isolation_level == IsolationLevel::SNAPSHOT_ISOLATION && delta_commit_info.is_locally_committed &&
delta_commit_info.start_or_commit_timestamp.logical_id < transaction->start_timestamp.logical_id) ||
(transaction->isolation_level == IsolationLevel::READ_COMMITTED && delta_commit_info.is_locally_committed) ||
(transaction->isolation_level == IsolationLevel::READ_UNCOMMITTED)) {
break;
}
// We shouldn't undo our newest changes because the user requested a NEW
// view of the database.
if (view == View::NEW && ts == commit_timestamp && cid <= transaction->command_id) {
if (view == View::NEW && delta_commit_info.start_or_commit_timestamp == commit_info.start_or_commit_timestamp &&
cid <= transaction->command_id) {
break;
}
// We shouldn't undo our older changes because the user requested a OLD view
// of the database.
if (view == View::OLD && ts == commit_timestamp && cid < transaction->command_id) {
if (view == View::OLD && delta_commit_info.start_or_commit_timestamp == commit_info.start_or_commit_timestamp &&
delta->command_id < transaction->command_id) {
break;
}
@ -66,7 +67,7 @@ inline void ApplyDeltasForRead(Transaction *transaction, const Delta *delta, Vie
callback(*delta);
// Move to the next delta.
delta = delta->next.load(std::memory_order_acquire);
delta = delta->next;
}
}
@ -79,8 +80,10 @@ template <typename TObj>
inline bool PrepareForWrite(Transaction *transaction, TObj *object) {
if (object->delta == nullptr) return true;
auto ts = object->delta->timestamp->load(std::memory_order_acquire);
if (ts == transaction->transaction_id || ts < transaction->start_timestamp) {
const auto &delta_commit_info = *object->delta->commit_info;
if (delta_commit_info.start_or_commit_timestamp == transaction->commit_info->start_or_commit_timestamp ||
(delta_commit_info.is_locally_committed &&
delta_commit_info.start_or_commit_timestamp < transaction->start_timestamp)) {
return true;
}
@ -94,8 +97,7 @@ inline bool PrepareForWrite(Transaction *transaction, TObj *object) {
/// a `DELETE_OBJECT` delta).
/// @throw std::bad_alloc
inline Delta *CreateDeleteObjectDelta(Transaction *transaction) {
transaction->EnsureCommitTimestampExists();
return &transaction->deltas.emplace_back(Delta::DeleteObjectTag(), transaction->commit_timestamp.get(),
return &transaction->deltas.emplace_back(Delta::DeleteObjectTag(), transaction->commit_info.get(),
transaction->command_id);
}
@ -104,8 +106,7 @@ inline Delta *CreateDeleteObjectDelta(Transaction *transaction) {
/// @throw std::bad_alloc
template <typename TObj, class... Args>
inline void CreateAndLinkDelta(Transaction *transaction, TObj *object, Args &&...args) {
transaction->EnsureCommitTimestampExists();
auto delta = &transaction->deltas.emplace_back(std::forward<Args>(args)..., transaction->commit_timestamp.get(),
auto delta = &transaction->deltas.emplace_back(std::forward<Args>(args)..., transaction->commit_info.get(),
transaction->command_id);
// The operations are written in such order so that both `next` and `prev`
@ -117,7 +118,7 @@ inline void CreateAndLinkDelta(Transaction *transaction, TObj *object, Args &&..
// TODO(antaljanosbenjamin): clang-tidy detects (in my opinion a false positive) issue in
// `Shard::Accessor::CreateEdge`.
// NOLINTNEXTLINE(clang-analyzer-core.NullDereference)
delta->next.store(object->delta, std::memory_order_release);
delta->next = object->delta;
// 2. We need to set the previous delta of the new delta to the object.
delta->prev.Set(object);
// 3. We need to set the previous delta of the existing delta to the new

View File

@ -1,2 +0,0 @@
# autogenerated files
rpc.hpp

View File

@ -1,44 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <chrono>
#include <optional>
#include <string>
namespace memgraph::storage::v3::replication {
struct ReplicationClientConfig {
std::optional<double> timeout;
// The default delay between main checking/pinging replicas is 1s because
// that seems like a reasonable timeframe in which main should notice a
// replica is down.
std::chrono::seconds replica_check_frequency{1};
struct SSL {
std::string key_file;
std::string cert_file;
};
std::optional<SSL> ssl;
};
struct ReplicationServerConfig {
struct SSL {
std::string key_file;
std::string cert_file;
std::string ca_file;
bool verify_peer;
};
std::optional<SSL> ssl;
};
} // namespace memgraph::storage::v3::replication

View File

@ -1,19 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <cstdint>
namespace memgraph::storage::v3::replication {
enum class ReplicationMode : std::uint8_t { SYNC, ASYNC };
enum class ReplicaState : std::uint8_t { READY, REPLICATING, RECOVERY, INVALID };
} // namespace memgraph::storage::v3::replication

View File

@ -1,617 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include "storage/v3/replication/replication_client.hpp"
#include <algorithm>
#include <type_traits>
#include "storage/v3/durability/durability.hpp"
#include "storage/v3/replication/config.hpp"
#include "storage/v3/replication/enums.hpp"
#include "storage/v3/transaction.hpp"
#include "utils/file_locker.hpp"
#include "utils/logging.hpp"
#include "utils/message.hpp"
namespace memgraph::storage::v3 {
namespace {
template <typename>
[[maybe_unused]] inline constexpr bool always_false_v = false;
} // namespace
////// ReplicationClient //////
Shard::ReplicationClient::ReplicationClient(std::string name, Shard *shard, const io::network::Endpoint &endpoint,
const replication::ReplicationMode mode,
const replication::ReplicationClientConfig &config)
: name_(std::move(name)), shard_(shard), mode_(mode) {
if (config.ssl) {
rpc_context_.emplace(config.ssl->key_file, config.ssl->cert_file);
} else {
rpc_context_.emplace();
}
rpc_client_.emplace(endpoint, &*rpc_context_);
TryInitializeClientSync();
if (config.timeout && replica_state_ != replication::ReplicaState::INVALID) {
timeout_.emplace(*config.timeout);
timeout_dispatcher_.emplace();
}
// Help the user to get the most accurate replica state possible.
if (config.replica_check_frequency > std::chrono::seconds(0)) {
replica_checker_.Run("Replica Checker", config.replica_check_frequency, [&] { FrequentCheck(); });
}
}
void Shard::ReplicationClient::TryInitializeClientAsync() {
thread_pool_.AddTask([this] {
rpc_client_->Abort();
this->TryInitializeClientSync();
});
}
void Shard::ReplicationClient::FrequentCheck() {
const auto is_success = std::invoke([this]() {
try {
auto stream{rpc_client_->Stream<replication::FrequentHeartbeatRpc>()};
const auto response = stream.AwaitResponse();
return response.success;
} catch (const rpc::RpcFailedException &) {
return false;
}
});
// States: READY, REPLICATING, RECOVERY, INVALID
// If success && ready, replicating, recovery -> stay the same because something good is going on.
// If success && INVALID -> [it's possible that replica came back to life] -> TryInitializeClient.
// If fail -> [replica is not reachable at all] -> INVALID state.
// NOTE: TryInitializeClient might return nothing if there is a branching point.
// NOTE: The early return pattern simplified the code, but the behavior should be as explained.
if (!is_success) {
replica_state_.store(replication::ReplicaState::INVALID);
return;
}
if (replica_state_.load() == replication::ReplicaState::INVALID) {
TryInitializeClientAsync();
}
}
/// @throws rpc::RpcFailedException
void Shard::ReplicationClient::InitializeClient() {
uint64_t current_commit_timestamp{kTimestampInitialId};
auto stream{rpc_client_->Stream<replication::HeartbeatRpc>(shard_->last_commit_timestamp_, shard_->epoch_id_)};
const auto response = stream.AwaitResponse();
std::optional<uint64_t> branching_point;
if (response.epoch_id != shard_->epoch_id_ && response.current_commit_timestamp != kTimestampInitialId) {
const auto &epoch_history = shard_->epoch_history_;
const auto epoch_info_iter =
std::find_if(epoch_history.crbegin(), epoch_history.crend(),
[&](const auto &epoch_info) { return epoch_info.first == response.epoch_id; });
if (epoch_info_iter == epoch_history.crend()) {
branching_point = 0;
} else if (epoch_info_iter->second != response.current_commit_timestamp) {
branching_point = epoch_info_iter->second;
}
}
if (branching_point) {
spdlog::error(
"Replica {} cannot be used with this instance. Please start a clean "
"instance of Memgraph server on the specified endpoint.",
name_);
return;
}
current_commit_timestamp = response.current_commit_timestamp;
spdlog::trace("Current timestamp on replica: {}", current_commit_timestamp);
spdlog::trace("Current timestamp on main: {}", shard_->last_commit_timestamp_);
if (current_commit_timestamp == shard_->last_commit_timestamp_) {
spdlog::debug("Replica '{}' up to date", name_);
std::unique_lock client_guard{client_lock_};
replica_state_.store(replication::ReplicaState::READY);
} else {
spdlog::debug("Replica '{}' is behind", name_);
{
std::unique_lock client_guard{client_lock_};
replica_state_.store(replication::ReplicaState::RECOVERY);
}
thread_pool_.AddTask([=, this] { this->RecoverReplica(current_commit_timestamp); });
}
}
void Shard::ReplicationClient::TryInitializeClientSync() {
try {
InitializeClient();
} catch (const rpc::RpcFailedException &) {
std::unique_lock client_guarde{client_lock_};
replica_state_.store(replication::ReplicaState::INVALID);
spdlog::error(utils::MessageWithLink("Failed to connect to replica {} at the endpoint {}.", name_,
rpc_client_->Endpoint(), "https://memgr.ph/replication"));
}
}
void Shard::ReplicationClient::HandleRpcFailure() {
spdlog::error(utils::MessageWithLink("Couldn't replicate data to {}.", name_, "https://memgr.ph/replication"));
TryInitializeClientAsync();
}
replication::SnapshotRes Shard::ReplicationClient::TransferSnapshot(const std::filesystem::path &path) {
auto stream{rpc_client_->Stream<replication::SnapshotRpc>()};
replication::Encoder encoder(stream.GetBuilder());
encoder.WriteFile(path);
return stream.AwaitResponse();
}
replication::WalFilesRes Shard::ReplicationClient::TransferWalFiles(
const std::vector<std::filesystem::path> &wal_files) {
MG_ASSERT(!wal_files.empty(), "Wal files list is empty!");
auto stream{rpc_client_->Stream<replication::WalFilesRpc>(wal_files.size())};
replication::Encoder encoder(stream.GetBuilder());
for (const auto &wal : wal_files) {
spdlog::debug("Sending wal file: {}", wal);
encoder.WriteFile(wal);
}
return stream.AwaitResponse();
}
void Shard::ReplicationClient::StartTransactionReplication(const uint64_t current_wal_seq_num) {
std::unique_lock guard(client_lock_);
const auto status = replica_state_.load();
switch (status) {
case replication::ReplicaState::RECOVERY:
spdlog::debug("Replica {} is behind MAIN instance", name_);
return;
case replication::ReplicaState::REPLICATING:
spdlog::debug("Replica {} missed a transaction", name_);
// We missed a transaction because we're still replicating
// the previous transaction so we need to go to RECOVERY
// state to catch up with the missing transaction
// We cannot queue the recovery process here because
// an error can happen while we're replicating the previous
// transaction after which the client should go to
// INVALID state before starting the recovery process
replica_state_.store(replication::ReplicaState::RECOVERY);
return;
case replication::ReplicaState::INVALID:
HandleRpcFailure();
return;
case replication::ReplicaState::READY:
MG_ASSERT(!replica_stream_);
try {
replica_stream_.emplace(ReplicaStream{this, shard_->last_commit_timestamp_, current_wal_seq_num});
replica_state_.store(replication::ReplicaState::REPLICATING);
} catch (const rpc::RpcFailedException &) {
replica_state_.store(replication::ReplicaState::INVALID);
HandleRpcFailure();
}
return;
}
}
void Shard::ReplicationClient::IfStreamingTransaction(const std::function<void(ReplicaStream &handler)> &callback) {
// We can only check the state because it guarantees to be only
// valid during a single transaction replication (if the assumption
// that this and other transaction replication functions can only be
// called from a one thread stands)
if (replica_state_ != replication::ReplicaState::REPLICATING) {
return;
}
try {
callback(*replica_stream_);
} catch (const rpc::RpcFailedException &) {
{
std::unique_lock client_guard{client_lock_};
replica_state_.store(replication::ReplicaState::INVALID);
}
HandleRpcFailure();
}
}
void Shard::ReplicationClient::FinalizeTransactionReplication() {
// We can only check the state because it guarantees to be only
// valid during a single transaction replication (if the assumption
// that this and other transaction replication functions can only be
// called from a one thread stands)
if (replica_state_ != replication::ReplicaState::REPLICATING) {
return;
}
if (mode_ == replication::ReplicationMode::ASYNC) {
thread_pool_.AddTask([this] { this->FinalizeTransactionReplicationInternal(); });
} else if (timeout_) {
MG_ASSERT(mode_ == replication::ReplicationMode::SYNC, "Only SYNC replica can have a timeout.");
MG_ASSERT(timeout_dispatcher_, "Timeout thread is missing");
timeout_dispatcher_->WaitForTaskToFinish();
timeout_dispatcher_->active = true;
thread_pool_.AddTask([&, this] {
this->FinalizeTransactionReplicationInternal();
std::unique_lock main_guard(timeout_dispatcher_->main_lock);
// TimerThread can finish waiting for timeout
timeout_dispatcher_->active = false;
// Notify the main thread
timeout_dispatcher_->main_cv.notify_one();
});
timeout_dispatcher_->StartTimeoutTask(*timeout_);
// Wait until one of the threads notifies us that they finished executing
// Both threads should first set the active flag to false
{
std::unique_lock main_guard(timeout_dispatcher_->main_lock);
timeout_dispatcher_->main_cv.wait(main_guard, [&] { return !timeout_dispatcher_->active.load(); });
}
// TODO (antonio2368): Document and/or polish SEMI-SYNC to ASYNC fallback.
if (replica_state_ == replication::ReplicaState::REPLICATING) {
mode_ = replication::ReplicationMode::ASYNC;
timeout_.reset();
// This can only happen if we timeouted so we are sure that
// Timeout task finished
// We need to delete timeout dispatcher AFTER the replication
// finished because it tries to acquire the timeout lock
// and acces the `active` variable`
thread_pool_.AddTask([this] { timeout_dispatcher_.reset(); });
}
} else {
FinalizeTransactionReplicationInternal();
}
}
void Shard::ReplicationClient::FinalizeTransactionReplicationInternal() {
MG_ASSERT(replica_stream_, "Missing stream for transaction deltas");
try {
auto response = replica_stream_->Finalize();
replica_stream_.reset();
std::unique_lock client_guard(client_lock_);
if (!response.success || replica_state_ == replication::ReplicaState::RECOVERY) {
replica_state_.store(replication::ReplicaState::RECOVERY);
thread_pool_.AddTask([&, this] { this->RecoverReplica(response.current_commit_timestamp); });
} else {
replica_state_.store(replication::ReplicaState::READY);
}
} catch (const rpc::RpcFailedException &) {
replica_stream_.reset();
{
std::unique_lock client_guard(client_lock_);
replica_state_.store(replication::ReplicaState::INVALID);
}
HandleRpcFailure();
}
}
void Shard::ReplicationClient::RecoverReplica(uint64_t replica_commit) {
while (true) {
auto file_locker = shard_->file_retainer_.AddLocker();
const auto steps = GetRecoverySteps(replica_commit, &file_locker);
for (const auto &recovery_step : steps) {
try {
std::visit(
[&, this]<typename T>(T &&arg) {
using StepType = std::remove_cvref_t<T>;
if constexpr (std::is_same_v<StepType, RecoverySnapshot>) {
spdlog::debug("Sending the latest snapshot file: {}", arg);
auto response = TransferSnapshot(arg);
replica_commit = response.current_commit_timestamp;
} else if constexpr (std::is_same_v<StepType, RecoveryWals>) {
spdlog::debug("Sending the latest wal files");
auto response = TransferWalFiles(arg);
replica_commit = response.current_commit_timestamp;
} else if constexpr (std::is_same_v<StepType, RecoveryCurrentWal>) {
if (shard_->wal_file_ && shard_->wal_file_->SequenceNumber() == arg.current_wal_seq_num) {
shard_->wal_file_->DisableFlushing();
spdlog::debug("Sending current wal file");
replica_commit = ReplicateCurrentWal();
shard_->wal_file_->EnableFlushing();
}
} else {
static_assert(always_false_v<T>, "Missing type from variant visitor");
}
},
recovery_step);
} catch (const rpc::RpcFailedException &) {
{
std::unique_lock client_guard{client_lock_};
replica_state_.store(replication::ReplicaState::INVALID);
}
HandleRpcFailure();
return;
}
}
spdlog::trace("Current timestamp on replica: {}", replica_commit);
// To avoid the situation where we read a correct commit timestamp in
// one thread, and after that another thread commits a different a
// transaction and THEN we set the state to READY in the first thread,
// we set this lock before checking the timestamp.
// We will detect that the state is invalid during the next commit,
// because replication::AppendDeltasRpc sends the last commit timestamp which
// replica checks if it's the same last commit timestamp it received
// and we will go to recovery.
// By adding this lock, we can avoid that, and go to RECOVERY immediately.
std::unique_lock client_guard{client_lock_};
SPDLOG_INFO("Replica timestamp: {}", replica_commit);
SPDLOG_INFO("Last commit: {}", shard_->last_commit_timestamp_);
if (shard_->last_commit_timestamp_ == replica_commit) {
replica_state_.store(replication::ReplicaState::READY);
return;
}
}
}
uint64_t Shard::ReplicationClient::ReplicateCurrentWal() {
const auto &wal_file = shard_->wal_file_;
auto stream = TransferCurrentWalFile();
stream.AppendFilename(wal_file->Path().filename());
utils::InputFile file;
MG_ASSERT(file.Open(shard_->wal_file_->Path()), "Failed to open current WAL file!");
const auto [buffer, buffer_size] = wal_file->CurrentFileBuffer();
stream.AppendSize(file.GetSize() + buffer_size);
stream.AppendFileData(&file);
stream.AppendBufferData(buffer, buffer_size);
auto response = stream.Finalize();
return response.current_commit_timestamp;
}
/// This method tries to find the optimal path for recoverying a single replica.
/// Based on the last commit transfered to replica it tries to update the
/// replica using durability files - WALs and Snapshots. WAL files are much
/// smaller in size as they contain only the Deltas (changes) made during the
/// transactions while Snapshots contain all the data. For that reason we prefer
/// WALs as much as possible. As the WAL file that is currently being updated
/// can change during the process we ignore it as much as possible. Also, it
/// uses the transaction lock so lokcing it can be really expensive. After we
/// fetch the list of finalized WALs, we try to find the longest chain of
/// sequential WALs, starting from the latest one, that will update the recovery
/// with the all missed updates. If the WAL chain cannot be created, replica is
/// behind by a lot, so we use the regular recovery process, we send the latest
/// snapshot and all the necessary WAL files, starting from the newest WAL that
/// contains a timestamp before the snapshot. If we registered the existence of
/// the current WAL, we add the sequence number we read from it to the recovery
/// process. After all the other steps are finished, if the current WAL contains
/// the same sequence number, it's the same WAL we read while fetching the
/// recovery steps, so we can safely send it to the replica.
/// We assume that the property of preserving at least 1 WAL before the snapshot
/// is satisfied as we extract the timestamp information from it.
std::vector<Shard::ReplicationClient::RecoveryStep> Shard::ReplicationClient::GetRecoverySteps(
const uint64_t replica_commit, utils::FileRetainer::FileLocker *file_locker) {
// First check if we can recover using the current wal file only
// otherwise save the seq_num of the current wal file
// This lock is also necessary to force the missed transaction to finish.
std::optional<uint64_t> current_wal_seq_num;
std::optional<uint64_t> current_wal_from_timestamp;
if (shard_->wal_file_) {
current_wal_seq_num.emplace(shard_->wal_file_->SequenceNumber());
current_wal_from_timestamp.emplace(shard_->wal_file_->FromTimestamp());
}
auto locker_acc = file_locker->Access();
auto wal_files = durability::GetWalFiles(shard_->wal_directory_, shard_->uuid_, current_wal_seq_num);
MG_ASSERT(wal_files, "Wal files could not be loaded");
auto snapshot_files = durability::GetSnapshotFiles(shard_->snapshot_directory_, shard_->uuid_);
std::optional<durability::SnapshotDurabilityInfo> latest_snapshot;
if (!snapshot_files.empty()) {
std::sort(snapshot_files.begin(), snapshot_files.end());
latest_snapshot.emplace(std::move(snapshot_files.back()));
}
std::vector<RecoveryStep> recovery_steps;
// No finalized WAL files were found. This means the difference is contained
// inside the current WAL or the snapshot.
if (wal_files->empty()) {
if (current_wal_from_timestamp && replica_commit >= *current_wal_from_timestamp) {
MG_ASSERT(current_wal_seq_num);
recovery_steps.emplace_back(RecoveryCurrentWal{*current_wal_seq_num});
return recovery_steps;
}
// Without the finalized WAL containing the current timestamp of replica,
// we cannot know if the difference is only in the current WAL or we need
// to send the snapshot.
if (latest_snapshot) {
locker_acc.AddPath(latest_snapshot->path);
recovery_steps.emplace_back(std::in_place_type_t<RecoverySnapshot>{}, std::move(latest_snapshot->path));
}
// if there are no finalized WAL files, snapshot left the current WAL
// as the WAL file containing a transaction before snapshot creation
// so we can be sure that the current WAL is present
MG_ASSERT(current_wal_seq_num);
recovery_steps.emplace_back(RecoveryCurrentWal{*current_wal_seq_num});
return recovery_steps;
}
// Find the longest chain of WALs for recovery.
// The chain consists ONLY of sequential WALs.
auto rwal_it = wal_files->rbegin();
// if the last finalized WAL is before the replica commit
// then we can recovery only from current WAL
if (rwal_it->to_timestamp <= replica_commit) {
MG_ASSERT(current_wal_seq_num);
recovery_steps.emplace_back(RecoveryCurrentWal{*current_wal_seq_num});
return recovery_steps;
}
uint64_t previous_seq_num{rwal_it->seq_num};
for (; rwal_it != wal_files->rend(); ++rwal_it) {
// If the difference between two consecutive wal files is not 0 or 1
// we have a missing WAL in our chain
if (previous_seq_num - rwal_it->seq_num > 1) {
break;
}
// Find first WAL that contains up to replica commit, i.e. WAL
// that is before the replica commit or conatins the replica commit
// as the last committed transaction OR we managed to find the first WAL
// file.
if (replica_commit >= rwal_it->from_timestamp || rwal_it->seq_num == 0) {
if (replica_commit >= rwal_it->to_timestamp) {
// We want the WAL after because the replica already contains all the
// commits from this WAL
--rwal_it;
}
std::vector<std::filesystem::path> wal_chain;
auto distance_from_first = std::distance(rwal_it, wal_files->rend() - 1);
// We have managed to create WAL chain
// We need to lock these files and add them to the chain
for (auto result_wal_it = wal_files->begin() + distance_from_first; result_wal_it != wal_files->end();
++result_wal_it) {
locker_acc.AddPath(result_wal_it->path);
wal_chain.push_back(std::move(result_wal_it->path));
}
recovery_steps.emplace_back(std::in_place_type_t<RecoveryWals>{}, std::move(wal_chain));
if (current_wal_seq_num) {
recovery_steps.emplace_back(RecoveryCurrentWal{*current_wal_seq_num});
}
return recovery_steps;
}
previous_seq_num = rwal_it->seq_num;
}
MG_ASSERT(latest_snapshot, "Invalid durability state, missing snapshot");
// We didn't manage to find a WAL chain, we need to send the latest snapshot
// with its WALs
locker_acc.AddPath(latest_snapshot->path);
recovery_steps.emplace_back(std::in_place_type_t<RecoverySnapshot>{}, std::move(latest_snapshot->path));
std::vector<std::filesystem::path> recovery_wal_files;
auto wal_it = wal_files->begin();
for (; wal_it != wal_files->end(); ++wal_it) {
// Assuming recovery process is correct the snashpot should
// always retain a single WAL that contains a transaction
// before its creation
if (latest_snapshot->start_timestamp < wal_it->to_timestamp) {
if (latest_snapshot->start_timestamp < wal_it->from_timestamp) {
MG_ASSERT(wal_it != wal_files->begin(), "Invalid durability files state");
--wal_it;
}
break;
}
}
for (; wal_it != wal_files->end(); ++wal_it) {
locker_acc.AddPath(wal_it->path);
recovery_wal_files.push_back(std::move(wal_it->path));
}
// We only have a WAL before the snapshot
if (recovery_wal_files.empty()) {
locker_acc.AddPath(wal_files->back().path);
recovery_wal_files.push_back(std::move(wal_files->back().path));
}
recovery_steps.emplace_back(std::in_place_type_t<RecoveryWals>{}, std::move(recovery_wal_files));
if (current_wal_seq_num) {
recovery_steps.emplace_back(RecoveryCurrentWal{*current_wal_seq_num});
}
return recovery_steps;
}
////// TimeoutDispatcher //////
void Shard::ReplicationClient::TimeoutDispatcher::WaitForTaskToFinish() {
// Wait for the previous timeout task to finish
std::unique_lock main_guard(main_lock);
main_cv.wait(main_guard, [&] { return finished; });
}
void Shard::ReplicationClient::TimeoutDispatcher::StartTimeoutTask(const double timeout) {
timeout_pool.AddTask([timeout, this] {
finished = false;
using std::chrono::steady_clock;
const auto timeout_duration =
std::chrono::duration_cast<steady_clock::duration>(std::chrono::duration<double>(timeout));
const auto end_time = steady_clock::now() + timeout_duration;
while (active && (steady_clock::now() < end_time)) {
std::this_thread::sleep_for(std::chrono::milliseconds(10));
}
std::unique_lock main_guard(main_lock);
finished = true;
active = false;
main_cv.notify_one();
});
}
////// ReplicaStream //////
Shard::ReplicationClient::ReplicaStream::ReplicaStream(ReplicationClient *self,
const uint64_t previous_commit_timestamp,
const uint64_t current_seq_num)
: self_(self),
stream_(self_->rpc_client_->Stream<replication::AppendDeltasRpc>(previous_commit_timestamp, current_seq_num)) {
replication::Encoder encoder{stream_.GetBuilder()};
encoder.WriteString(self_->shard_->epoch_id_);
}
void Shard::ReplicationClient::ReplicaStream::AppendDelta(const Delta &delta, const Vertex &vertex,
uint64_t final_commit_timestamp) {
replication::Encoder encoder(stream_.GetBuilder());
EncodeDelta(&encoder, &self_->shard_->name_id_mapper_, self_->shard_->config_.items, delta, vertex,
final_commit_timestamp);
}
void Shard::ReplicationClient::ReplicaStream::AppendDelta(const Delta &delta, const Edge &edge,
uint64_t final_commit_timestamp) {
replication::Encoder encoder(stream_.GetBuilder());
EncodeDelta(&encoder, &self_->shard_->name_id_mapper_, delta, edge, final_commit_timestamp);
}
void Shard::ReplicationClient::ReplicaStream::AppendTransactionEnd(uint64_t final_commit_timestamp) {
replication::Encoder encoder(stream_.GetBuilder());
EncodeTransactionEnd(&encoder, final_commit_timestamp);
}
void Shard::ReplicationClient::ReplicaStream::AppendOperation(durability::StorageGlobalOperation operation,
LabelId label, const std::set<PropertyId> &properties,
uint64_t timestamp) {
replication::Encoder encoder(stream_.GetBuilder());
EncodeOperation(&encoder, &self_->shard_->name_id_mapper_, operation, label, properties, timestamp);
}
replication::AppendDeltasRes Shard::ReplicationClient::ReplicaStream::Finalize() { return stream_.AwaitResponse(); }
////// CurrentWalHandler //////
Shard::ReplicationClient::CurrentWalHandler::CurrentWalHandler(ReplicationClient *self)
: self_(self), stream_(self_->rpc_client_->Stream<replication::CurrentWalRpc>()) {}
void Shard::ReplicationClient::CurrentWalHandler::AppendFilename(const std::string &filename) {
replication::Encoder encoder(stream_.GetBuilder());
encoder.WriteString(filename);
}
void Shard::ReplicationClient::CurrentWalHandler::AppendSize(const size_t size) {
replication::Encoder encoder(stream_.GetBuilder());
encoder.WriteUint(size);
}
void Shard::ReplicationClient::CurrentWalHandler::AppendFileData(utils::InputFile *file) {
replication::Encoder encoder(stream_.GetBuilder());
encoder.WriteFileData(file);
}
void Shard::ReplicationClient::CurrentWalHandler::AppendBufferData(const uint8_t *buffer, const size_t buffer_size) {
replication::Encoder encoder(stream_.GetBuilder());
encoder.WriteBuffer(buffer, buffer_size);
}
replication::CurrentWalRes Shard::ReplicationClient::CurrentWalHandler::Finalize() { return stream_.AwaitResponse(); }
} // namespace memgraph::storage::v3

View File

@ -1,203 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <atomic>
#include <chrono>
#include <thread>
#include <variant>
#include "rpc/client.hpp"
#include "storage/v3/config.hpp"
#include "storage/v3/delta.hpp"
#include "storage/v3/durability/wal.hpp"
#include "storage/v3/id_types.hpp"
#include "storage/v3/mvcc.hpp"
#include "storage/v3/name_id_mapper.hpp"
#include "storage/v3/property_value.hpp"
#include "storage/v3/replication/config.hpp"
#include "storage/v3/replication/enums.hpp"
#include "storage/v3/replication/rpc.hpp"
#include "storage/v3/replication/serialization.hpp"
#include "storage/v3/shard.hpp"
#include "utils/file.hpp"
#include "utils/file_locker.hpp"
#include "utils/spin_lock.hpp"
#include "utils/synchronized.hpp"
#include "utils/thread_pool.hpp"
namespace memgraph::storage::v3 {
class Shard::ReplicationClient {
public:
ReplicationClient(std::string name, Shard *shard, const io::network::Endpoint &endpoint,
replication::ReplicationMode mode, const replication::ReplicationClientConfig &config = {});
// Handler used for transfering the current transaction.
class ReplicaStream {
private:
friend class ReplicationClient;
explicit ReplicaStream(ReplicationClient *self, uint64_t previous_commit_timestamp, uint64_t current_seq_num);
public:
/// @throw rpc::RpcFailedException
void AppendDelta(const Delta &delta, const Vertex &vertex, uint64_t final_commit_timestamp);
/// @throw rpc::RpcFailedException
void AppendDelta(const Delta &delta, const Edge &edge, uint64_t final_commit_timestamp);
/// @throw rpc::RpcFailedException
void AppendTransactionEnd(uint64_t final_commit_timestamp);
/// @throw rpc::RpcFailedException
void AppendOperation(durability::StorageGlobalOperation operation, LabelId label,
const std::set<PropertyId> &properties, uint64_t timestamp);
private:
/// @throw rpc::RpcFailedException
replication::AppendDeltasRes Finalize();
ReplicationClient *self_;
rpc::Client::StreamHandler<replication::AppendDeltasRpc> stream_;
};
// Handler for transfering the current WAL file whose data is
// contained in the internal buffer and the file.
class CurrentWalHandler {
private:
friend class ReplicationClient;
explicit CurrentWalHandler(ReplicationClient *self);
public:
void AppendFilename(const std::string &filename);
void AppendSize(size_t size);
void AppendFileData(utils::InputFile *file);
void AppendBufferData(const uint8_t *buffer, size_t buffer_size);
/// @throw rpc::RpcFailedException
replication::CurrentWalRes Finalize();
private:
ReplicationClient *self_;
rpc::Client::StreamHandler<replication::CurrentWalRpc> stream_;
};
void StartTransactionReplication(uint64_t current_wal_seq_num);
// Replication clients can be removed at any point
// so to avoid any complexity of checking if the client was removed whenever
// we want to send part of transaction and to avoid adding some GC logic this
// function will run a callback if, after previously callling
// StartTransactionReplication, stream is created.
void IfStreamingTransaction(const std::function<void(ReplicaStream &handler)> &callback);
void FinalizeTransactionReplication();
// Transfer the snapshot file.
// @param path Path of the snapshot file.
replication::SnapshotRes TransferSnapshot(const std::filesystem::path &path);
CurrentWalHandler TransferCurrentWalFile() { return CurrentWalHandler{this}; }
// Transfer the WAL files
replication::WalFilesRes TransferWalFiles(const std::vector<std::filesystem::path> &wal_files);
const auto &Name() const { return name_; }
auto State() const { return replica_state_.load(); }
auto Mode() const { return mode_; }
auto Timeout() const { return timeout_; }
const auto &Endpoint() const { return rpc_client_->Endpoint(); }
private:
void FinalizeTransactionReplicationInternal();
void RecoverReplica(uint64_t replica_commit);
uint64_t ReplicateCurrentWal();
using RecoveryWals = std::vector<std::filesystem::path>;
struct RecoveryCurrentWal {
uint64_t current_wal_seq_num;
explicit RecoveryCurrentWal(const uint64_t current_wal_seq_num) : current_wal_seq_num(current_wal_seq_num) {}
};
using RecoverySnapshot = std::filesystem::path;
using RecoveryStep = std::variant<RecoverySnapshot, RecoveryWals, RecoveryCurrentWal>;
std::vector<RecoveryStep> GetRecoverySteps(uint64_t replica_commit, utils::FileRetainer::FileLocker *file_locker);
void FrequentCheck();
void InitializeClient();
void TryInitializeClientSync();
void TryInitializeClientAsync();
void HandleRpcFailure();
std::string name_;
Shard *shard_;
std::optional<communication::ClientContext> rpc_context_;
std::optional<rpc::Client> rpc_client_;
std::optional<ReplicaStream> replica_stream_;
replication::ReplicationMode mode_{replication::ReplicationMode::SYNC};
// Dispatcher class for timeout tasks
struct TimeoutDispatcher {
explicit TimeoutDispatcher(){};
void WaitForTaskToFinish();
void StartTimeoutTask(double timeout);
// If the Timeout task should continue waiting
std::atomic<bool> active{false};
std::mutex main_lock;
std::condition_variable main_cv;
private:
// if the Timeout task finished executing
bool finished{true};
utils::ThreadPool timeout_pool{1};
};
std::optional<double> timeout_;
std::optional<TimeoutDispatcher> timeout_dispatcher_;
utils::SpinLock client_lock_;
// This thread pool is used for background tasks so we don't
// block the main storage thread
// We use only 1 thread for 2 reasons:
// - background tasks ALWAYS contain some kind of RPC communication.
// We can't have multiple RPC communication from a same client
// because that's not logically valid (e.g. you cannot send a snapshot
// and WAL at a same time because WAL will arrive earlier and be applied
// before the snapshot which is not correct)
// - the implementation is simplified as we have a total control of what
// this pool is executing. Also, we can simply queue multiple tasks
// and be sure of the execution order.
// Not having mulitple possible threads in the same client allows us
// to ignore concurrency problems inside the client.
utils::ThreadPool thread_pool_{1};
std::atomic<replication::ReplicaState> replica_state_{replication::ReplicaState::INVALID};
utils::Scheduler replica_checker_;
};
} // namespace memgraph::storage::v3

View File

@ -1,570 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include "storage/v3/replication/replication_server.hpp"
#include <atomic>
#include <filesystem>
#include "storage/v3/durability/durability.hpp"
#include "storage/v3/durability/paths.hpp"
#include "storage/v3/durability/serialization.hpp"
#include "storage/v3/durability/snapshot.hpp"
#include "storage/v3/durability/version.hpp"
#include "storage/v3/durability/wal.hpp"
#include "storage/v3/replication/config.hpp"
#include "storage/v3/transaction.hpp"
#include "utils/exceptions.hpp"
namespace memgraph::storage::v3 {
namespace {
std::pair<uint64_t, durability::WalDeltaData> ReadDelta(durability::BaseDecoder *decoder) {
try {
auto timestamp = ReadWalDeltaHeader(decoder);
SPDLOG_INFO(" Timestamp {}", timestamp);
auto delta = ReadWalDeltaData(decoder);
return {timestamp, delta};
} catch (const slk::SlkReaderException &) {
throw utils::BasicException("Missing data!");
} catch (const durability::RecoveryFailure &) {
throw utils::BasicException("Invalid data!");
}
};
} // namespace
Shard::ReplicationServer::ReplicationServer(Shard *shard, io::network::Endpoint endpoint,
const replication::ReplicationServerConfig &config)
: shard_(shard) {
// Create RPC server.
if (config.ssl) {
rpc_server_context_.emplace(config.ssl->key_file, config.ssl->cert_file, config.ssl->ca_file,
config.ssl->verify_peer);
} else {
rpc_server_context_.emplace();
}
// NOTE: The replication server must have a single thread for processing
// because there is no need for more processing threads - each replica can
// have only a single main server. Also, the single-threaded guarantee
// simplifies the rest of the implementation.
rpc_server_.emplace(std::move(endpoint), &*rpc_server_context_,
/* workers_count = */ 1);
rpc_server_->Register<replication::HeartbeatRpc>([this](auto *req_reader, auto *res_builder) {
spdlog::debug("Received HeartbeatRpc");
this->HeartbeatHandler(req_reader, res_builder);
});
rpc_server_->Register<replication::FrequentHeartbeatRpc>([](auto *req_reader, auto *res_builder) {
spdlog::debug("Received FrequentHeartbeatRpc");
FrequentHeartbeatHandler(req_reader, res_builder);
});
rpc_server_->Register<replication::AppendDeltasRpc>([this](auto *req_reader, auto *res_builder) {
spdlog::debug("Received AppendDeltasRpc");
this->AppendDeltasHandler(req_reader, res_builder);
});
rpc_server_->Register<replication::SnapshotRpc>([this](auto *req_reader, auto *res_builder) {
spdlog::debug("Received SnapshotRpc");
this->SnapshotHandler(req_reader, res_builder);
});
rpc_server_->Register<replication::WalFilesRpc>([this](auto *req_reader, auto *res_builder) {
spdlog::debug("Received WalFilesRpc");
this->WalFilesHandler(req_reader, res_builder);
});
rpc_server_->Register<replication::CurrentWalRpc>([this](auto *req_reader, auto *res_builder) {
spdlog::debug("Received CurrentWalRpc");
this->CurrentWalHandler(req_reader, res_builder);
});
rpc_server_->Start();
}
void Shard::ReplicationServer::HeartbeatHandler(slk::Reader *req_reader, slk::Builder *res_builder) {
replication::HeartbeatReq req;
slk::Load(&req, req_reader);
replication::HeartbeatRes res{true, shard_->last_commit_timestamp_, shard_->epoch_id_};
slk::Save(res, res_builder);
}
void Shard::ReplicationServer::FrequentHeartbeatHandler(slk::Reader *req_reader, slk::Builder *res_builder) {
replication::FrequentHeartbeatReq req;
slk::Load(&req, req_reader);
replication::FrequentHeartbeatRes res{true};
slk::Save(res, res_builder);
}
void Shard::ReplicationServer::AppendDeltasHandler(slk::Reader *req_reader, slk::Builder *res_builder) {
replication::AppendDeltasReq req;
slk::Load(&req, req_reader);
replication::Decoder decoder(req_reader);
auto maybe_epoch_id = decoder.ReadString();
MG_ASSERT(maybe_epoch_id, "Invalid replication message");
if (*maybe_epoch_id != shard_->epoch_id_) {
shard_->epoch_history_.emplace_back(std::move(shard_->epoch_id_), shard_->last_commit_timestamp_);
shard_->epoch_id_ = std::move(*maybe_epoch_id);
}
if (shard_->wal_file_) {
if (req.seq_num > shard_->wal_file_->SequenceNumber() || *maybe_epoch_id != shard_->epoch_id_) {
shard_->wal_file_->FinalizeWal();
shard_->wal_file_.reset();
shard_->wal_seq_num_ = req.seq_num;
} else {
MG_ASSERT(shard_->wal_file_->SequenceNumber() == req.seq_num, "Invalid sequence number of current wal file");
shard_->wal_seq_num_ = req.seq_num + 1;
}
} else {
shard_->wal_seq_num_ = req.seq_num;
}
if (req.previous_commit_timestamp != shard_->last_commit_timestamp_) {
// Empty the stream
bool transaction_complete = false;
while (!transaction_complete) {
SPDLOG_INFO("Skipping delta");
const auto [timestamp, delta] = ReadDelta(&decoder);
transaction_complete = durability::IsWalDeltaDataTypeTransactionEnd(delta.type);
}
replication::AppendDeltasRes res{false, shard_->last_commit_timestamp_};
slk::Save(res, res_builder);
return;
}
ReadAndApplyDelta(&decoder);
replication::AppendDeltasRes res{true, shard_->last_commit_timestamp_};
slk::Save(res, res_builder);
}
void Shard::ReplicationServer::SnapshotHandler(slk::Reader *req_reader, slk::Builder *res_builder) {
replication::SnapshotReq req;
slk::Load(&req, req_reader);
replication::Decoder decoder(req_reader);
utils::EnsureDirOrDie(shard_->snapshot_directory_);
const auto maybe_snapshot_path = decoder.ReadFile(shard_->snapshot_directory_);
MG_ASSERT(maybe_snapshot_path, "Failed to load snapshot!");
spdlog::info("Received snapshot saved to {}", *maybe_snapshot_path);
// Clear the database
shard_->vertices_.clear();
shard_->edges_.clear();
shard_->constraints_ = Constraints();
shard_->indices_.label_index =
LabelIndex(&shard_->indices_, &shard_->constraints_, shard_->config_.items, shard_->vertex_validator_);
shard_->indices_.label_property_index =
LabelPropertyIndex(&shard_->indices_, &shard_->constraints_, shard_->config_.items, shard_->vertex_validator_);
try {
spdlog::debug("Loading snapshot");
auto recovered_snapshot = durability::RecoveredSnapshot{};
// durability::LoadSnapshot(*maybe_snapshot_path, &shard_->vertices_, &shard_->edges_,
// &shard_->epoch_history_,
// &shard_->name_id_mapper_, &shard_->edge_count_, shard_->config_.items);
spdlog::debug("Snapshot loaded successfully");
// If this step is present it should always be the first step of
// the recovery so we use the UUID we read from snasphost
shard_->uuid_ = std::move(recovered_snapshot.snapshot_info.uuid);
shard_->epoch_id_ = std::move(recovered_snapshot.snapshot_info.epoch_id);
const auto &recovery_info = recovered_snapshot.recovery_info;
shard_->timestamp_ = std::max(shard_->timestamp_, recovery_info.next_timestamp);
// durability::RecoverIndicesAndConstraints(recovered_snapshot.indices_constraints, &shard_->indices_,
// &shard_->constraints_, &shard_->vertices_);
} catch (const durability::RecoveryFailure &e) {
LOG_FATAL("Couldn't load the snapshot because of: {}", e.what());
}
replication::SnapshotRes res{true, shard_->last_commit_timestamp_};
slk::Save(res, res_builder);
// Delete other durability files
auto snapshot_files = durability::GetSnapshotFiles(shard_->snapshot_directory_, shard_->uuid_);
for (const auto &[path, uuid, _] : snapshot_files) {
if (path != *maybe_snapshot_path) {
shard_->file_retainer_.DeleteFile(path);
}
}
auto wal_files = durability::GetWalFiles(shard_->wal_directory_, shard_->uuid_);
if (wal_files) {
for (const auto &wal_file : *wal_files) {
shard_->file_retainer_.DeleteFile(wal_file.path);
}
shard_->wal_file_.reset();
}
}
void Shard::ReplicationServer::WalFilesHandler(slk::Reader *req_reader, slk::Builder *res_builder) {
replication::WalFilesReq req;
slk::Load(&req, req_reader);
const auto wal_file_number = req.file_number;
spdlog::debug("Received WAL files: {}", wal_file_number);
replication::Decoder decoder(req_reader);
utils::EnsureDirOrDie(shard_->wal_directory_);
for (auto i = 0; i < wal_file_number; ++i) {
LoadWal(&decoder);
}
replication::WalFilesRes res{true, shard_->last_commit_timestamp_};
slk::Save(res, res_builder);
}
void Shard::ReplicationServer::CurrentWalHandler(slk::Reader *req_reader, slk::Builder *res_builder) {
replication::CurrentWalReq req;
slk::Load(&req, req_reader);
replication::Decoder decoder(req_reader);
utils::EnsureDirOrDie(shard_->wal_directory_);
LoadWal(&decoder);
replication::CurrentWalRes res{true, shard_->last_commit_timestamp_};
slk::Save(res, res_builder);
}
void Shard::ReplicationServer::LoadWal(replication::Decoder *decoder) {
const auto temp_wal_directory = std::filesystem::temp_directory_path() / "memgraph" / durability::kWalDirectory;
utils::EnsureDir(temp_wal_directory);
auto maybe_wal_path = decoder->ReadFile(temp_wal_directory);
MG_ASSERT(maybe_wal_path, "Failed to load WAL!");
spdlog::trace("Received WAL saved to {}", *maybe_wal_path);
try {
auto wal_info = durability::ReadWalInfo(*maybe_wal_path);
if (wal_info.seq_num == 0) {
shard_->uuid_ = wal_info.uuid;
}
if (wal_info.epoch_id != shard_->epoch_id_) {
shard_->epoch_history_.emplace_back(wal_info.epoch_id, shard_->last_commit_timestamp_);
shard_->epoch_id_ = std::move(wal_info.epoch_id);
}
if (shard_->wal_file_) {
if (shard_->wal_file_->SequenceNumber() != wal_info.seq_num) {
shard_->wal_file_->FinalizeWal();
shard_->wal_seq_num_ = wal_info.seq_num;
shard_->wal_file_.reset();
}
} else {
shard_->wal_seq_num_ = wal_info.seq_num;
}
durability::Decoder wal;
const auto version = wal.Initialize(*maybe_wal_path, durability::kWalMagic);
if (!version) throw durability::RecoveryFailure("Couldn't read WAL magic and/or version!");
if (!durability::IsVersionSupported(*version)) throw durability::RecoveryFailure("Invalid WAL version!");
wal.SetPosition(wal_info.offset_deltas);
for (size_t i = 0; i < wal_info.num_deltas;) {
i += ReadAndApplyDelta(&wal);
}
spdlog::debug("{} loaded successfully", *maybe_wal_path);
} catch (const durability::RecoveryFailure &e) {
LOG_FATAL("Couldn't recover WAL deltas from {} because of: {}", *maybe_wal_path, e.what());
}
}
Shard::ReplicationServer::~ReplicationServer() {
if (rpc_server_) {
rpc_server_->Shutdown();
rpc_server_->AwaitShutdown();
}
}
uint64_t Shard::ReplicationServer::ReadAndApplyDelta(durability::BaseDecoder *decoder) {
auto edge_acc = shard_->edges_.access();
// auto vertex_acc = shard_->vertices_.access();
std::optional<std::pair<uint64_t, Shard::Accessor>> commit_timestamp_and_accessor;
// auto get_transaction = [this, &commit_timestamp_and_accessor](uint64_t commit_timestamp) {
// if (!commit_timestamp_and_accessor) {
// commit_timestamp_and_accessor.emplace(commit_timestamp, shard_->Access());
// } else if (commit_timestamp_and_accessor->first != commit_timestamp) {
// throw utils::BasicException("Received more than one transaction!");
// }
// return &commit_timestamp_and_accessor->second;
// };
uint64_t applied_deltas = 0;
auto max_commit_timestamp = shard_->last_commit_timestamp_;
for (bool transaction_complete = false; !transaction_complete; ++applied_deltas) {
const auto [timestamp, delta] = ReadDelta(decoder);
if (timestamp > max_commit_timestamp) {
max_commit_timestamp = timestamp;
}
transaction_complete = durability::IsWalDeltaDataTypeTransactionEnd(delta.type);
if (timestamp < shard_->timestamp_) {
continue;
}
// SPDLOG_INFO(" Delta {}", applied_deltas);
// switch (delta.type) {
// case durability::WalDeltaData::Type::VERTEX_CREATE: {
// spdlog::trace(" Create vertex {}", delta.vertex_create_delete.gid.AsUint());
// auto *transaction = get_transaction(timestamp);
// transaction->CreateVertex(delta.vertex_create_delete.gid);
// break;
// }
// case durability::WalDeltaData::Type::VERTEX_DELETE: {
// spdlog::trace(" Delete vertex {}", delta.vertex_create_delete.gid.AsUint());
// auto *transaction = get_transaction(timestamp);
// auto vertex = transaction->FindVertex(delta.vertex_create_delete.gid, View::NEW);
// if (!vertex) throw utils::BasicException("Invalid transaction!");
// auto ret = transaction->DeleteVertex(&*vertex);
// if (ret.HasError() || !ret.GetValue()) throw utils::BasicException("Invalid transaction!");
// break;
// }
// case durability::WalDeltaData::Type::VERTEX_ADD_LABEL: {
// spdlog::trace(" Vertex {} add label {}", delta.vertex_add_remove_label.gid.AsUint(),
// delta.vertex_add_remove_label.label);
// auto *transaction = get_transaction(timestamp);
// auto vertex = transaction->FindVertex(delta.vertex_add_remove_label.gid, View::NEW);
// if (!vertex) throw utils::BasicException("Invalid transaction!");
// auto ret = vertex->AddLabel(transaction->NameToLabel(delta.vertex_add_remove_label.label));
// if (ret.HasError() || !ret.GetValue()) throw utils::BasicException("Invalid transaction!");
// break;
// }
// case durability::WalDeltaData::Type::VERTEX_REMOVE_LABEL: {
// spdlog::trace(" Vertex {} remove label {}", delta.vertex_add_remove_label.gid.AsUint(),
// delta.vertex_add_remove_label.label);
// auto *transaction = get_transaction(timestamp);
// auto vertex = transaction->FindVertex(delta.vertex_add_remove_label.gid, View::NEW);
// if (!vertex) throw utils::BasicException("Invalid transaction!");
// auto ret = vertex->RemoveLabel(transaction->NameToLabel(delta.vertex_add_remove_label.label));
// if (ret.HasError() || !ret.GetValue()) throw utils::BasicException("Invalid transaction!");
// break;
// }
// case durability::WalDeltaData::Type::VERTEX_SET_PROPERTY: {
// spdlog::trace(" Vertex {} set property {} to {}", delta.vertex_edge_set_property.gid.AsUint(),
// delta.vertex_edge_set_property.property, delta.vertex_edge_set_property.value);
// auto *transaction = get_transaction(timestamp);
// auto vertex = transaction->FindVertex(delta.vertex_edge_set_property.gid, View::NEW);
// if (!vertex) throw utils::BasicException("Invalid transaction!");
// auto ret = vertex->SetProperty(transaction->NameToProperty(delta.vertex_edge_set_property.property),
// delta.vertex_edge_set_property.value);
// if (ret.HasError()) throw utils::BasicException("Invalid transaction!");
// break;
// }
// case durability::WalDeltaData::Type::EDGE_CREATE: {
// spdlog::trace(" Create edge {} of type {} from vertex {} to vertex {}",
// delta.edge_create_delete.gid.AsUint(), delta.edge_create_delete.edge_type,
// delta.edge_create_delete.from_vertex.AsUint(), delta.edge_create_delete.to_vertex.AsUint());
// auto *transaction = get_transaction(timestamp);
// auto from_vertex = transaction->FindVertex(delta.edge_create_delete.from_vertex, View::NEW);
// if (!from_vertex) throw utils::BasicException("Invalid transaction!");
// auto to_vertex = transaction->FindVertex(delta.edge_create_delete.to_vertex, View::NEW);
// if (!to_vertex) throw utils::BasicException("Invalid transaction!");
// auto edge = transaction->CreateEdge(&*from_vertex, &*to_vertex,
// transaction->NameToEdgeType(delta.edge_create_delete.edge_type),
// delta.edge_create_delete.gid);
// if (edge.HasError()) throw utils::BasicException("Invalid transaction!");
// break;
// }
// case durability::WalDeltaData::Type::EDGE_DELETE: {
// spdlog::trace(" Delete edge {} of type {} from vertex {} to vertex {}",
// delta.edge_create_delete.gid.AsUint(), delta.edge_create_delete.edge_type,
// delta.edge_create_delete.from_vertex.AsUint(), delta.edge_create_delete.to_vertex.AsUint());
// auto *transaction = get_transaction(timestamp);
// auto from_vertex = transaction->FindVertex(delta.edge_create_delete.from_vertex, View::NEW);
// if (!from_vertex) throw utils::BasicException("Invalid transaction!");
// auto to_vertex = transaction->FindVertex(delta.edge_create_delete.to_vertex, View::NEW);
// if (!to_vertex) throw utils::BasicException("Invalid transaction!");
// auto edges = from_vertex->OutEdges(View::NEW,
// {transaction->NameToEdgeType(delta.edge_create_delete.edge_type)},
// &*to_vertex);
// if (edges.HasError()) throw utils::BasicException("Invalid transaction!");
// if (edges->size() != 1) throw utils::BasicException("Invalid transaction!");
// auto &edge = (*edges)[0];
// auto ret = transaction->DeleteEdge(&edge);
// if (ret.HasError()) throw utils::BasicException("Invalid transaction!");
// break;
// }
// case durability::WalDeltaData::Type::EDGE_SET_PROPERTY: {
// spdlog::trace(" Edge {} set property {} to {}", delta.vertex_edge_set_property.gid.AsUint(),
// delta.vertex_edge_set_property.property, delta.vertex_edge_set_property.value);
// if (!shard_->config_.items.properties_on_edges)
// throw utils::BasicException(
// "Can't set properties on edges because properties on edges "
// "are disabled!");
// // auto *transaction = get_transaction(timestamp);
// // The following block of code effectively implements `FindEdge` and
// // yields an accessor that is only valid for managing the edge's
// // properties.
// auto edge = edge_acc.find(delta.vertex_edge_set_property.gid);
// if (edge == edge_acc.end()) throw utils::BasicException("Invalid transaction!");
// // The edge visibility check must be done here manually because we
// // don't allow direct access to the edges through the public API.
// {
// auto is_visible = !edge->deleted;
// auto *delta = edge->delta;
// ApplyDeltasForRead(&transaction->transaction_, delta, View::NEW, [&is_visible](const Delta &delta) {
// switch (delta.action) {
// case Delta::Action::ADD_LABEL:
// case Delta::Action::REMOVE_LABEL:
// case Delta::Action::SET_PROPERTY:
// case Delta::Action::ADD_IN_EDGE:
// case Delta::Action::ADD_OUT_EDGE:
// case Delta::Action::REMOVE_IN_EDGE:
// case Delta::Action::REMOVE_OUT_EDGE:
// break;
// case Delta::Action::RECREATE_OBJECT: {
// is_visible = true;
// break;
// }
// case Delta::Action::DELETE_OBJECT: {
// is_visible = false;
// break;
// }
// }
// });
// if (!is_visible) throw utils::BasicException("Invalid transaction!");
// }
// EdgeRef edge_ref(&*edge);
// // Here we create an edge accessor that we will use to get the
// // properties of the edge. The accessor is created with an invalid
// // type and invalid from/to pointers because we don't know them
// // here, but that isn't an issue because we won't use that part of
// // the API here.
// auto ea = EdgeAccessor{edge_ref,
// EdgeTypeId::FromUint(0UL),
// nullptr,
// nullptr,
// &transaction->transaction_,
// &shard_->indices_,
// &shard_->constraints_,
// shard_->config_.items,
// shard_->schema_validator_};
// auto ret = ea.SetProperty(transaction->NameToProperty(delta.vertex_edge_set_property.property),
// delta.vertex_edge_set_property.value);
// if (ret.HasError()) throw utils::BasicException("Invalid transaction!");
// break;
// }
// case durability::WalDeltaData::Type::TRANSACTION_END: {
// spdlog::trace(" Transaction end");
// if (!commit_timestamp_and_accessor || commit_timestamp_and_accessor->first != timestamp)
// throw utils::BasicException("Invalid data!");
// auto ret = commit_timestamp_and_accessor->second.Commit(commit_timestamp_and_accessor->first);
// if (ret.HasError()) throw utils::BasicException("Invalid transaction!");
// commit_timestamp_and_accessor = std::nullopt;
// break;
// }
// case durability::WalDeltaData::Type::LABEL_INDEX_CREATE: {
// spdlog::trace(" Create label index on :{}", delta.operation_label.label);
// // Need to send the timestamp
// if (commit_timestamp_and_accessor) throw utils::BasicException("Invalid transaction!");
// if (!shard_->CreateIndex(shard_->NameToLabel(delta.operation_label.label), timestamp))
// throw utils::BasicException("Invalid transaction!");
// break;
// }
// case durability::WalDeltaData::Type::LABEL_INDEX_DROP: {
// spdlog::trace(" Drop label index on :{}", delta.operation_label.label);
// if (commit_timestamp_and_accessor) throw utils::BasicException("Invalid transaction!");
// if (!shard_->DropIndex(shard_->NameToLabel(delta.operation_label.label), timestamp))
// throw utils::BasicException("Invalid transaction!");
// break;
// }
// case durability::WalDeltaData::Type::LABEL_PROPERTY_INDEX_CREATE: {
// spdlog::trace(" Create label+property index on :{} ({})", delta.operation_label_property.label,
// delta.operation_label_property.property);
// if (commit_timestamp_and_accessor) throw utils::BasicException("Invalid transaction!");
// if (!shard_->CreateIndex(shard_->NameToLabel(delta.operation_label_property.label),
// shard_->NameToProperty(delta.operation_label_property.property), timestamp))
// throw utils::BasicException("Invalid transaction!");
// break;
// }
// case durability::WalDeltaData::Type::LABEL_PROPERTY_INDEX_DROP: {
// spdlog::trace(" Drop label+property index on :{} ({})", delta.operation_label_property.label,
// delta.operation_label_property.property);
// if (commit_timestamp_and_accessor) throw utils::BasicException("Invalid transaction!");
// if (!shard_->DropIndex(shard_->NameToLabel(delta.operation_label_property.label),
// shard_->NameToProperty(delta.operation_label_property.property), timestamp))
// throw utils::BasicException("Invalid transaction!");
// break;
// }
// case durability::WalDeltaData::Type::EXISTENCE_CONSTRAINT_CREATE: {
// spdlog::trace(" Create existence constraint on :{} ({})", delta.operation_label_property.label,
// delta.operation_label_property.property);
// if (commit_timestamp_and_accessor) throw utils::BasicException("Invalid transaction!");
// auto ret = shard_->CreateExistenceConstraint(
// shard_->NameToLabel(delta.operation_label_property.label),
// shard_->NameToProperty(delta.operation_label_property.property), timestamp);
// if (!ret.HasValue() || !ret.GetValue()) throw utils::BasicException("Invalid transaction!");
// break;
// }
// case durability::WalDeltaData::Type::EXISTENCE_CONSTRAINT_DROP: {
// spdlog::trace(" Drop existence constraint on :{} ({})", delta.operation_label_property.label,
// delta.operation_label_property.property);
// if (commit_timestamp_and_accessor) throw utils::BasicException("Invalid transaction!");
// if (!shard_->DropExistenceConstraint(shard_->NameToLabel(delta.operation_label_property.label),
// shard_->NameToProperty(delta.operation_label_property.property),
// timestamp))
// throw utils::BasicException("Invalid transaction!");
// break;
// }
// case durability::WalDeltaData::Type::UNIQUE_CONSTRAINT_CREATE: {
// std::stringstream ss;
// utils::PrintIterable(ss, delta.operation_label_properties.properties);
// spdlog::trace(" Create unique constraint on :{} ({})", delta.operation_label_properties.label,
// ss.str()); if (commit_timestamp_and_accessor) throw utils::BasicException("Invalid transaction!");
// std::set<PropertyId> properties;
// for (const auto &prop : delta.operation_label_properties.properties) {
// properties.emplace(shard_->NameToProperty(prop));
// }
// auto ret = shard_->CreateUniqueConstraint(shard_->NameToLabel(delta.operation_label_properties.label),
// properties, timestamp);
// if (!ret.HasValue() || ret.GetValue() != UniqueConstraints::CreationStatus::SUCCESS)
// throw utils::BasicException("Invalid transaction!");
// break;
// }
// case durability::WalDeltaData::Type::UNIQUE_CONSTRAINT_DROP: {
// std::stringstream ss;
// utils::PrintIterable(ss, delta.operation_label_properties.properties);
// spdlog::trace(" Drop unique constraint on :{} ({})", delta.operation_label_properties.label,
// ss.str()); if (commit_timestamp_and_accessor) throw utils::BasicException("Invalid transaction!");
// std::set<PropertyId> properties;
// for (const auto &prop : delta.operation_label_properties.properties) {
// properties.emplace(shard_->NameToProperty(prop));
// }
// auto ret = shard_->DropUniqueConstraint(shard_->NameToLabel(delta.operation_label_properties.label),
// properties, timestamp);
// if (ret != UniqueConstraints::DeletionStatus::SUCCESS) throw utils::BasicException("Invalid transaction!");
// break;
// }
// }
}
if (commit_timestamp_and_accessor) throw utils::BasicException("Invalid data!");
shard_->last_commit_timestamp_ = max_commit_timestamp;
return applied_deltas;
}
} // namespace memgraph::storage::v3

View File

@ -1,47 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include "storage/v3/shard.hpp"
namespace memgraph::storage::v3 {
class Shard::ReplicationServer {
public:
explicit ReplicationServer(Shard *shard, io::network::Endpoint endpoint,
const replication::ReplicationServerConfig &config);
ReplicationServer(const ReplicationServer &) = delete;
ReplicationServer(ReplicationServer &&) = delete;
ReplicationServer &operator=(const ReplicationServer &) = delete;
ReplicationServer &operator=(ReplicationServer &&) = delete;
~ReplicationServer();
private:
// RPC handlers
void HeartbeatHandler(slk::Reader *req_reader, slk::Builder *res_builder);
static void FrequentHeartbeatHandler(slk::Reader *req_reader, slk::Builder *res_builder);
void AppendDeltasHandler(slk::Reader *req_reader, slk::Builder *res_builder);
void SnapshotHandler(slk::Reader *req_reader, slk::Builder *res_builder);
void WalFilesHandler(slk::Reader *req_reader, slk::Builder *res_builder);
void CurrentWalHandler(slk::Reader *req_reader, slk::Builder *res_builder);
void LoadWal(replication::Decoder *decoder);
uint64_t ReadAndApplyDelta(durability::BaseDecoder *decoder);
std::optional<communication::ServerContext> rpc_server_context_;
std::optional<rpc::Server> rpc_server_;
Shard *shard_;
};
} // namespace memgraph::storage::v3

View File

@ -1,74 +0,0 @@
;; Copyright 2022 Memgraph Ltd.
;;
;; Use of this software is governed by the Business Source License
;; included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
;; License, and you may not use this file except in compliance with the Business Source License.
;;
;; As of the Change Date specified in that file, in accordance with
;; the Business Source License, use of this software will be governed
;; by the Apache License, Version 2.0, included in the file
;; licenses/APL.txt.
#>cpp
#pragma once
#include <cstdint>
#include <cstring>
#include <string>
#include "rpc/messages.hpp"
#include "slk/serialization.hpp"
#include "slk/streams.hpp"
cpp<#
(lcp:namespace memgraph)
(lcp:namespace storage)
(lcp:namespace v3)
(lcp:namespace replication)
(lcp:define-rpc append-deltas
;; The actual deltas are sent as additional data using the RPC client's
;; streaming API for additional data.
(:request
((previous-commit-timestamp :uint64_t)
(seq-num :uint64_t)))
(:response
((success :bool)
(current-commit-timestamp :uint64_t))))
(lcp:define-rpc heartbeat
(:request
((main-commit-timestamp :uint64_t)
(epoch-id "std::string")))
(:response
((success :bool)
(current-commit-timestamp :uint64_t)
(epoch-id "std::string"))))
;; FrequentHearthbeat is required because calling Heartbeat takes the storage lock.
;; Configured by `replication_replica_check_delay`.
(lcp:define-rpc frequent-heartbeat
(:request ())
(:response ((success :bool))))
(lcp:define-rpc snapshot
(:request ())
(:response
((success :bool)
(current-commit-timestamp :uint64_t))))
(lcp:define-rpc wal-files
(:request ((file-number :uint64_t)))
(:response
((success :bool)
(current-commit-timestamp :uint64_t))))
(lcp:define-rpc current-wal
(:request ())
(:response
((success :bool)
(current-commit-timestamp :uint64_t))))
(lcp:pop-namespace) ;; replication
(lcp:pop-namespace) ;; v3
(lcp:pop-namespace) ;; storage
(lcp:pop-namespace) ;; memgraph

View File

@ -1,149 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include "storage/v3/replication/serialization.hpp"
namespace memgraph::storage::v3::replication {
////// Encoder //////
void Encoder::WriteMarker(durability::Marker marker) { slk::Save(marker, builder_); }
void Encoder::WriteBool(bool value) {
WriteMarker(durability::Marker::TYPE_BOOL);
slk::Save(value, builder_);
}
void Encoder::WriteUint(uint64_t value) {
WriteMarker(durability::Marker::TYPE_INT);
slk::Save(value, builder_);
}
void Encoder::WriteDouble(double value) {
WriteMarker(durability::Marker::TYPE_DOUBLE);
slk::Save(value, builder_);
}
void Encoder::WriteString(const std::string_view &value) {
WriteMarker(durability::Marker::TYPE_STRING);
slk::Save(value, builder_);
}
void Encoder::WritePropertyValue(const PropertyValue &value) {
WriteMarker(durability::Marker::TYPE_PROPERTY_VALUE);
slk::Save(value, builder_);
}
void Encoder::WriteBuffer(const uint8_t *buffer, const size_t buffer_size) { builder_->Save(buffer, buffer_size); }
void Encoder::WriteFileData(utils::InputFile *file) {
auto file_size = file->GetSize();
uint8_t buffer[utils::kFileBufferSize];
while (file_size > 0) {
const auto chunk_size = std::min(file_size, utils::kFileBufferSize);
file->Read(buffer, chunk_size);
WriteBuffer(buffer, chunk_size);
file_size -= chunk_size;
}
}
void Encoder::WriteFile(const std::filesystem::path &path) {
utils::InputFile file;
MG_ASSERT(file.Open(path), "Failed to open file {}", path);
MG_ASSERT(path.has_filename(), "Path does not have a filename!");
const auto &filename = path.filename().generic_string();
WriteString(filename);
auto file_size = file.GetSize();
WriteUint(file_size);
WriteFileData(&file);
file.Close();
}
////// Decoder //////
std::optional<durability::Marker> Decoder::ReadMarker() {
durability::Marker marker{durability::Marker::TYPE_NULL};
slk::Load(&marker, reader_);
return marker;
}
std::optional<bool> Decoder::ReadBool() {
if (const auto marker = ReadMarker(); !marker || marker != durability::Marker::TYPE_BOOL) return std::nullopt;
bool value{false};
slk::Load(&value, reader_);
return value;
}
std::optional<uint64_t> Decoder::ReadUint() {
if (const auto marker = ReadMarker(); !marker || marker != durability::Marker::TYPE_INT) return std::nullopt;
uint64_t value{0};
slk::Load(&value, reader_);
return value;
}
std::optional<double> Decoder::ReadDouble() {
if (const auto marker = ReadMarker(); !marker || marker != durability::Marker::TYPE_DOUBLE) return std::nullopt;
double value{0.0};
slk::Load(&value, reader_);
return value;
}
std::optional<std::string> Decoder::ReadString() {
if (const auto marker = ReadMarker(); !marker || marker != durability::Marker::TYPE_STRING) return std::nullopt;
std::string value;
slk::Load(&value, reader_);
return std::move(value);
}
std::optional<PropertyValue> Decoder::ReadPropertyValue() {
if (const auto marker = ReadMarker(); !marker || marker != durability::Marker::TYPE_PROPERTY_VALUE)
return std::nullopt;
PropertyValue value;
slk::Load(&value, reader_);
return std::move(value);
}
bool Decoder::SkipString() {
if (const auto marker = ReadMarker(); !marker || marker != durability::Marker::TYPE_STRING) return false;
std::string value;
slk::Load(&value, reader_);
return true;
}
bool Decoder::SkipPropertyValue() {
if (const auto marker = ReadMarker(); !marker || marker != durability::Marker::TYPE_PROPERTY_VALUE) return false;
PropertyValue value;
slk::Load(&value, reader_);
return true;
}
std::optional<std::filesystem::path> Decoder::ReadFile(const std::filesystem::path &directory,
const std::string &suffix) {
MG_ASSERT(std::filesystem::exists(directory) && std::filesystem::is_directory(directory),
"Sent path for streamed files should be a valid directory!");
utils::OutputFile file;
const auto maybe_filename = ReadString();
MG_ASSERT(maybe_filename, "Filename missing for the file");
const auto filename = *maybe_filename + suffix;
auto path = directory / filename;
file.Open(path, utils::OutputFile::Mode::OVERWRITE_EXISTING);
std::optional<size_t> maybe_file_size = ReadUint();
MG_ASSERT(maybe_file_size, "File size missing");
auto file_size = *maybe_file_size;
uint8_t buffer[utils::kFileBufferSize];
while (file_size > 0) {
const auto chunk_size = std::min(file_size, utils::kFileBufferSize);
reader_->Load(buffer, chunk_size);
file.Write(buffer, chunk_size);
file_size -= chunk_size;
}
file.Close();
return std::move(path);
}
} // namespace memgraph::storage::v3::replication

View File

@ -1,80 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <filesystem>
#include "slk/streams.hpp"
#include "storage/v3/durability/serialization.hpp"
#include "storage/v3/replication/slk.hpp"
#include "utils/cast.hpp"
#include "utils/file.hpp"
namespace memgraph::storage::v3::replication {
class Encoder final : public durability::BaseEncoder {
public:
explicit Encoder(slk::Builder *builder) : builder_(builder) {}
void WriteMarker(durability::Marker marker) override;
void WriteBool(bool value) override;
void WriteUint(uint64_t value) override;
void WriteDouble(double value) override;
void WriteString(const std::string_view &value) override;
void WritePropertyValue(const PropertyValue &value) override;
void WriteBuffer(const uint8_t *buffer, size_t buffer_size);
void WriteFileData(utils::InputFile *file);
void WriteFile(const std::filesystem::path &path);
private:
slk::Builder *builder_;
};
class Decoder final : public durability::BaseDecoder {
public:
explicit Decoder(slk::Reader *reader) : reader_(reader) {}
std::optional<durability::Marker> ReadMarker() override;
std::optional<bool> ReadBool() override;
std::optional<uint64_t> ReadUint() override;
std::optional<double> ReadDouble() override;
std::optional<std::string> ReadString() override;
std::optional<PropertyValue> ReadPropertyValue() override;
bool SkipString() override;
bool SkipPropertyValue() override;
/// Read the file and save it inside the specified directory.
/// @param directory Directory which will contain the read file.
/// @param suffix Suffix to be added to the received file's filename.
/// @return If the read was successful, path to the read file.
std::optional<std::filesystem::path> ReadFile(const std::filesystem::path &directory, const std::string &suffix = "");
private:
slk::Reader *reader_;
};
} // namespace memgraph::storage::v3::replication

View File

@ -1,169 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include "storage/v3/replication/slk.hpp"
#include <type_traits>
#include "storage/v3/property_value.hpp"
#include "storage/v3/temporal.hpp"
#include "utils/cast.hpp"
namespace memgraph::slk {
void Save(const storage::v3::Gid &gid, slk::Builder *builder) { slk::Save(gid.AsUint(), builder); }
void Load(storage::v3::Gid *gid, slk::Reader *reader) {
uint64_t value{0};
slk::Load(&value, reader);
*gid = storage::v3::Gid::FromUint(value);
}
void Load(storage::v3::PropertyValue::Type *type, slk::Reader *reader) {
using PVTypeUnderlyingType = std::underlying_type_t<storage::v3::PropertyValue::Type>;
PVTypeUnderlyingType value{};
slk::Load(&value, reader);
bool valid{false};
switch (value) {
case utils::UnderlyingCast(storage::v3::PropertyValue::Type::Null):
case utils::UnderlyingCast(storage::v3::PropertyValue::Type::Bool):
case utils::UnderlyingCast(storage::v3::PropertyValue::Type::Int):
case utils::UnderlyingCast(storage::v3::PropertyValue::Type::Double):
case utils::UnderlyingCast(storage::v3::PropertyValue::Type::String):
case utils::UnderlyingCast(storage::v3::PropertyValue::Type::List):
case utils::UnderlyingCast(storage::v3::PropertyValue::Type::Map):
case utils::UnderlyingCast(storage::v3::PropertyValue::Type::TemporalData):
valid = true;
break;
default:
valid = false;
break;
}
if (!valid) throw slk::SlkDecodeException("Trying to load unknown storage::v3::PropertyValue!");
*type = static_cast<storage::v3::PropertyValue::Type>(value);
}
void Save(const storage::v3::PropertyValue &value, slk::Builder *builder) {
switch (value.type()) {
case storage::v3::PropertyValue::Type::Null:
slk::Save(storage::v3::PropertyValue::Type::Null, builder);
return;
case storage::v3::PropertyValue::Type::Bool:
slk::Save(storage::v3::PropertyValue::Type::Bool, builder);
slk::Save(value.ValueBool(), builder);
return;
case storage::v3::PropertyValue::Type::Int:
slk::Save(storage::v3::PropertyValue::Type::Int, builder);
slk::Save(value.ValueInt(), builder);
return;
case storage::v3::PropertyValue::Type::Double:
slk::Save(storage::v3::PropertyValue::Type::Double, builder);
slk::Save(value.ValueDouble(), builder);
return;
case storage::v3::PropertyValue::Type::String:
slk::Save(storage::v3::PropertyValue::Type::String, builder);
slk::Save(value.ValueString(), builder);
return;
case storage::v3::PropertyValue::Type::List: {
slk::Save(storage::v3::PropertyValue::Type::List, builder);
const auto &values = value.ValueList();
size_t size = values.size();
slk::Save(size, builder);
for (const auto &v : values) {
slk::Save(v, builder);
}
return;
}
case storage::v3::PropertyValue::Type::Map: {
slk::Save(storage::v3::PropertyValue::Type::Map, builder);
const auto &map = value.ValueMap();
size_t size = map.size();
slk::Save(size, builder);
for (const auto &kv : map) {
slk::Save(kv, builder);
}
return;
}
case storage::v3::PropertyValue::Type::TemporalData: {
slk::Save(storage::v3::PropertyValue::Type::TemporalData, builder);
const auto temporal_data = value.ValueTemporalData();
slk::Save(temporal_data.type, builder);
slk::Save(temporal_data.microseconds, builder);
return;
}
}
}
void Load(storage::v3::PropertyValue *value, slk::Reader *reader) {
storage::v3::PropertyValue::Type type{};
slk::Load(&type, reader);
switch (type) {
case storage::v3::PropertyValue::Type::Null:
*value = storage::v3::PropertyValue();
return;
case storage::v3::PropertyValue::Type::Bool: {
bool v{false};
slk::Load(&v, reader);
*value = storage::v3::PropertyValue(v);
return;
}
case storage::v3::PropertyValue::Type::Int: {
int64_t v{0};
slk::Load(&v, reader);
*value = storage::v3::PropertyValue(v);
return;
}
case storage::v3::PropertyValue::Type::Double: {
double v{0.0};
slk::Load(&v, reader);
*value = storage::v3::PropertyValue(v);
return;
}
case storage::v3::PropertyValue::Type::String: {
std::string v;
slk::Load(&v, reader);
*value = storage::v3::PropertyValue(std::move(v));
return;
}
case storage::v3::PropertyValue::Type::List: {
size_t size{0};
slk::Load(&size, reader);
std::vector<storage::v3::PropertyValue> list(size);
for (size_t i = 0; i < size; ++i) {
slk::Load(&list[i], reader);
}
*value = storage::v3::PropertyValue(std::move(list));
return;
}
case storage::v3::PropertyValue::Type::Map: {
size_t size{0};
slk::Load(&size, reader);
std::map<std::string, storage::v3::PropertyValue> map;
for (size_t i = 0; i < size; ++i) {
std::pair<std::string, storage::v3::PropertyValue> kv;
slk::Load(&kv, reader);
map.insert(kv);
}
*value = storage::v3::PropertyValue(std::move(map));
return;
}
case storage::v3::PropertyValue::Type::TemporalData: {
storage::v3::TemporalType temporal_type{};
slk::Load(&temporal_type, reader);
int64_t microseconds{0};
slk::Load(&microseconds, reader);
*value = storage::v3::PropertyValue(storage::v3::TemporalData{temporal_type, microseconds});
return;
}
}
}
} // namespace memgraph::slk

View File

@ -1,41 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include "slk/serialization.hpp"
#include "storage/v3/durability/marker.hpp"
#include "storage/v3/id_types.hpp"
#include "storage/v3/property_value.hpp"
#include "utils/concepts.hpp"
namespace memgraph::slk {
void Save(const storage::v3::Gid &gid, slk::Builder *builder);
void Load(storage::v3::Gid *gid, slk::Reader *reader);
void Save(const storage::v3::PropertyValue &value, slk::Builder *builder);
void Load(storage::v3::PropertyValue *value, slk::Reader *reader);
template <utils::Enum T>
void Save(const T &enum_value, slk::Builder *builder) {
slk::Save(utils::UnderlyingCast(enum_value), builder);
}
template <utils::Enum T>
void Load(T *enum_value, slk::Reader *reader) {
using UnderlyingType = std::underlying_type_t<T>;
UnderlyingType value;
slk::Load(&value, reader);
*enum_value = static_cast<T>(value);
}
} // namespace memgraph::slk

File diff suppressed because it is too large Load Diff

View File

@ -11,7 +11,6 @@
#pragma once
#include <atomic>
#include <cstdint>
#include <filesystem>
#include <map>
@ -21,13 +20,11 @@
#include <variant>
#include <vector>
#include "coordinator/hybrid_logical_clock.hpp"
#include "io/network/endpoint.hpp"
#include "io/time.hpp"
#include "kvstore/kvstore.hpp"
#include "storage/v3/commit_log.hpp"
#include "storage/v3/config.hpp"
#include "storage/v3/constraints.hpp"
#include "storage/v3/durability/metadata.hpp"
#include "storage/v3/durability/wal.hpp"
#include "storage/v3/edge.hpp"
#include "storage/v3/edge_accessor.hpp"
#include "storage/v3/id_types.hpp"
@ -46,6 +43,7 @@
#include "storage/v3/vertex_accessor.hpp"
#include "storage/v3/vertex_id.hpp"
#include "storage/v3/vertices_skip_list.hpp"
#include "storage/v3/view.hpp"
#include "utils/exceptions.hpp"
#include "utils/file_locker.hpp"
#include "utils/on_scope_exit.hpp"
@ -55,13 +53,6 @@
#include "utils/synchronized.hpp"
#include "utils/uuid.hpp"
/// REPLICATION ///
#include "rpc/server.hpp"
#include "storage/v3/replication/config.hpp"
#include "storage/v3/replication/enums.hpp"
#include "storage/v3/replication/rpc.hpp"
#include "storage/v3/replication/serialization.hpp"
namespace memgraph::storage::v3 {
// The storage is based on this paper:
@ -78,7 +69,6 @@ class AllVerticesIterable final {
Transaction *transaction_;
View view_;
Indices *indices_;
Constraints *constraints_;
Config::Items config_;
const VertexValidator *vertex_validator_;
const Schemas *schemas_;
@ -102,13 +92,11 @@ class AllVerticesIterable final {
};
AllVerticesIterable(VerticesSkipList::Accessor vertices_accessor, Transaction *transaction, View view,
Indices *indices, Constraints *constraints, Config::Items config,
const VertexValidator &vertex_validator)
Indices *indices, Config::Items config, const VertexValidator &vertex_validator)
: vertices_accessor_(std::move(vertices_accessor)),
transaction_(transaction),
view_(view),
indices_(indices),
constraints_(constraints),
config_(config),
vertex_validator_{&vertex_validator} {}
@ -184,13 +172,6 @@ struct IndicesInfo {
std::vector<std::pair<LabelId, PropertyId>> label_property;
};
/// Structure used to return information about existing constraints in the
/// storage.
struct ConstraintsInfo {
std::vector<std::pair<LabelId, PropertyId>> existence;
std::vector<std::pair<LabelId, std::set<PropertyId>>> unique;
};
/// Structure used to return information about existing schemas in the storage
struct SchemasInfo {
Schemas::SchemasList schemas;
@ -202,11 +183,8 @@ struct StorageInfo {
uint64_t edge_count;
double average_degree;
uint64_t memory_usage;
uint64_t disk_usage;
};
enum class ReplicationRole : uint8_t { MAIN, REPLICA };
class Shard final {
public:
/// @throw std::system_error
@ -224,19 +202,9 @@ class Shard final {
private:
friend class Shard;
explicit Accessor(Shard *shard, IsolationLevel isolation_level);
Accessor(Shard &shard, Transaction &transaction);
public:
Accessor(const Accessor &) = delete;
Accessor &operator=(const Accessor &) = delete;
Accessor &operator=(Accessor &&other) = delete;
// NOTE: After the accessor is moved, all objects derived from it (accessors
// and iterators) are *invalid*. You have to get all derived objects again.
Accessor(Accessor &&other) noexcept;
~Accessor();
// TODO(gvolfing) this is just a workaround for stitching remove this later.
LabelId GetPrimaryLabel() const noexcept { return shard_->primary_label_; }
@ -253,9 +221,8 @@ class Shard final {
std::optional<VertexAccessor> FindVertex(std::vector<PropertyValue> primary_key, View view);
VerticesIterable Vertices(View view) {
return VerticesIterable(AllVerticesIterable(shard_->vertices_.access(), &transaction_, view, &shard_->indices_,
&shard_->constraints_, shard_->config_.items,
shard_->vertex_validator_));
return VerticesIterable(AllVerticesIterable(shard_->vertices_.access(), transaction_, view, &shard_->indices_,
shard_->config_.items, shard_->vertex_validator_));
}
VerticesIterable Vertices(LabelId label, View view);
@ -338,41 +305,28 @@ class Shard final {
return {shard_->indices_.label_index.ListIndices(), shard_->indices_.label_property_index.ListIndices()};
}
ConstraintsInfo ListAllConstraints() const {
return {ListExistenceConstraints(shard_->constraints_),
shard_->constraints_.unique_constraints.ListConstraints()};
}
const SchemaValidator &GetSchemaValidator() const;
SchemasInfo ListAllSchemas() const { return {shard_->schemas_.ListSchemas()}; }
void AdvanceCommand();
/// Commit returns `ConstraintViolation` if the changes made by this
/// transaction violate an existence or unique constraint. In that case the
/// transaction is automatically aborted. Otherwise, void is returned.
/// @throw std::bad_alloc
utils::BasicResult<ConstraintViolation, void> Commit(std::optional<uint64_t> desired_commit_timestamp = {});
void Commit(coordinator::Hlc commit_timestamp);
/// @throw std::bad_alloc
void Abort();
void FinalizeTransaction();
private:
/// @throw std::bad_alloc
VertexAccessor CreateVertex(Gid gid, LabelId primary_label);
Shard *shard_;
Transaction transaction_;
std::optional<uint64_t> commit_timestamp_;
bool is_transaction_active_;
Transaction *transaction_;
Config::Items config_;
};
Accessor Access(std::optional<IsolationLevel> override_isolation_level = {}) {
return Accessor{this, override_isolation_level.value_or(isolation_level_)};
Accessor Access(coordinator::Hlc start_timestamp, std::optional<IsolationLevel> override_isolation_level = {}) {
return Accessor{*this, GetTransaction(start_timestamp, override_isolation_level.value_or(isolation_level_))};
}
LabelId NameToLabel(std::string_view name) const;
@ -399,45 +353,6 @@ class Shard final {
IndicesInfo ListAllIndices() const;
/// Creates an existence constraint. Returns true if the constraint was
/// successfully added, false if it already exists and a `ConstraintViolation`
/// if there is an existing vertex violating the constraint.
///
/// @throw std::bad_alloc
/// @throw std::length_error
utils::BasicResult<ConstraintViolation, bool> CreateExistenceConstraint(
LabelId label, PropertyId property, std::optional<uint64_t> desired_commit_timestamp = {});
/// Removes an existence constraint. Returns true if the constraint was
/// removed, and false if it doesn't exist.
bool DropExistenceConstraint(LabelId label, PropertyId property,
std::optional<uint64_t> desired_commit_timestamp = {});
/// Creates a unique constraint. In the case of two vertices violating the
/// constraint, it returns `ConstraintViolation`. Otherwise returns a
/// `UniqueConstraints::CreationStatus` enum with the following possibilities:
/// * `SUCCESS` if the constraint was successfully created,
/// * `ALREADY_EXISTS` if the constraint already existed,
/// * `EMPTY_PROPERTIES` if the property set is empty, or
// * `PROPERTIES_SIZE_LIMIT_EXCEEDED` if the property set exceeds the
// limit of maximum number of properties.
///
/// @throw std::bad_alloc
utils::BasicResult<ConstraintViolation, UniqueConstraints::CreationStatus> CreateUniqueConstraint(
LabelId label, const std::set<PropertyId> &properties, std::optional<uint64_t> desired_commit_timestamp = {});
/// Removes a unique constraint. Returns `UniqueConstraints::DeletionStatus`
/// enum with the following possibilities:
/// * `SUCCESS` if constraint was successfully removed,
/// * `NOT_FOUND` if the specified constraint was not found,
/// * `EMPTY_PROPERTIES` if the property set is empty, or
/// * `PROPERTIES_SIZE_LIMIT_EXCEEDED` if the property set exceeds the
// limit of maximum number of properties.
UniqueConstraints::DeletionStatus DropUniqueConstraint(LabelId label, const std::set<PropertyId> &properties,
std::optional<uint64_t> desired_commit_timestamp = {});
ConstraintsInfo ListAllConstraints() const;
SchemasInfo ListAllSchemas() const;
const Schemas::Schema *GetSchema(LabelId primary_label) const;
@ -448,75 +363,15 @@ class Shard final {
StorageInfo GetInfo() const;
bool LockPath();
bool UnlockPath();
bool SetReplicaRole(io::network::Endpoint endpoint, const replication::ReplicationServerConfig &config = {});
bool SetMainReplicationRole();
enum class RegisterReplicaError : uint8_t {
NAME_EXISTS,
END_POINT_EXISTS,
CONNECTION_FAILED,
COULD_NOT_BE_PERSISTED
};
/// @pre The instance should have a MAIN role
/// @pre Timeout can only be set for SYNC replication
utils::BasicResult<RegisterReplicaError, void> RegisterReplica(
std::string name, io::network::Endpoint endpoint, replication::ReplicationMode replication_mode,
const replication::ReplicationClientConfig &config = {});
/// @pre The instance should have a MAIN role
bool UnregisterReplica(std::string_view name);
std::optional<replication::ReplicaState> GetReplicaState(std::string_view name);
ReplicationRole GetReplicationRole() const;
struct ReplicaInfo {
std::string name;
replication::ReplicationMode mode;
std::optional<double> timeout;
io::network::Endpoint endpoint;
replication::ReplicaState state;
};
std::vector<ReplicaInfo> ReplicasInfo();
void FreeMemory();
void SetIsolationLevel(IsolationLevel isolation_level);
enum class CreateSnapshotError : uint8_t { DisabledForReplica };
utils::BasicResult<CreateSnapshotError> CreateSnapshot();
// Might invalidate accessors
void CollectGarbage(io::Time current_time);
void StoreMapping(std::unordered_map<uint64_t, std::string> id_to_name);
private:
Transaction CreateTransaction(IsolationLevel isolation_level);
/// The force parameter determines the behaviour of the garbage collector.
/// If it's set to true, it will behave as a global operation, i.e. it can't
/// be part of a transaction, and no other transaction can be active at the same time.
/// This allows it to delete immediately vertices without worrying that some other
/// transaction is possibly using it. If there are active transactions when this method
/// is called with force set to true, it will fallback to the same method with the force
/// set to false.
/// If it's set to false, it will execute in parallel with other transactions, ensuring
/// that no object in use can be deleted.
/// @throw std::system_error
/// @throw std::bad_alloc
template <bool force>
void CollectGarbage();
bool InitializeWalFile();
void FinalizeWalFile();
void AppendToWal(const Transaction &transaction, uint64_t final_commit_timestamp);
void AppendToWal(durability::StorageGlobalOperation operation, LabelId label, const std::set<PropertyId> &properties,
uint64_t final_commit_timestamp);
Transaction &GetTransaction(coordinator::Hlc start_timestamp, IsolationLevel isolation_level);
uint64_t CommitTimestamp(std::optional<uint64_t> desired_commit_timestamp = {});
@ -537,45 +392,22 @@ class Shard final {
SchemaValidator schema_validator_;
VertexValidator vertex_validator_;
Constraints constraints_;
Indices indices_;
Schemas schemas_;
// Transaction engine
uint64_t timestamp_{kTimestampInitialId};
uint64_t transaction_id_{kTransactionInitialId};
// TODO: This isn't really a commit log, it doesn't even care if a
// transaction commited or aborted. We could probably combine this with
// `timestamp_` in a sensible unit, something like TransactionClock or
// whatever.
std::optional<CommitLog> commit_log_;
std::list<Transaction> committed_transactions_;
std::list<Transaction *> committed_transactions_;
IsolationLevel isolation_level_;
Config config_;
// Undo buffers that were unlinked and now are waiting to be freed.
std::list<std::pair<uint64_t, std::list<Delta>>> garbage_undo_buffers_;
// Vertices that are logically deleted but still have to be removed from
// indices before removing them from the main storage.
std::list<PrimaryKey> deleted_vertices_;
// Vertices that are logically deleted and removed from indices and now wait
// to be removed from the main storage.
std::list<std::pair<uint64_t, PrimaryKey>> garbage_vertices_;
// Edges that are logically deleted and wait to be removed from the main
// storage.
std::list<Gid> deleted_edges_;
// Durability
std::filesystem::path snapshot_directory_;
std::filesystem::path wal_directory_;
std::filesystem::path lock_file_path_;
utils::OutputFile lock_file_handle_;
// UUID used to distinguish snapshots and to link snapshots to WALs
std::string uuid_;
// Sequence number used to keep track of the chain of WALs.
@ -601,7 +433,6 @@ class Shard final {
// epoch.
std::deque<std::pair<std::string, uint64_t>> epoch_history_;
std::optional<durability::WalFile> wal_file_;
uint64_t wal_unsynced_transactions_{0};
utils::FileRetainer file_retainer_;
@ -609,27 +440,10 @@ class Shard final {
// Global locker that is used for clients file locking
utils::FileRetainer::FileLocker global_locker_;
// Last commited timestamp
uint64_t last_commit_timestamp_{kTimestampInitialId};
class ReplicationServer;
std::unique_ptr<ReplicationServer> replication_server_{nullptr};
class ReplicationClient;
// We create ReplicationClient using unique_ptr so we can move
// newly created client into the vector.
// We cannot move the client directly because it contains ThreadPool
// which cannot be moved. Also, the move is necessary because
// we don't want to create the client directly inside the vector
// because that would require the lock on the list putting all
// commits (they iterate list of clients) to halt.
// This way we can initialize client in main thread which means
// that we can immediately notify the user if the initialization
// failed.
using ReplicationClientList = utils::Synchronized<std::vector<std::unique_ptr<ReplicationClient>>, utils::SpinLock>;
ReplicationClientList replication_clients_;
ReplicationRole replication_role_{ReplicationRole::MAIN};
// Holds all of the (in progress, committed and aborted) transactions that are read or write to this shard, but
// haven't been cleaned up yet
std::map<uint64_t, std::unique_ptr<Transaction>> start_logical_id_to_transaction_{};
bool has_any_transaction_aborted_since_last_gc{false};
};
} // namespace memgraph::storage::v3

View File

@ -12,6 +12,7 @@
#include <iterator>
#include <utility>
#include "query/v2/requests.hpp"
#include "storage/v3/shard_rsm.hpp"
#include "storage/v3/vertex_accessor.hpp"
@ -193,7 +194,7 @@ Value ConstructValueVertex(const memgraph::storage::v3::VertexAccessor &acc, mem
namespace memgraph::storage::v3 {
msgs::WriteResponses ShardRsm::ApplyWrite(msgs::CreateVerticesRequest &&req) {
auto acc = shard_->Access();
auto acc = shard_->Access(req.transaction_id);
// Workaround untill we have access to CreateVertexAndValidate()
// with the new signature that does not require the primary label.
@ -243,23 +244,12 @@ msgs::WriteResponses ShardRsm::ApplyWrite(msgs::CreateVerticesRequest &&req) {
}
}
msgs::CreateVerticesResponse resp{};
resp.success = action_successful;
if (action_successful) {
auto result = acc.Commit(req.transaction_id.logical_id);
if (result.HasError()) {
resp.success = false;
spdlog::debug(&"ConstraintViolation, commiting vertices was unsuccesfull with transaction id: "[req.transaction_id
.logical_id]);
}
}
return resp;
return msgs::CreateVerticesResponse{action_successful};
}
msgs::WriteResponses ShardRsm::ApplyWrite(msgs::DeleteVerticesRequest &&req) {
bool action_successful = true;
auto acc = shard_->Access();
auto acc = shard_->Access(req.transaction_id);
for (auto &propval : req.primary_keys) {
if (!action_successful) {
@ -269,9 +259,8 @@ msgs::WriteResponses ShardRsm::ApplyWrite(msgs::DeleteVerticesRequest &&req) {
auto vertex_acc = acc.FindVertex(ConvertPropertyVector(std::move(propval)), View::OLD);
if (!vertex_acc) {
spdlog::debug(
&"Error while trying to delete vertex. Vertex to delete does not exist. Transaction id: "[req.transaction_id
.logical_id]);
spdlog::debug("Error while trying to delete vertex. Vertex to delete does not exist. Transaction id: {}",
req.transaction_id.logical_id);
action_successful = false;
} else {
// TODO(gvolfing)
@ -282,7 +271,7 @@ msgs::WriteResponses ShardRsm::ApplyWrite(msgs::DeleteVerticesRequest &&req) {
auto result = acc.DeleteVertex(&vertex_acc.value());
if (result.HasError() || !(result.GetValue().has_value())) {
action_successful = false;
spdlog::debug(&"Error while trying to delete vertex. Transaction id: "[req.transaction_id.logical_id]);
spdlog::debug("Error while trying to delete vertex. Transaction id: {}", req.transaction_id.logical_id);
}
break;
@ -291,8 +280,8 @@ msgs::WriteResponses ShardRsm::ApplyWrite(msgs::DeleteVerticesRequest &&req) {
auto result = acc.DetachDeleteVertex(&vertex_acc.value());
if (result.HasError() || !(result.GetValue().has_value())) {
action_successful = false;
spdlog::debug(
&"Error while trying to detach and delete vertex. Transaction id: "[req.transaction_id.logical_id]);
spdlog::debug("Error while trying to detach and delete vertex. Transaction id: {}",
req.transaction_id.logical_id);
}
break;
@ -301,23 +290,11 @@ msgs::WriteResponses ShardRsm::ApplyWrite(msgs::DeleteVerticesRequest &&req) {
}
}
msgs::DeleteVerticesResponse resp{};
resp.success = action_successful;
if (action_successful) {
auto result = acc.Commit(req.transaction_id.logical_id);
if (result.HasError()) {
resp.success = false;
spdlog::debug(&"ConstraintViolation, commiting vertices was unsuccesfull with transaction id: "[req.transaction_id
.logical_id]);
}
}
return resp;
return msgs::DeleteVerticesResponse{action_successful};
}
msgs::WriteResponses ShardRsm::ApplyWrite(msgs::CreateEdgesRequest &&req) {
auto acc = shard_->Access();
auto acc = shard_->Access(req.transaction_id);
bool action_successful = true;
for (auto &edge : req.edges) {
@ -329,8 +306,8 @@ msgs::WriteResponses ShardRsm::ApplyWrite(msgs::CreateEdgesRequest &&req) {
if (!vertex_from_acc || !vertex_to_acc) {
action_successful = false;
spdlog::debug(
&"Error while trying to insert edge, vertex does not exist. Transaction id: "[req.transaction_id.logical_id]);
spdlog::debug("Error while trying to insert edge, vertex does not exist. Transaction id: {}",
req.transaction_id.logical_id);
break;
}
@ -341,30 +318,16 @@ msgs::WriteResponses ShardRsm::ApplyWrite(msgs::CreateEdgesRequest &&req) {
if (edge_acc.HasError()) {
action_successful = false;
spdlog::debug(&"Creating edge was not successful. Transaction id: "[req.transaction_id.logical_id]);
spdlog::debug("Creating edge was not successful. Transaction id: {}", req.transaction_id.logical_id);
break;
}
}
msgs::CreateEdgesResponse resp{};
resp.success = action_successful;
if (action_successful) {
auto result = acc.Commit(req.transaction_id.logical_id);
if (result.HasError()) {
resp.success = false;
spdlog::debug(
&"ConstraintViolation, commiting edge creation was unsuccesfull with transaction id: "[req.transaction_id
.logical_id]);
}
}
return resp;
return msgs::CreateEdgesResponse{action_successful};
}
msgs::ReadResponses ShardRsm::HandleRead(msgs::ScanVerticesRequest &&req) {
auto acc = shard_->Access();
auto acc = shard_->Access(req.transaction_id);
bool action_successful = true;
std::vector<msgs::ScanResultRow> results;
@ -421,6 +384,11 @@ msgs::ReadResponses ShardRsm::HandleRead(msgs::ScanVerticesRequest &&req) {
return resp;
}
msgs::WriteResponses ShardRsm::ApplyWrite(msgs::CommitRequest &&req) {
shard_->Access(req.transaction_id).Commit(req.commit_timestamp);
return msgs::CommitResponse{true};
};
// NOLINTNEXTLINE(readability-convert-member-functions-to-static)
msgs::WriteResponses ShardRsm::ApplyWrite(msgs::UpdateVerticesRequest && /*req*/) {
return msgs::UpdateVerticesResponse{};

View File

@ -14,7 +14,6 @@
#include <memory>
#include <variant>
#include <openssl/ec.h>
#include "query/v2/requests.hpp"
#include "storage/v3/shard.hpp"
#include "storage/v3/vertex_accessor.hpp"
@ -39,6 +38,8 @@ class ShardRsm {
msgs::WriteResponses ApplyWrite(msgs::DeleteEdgesRequest &&req);
msgs::WriteResponses ApplyWrite(msgs::UpdateEdgesRequest &&req);
msgs::WriteResponses ApplyWrite(msgs::CommitRequest &&req);
public:
explicit ShardRsm(std::unique_ptr<Shard> &&shard) : shard_(std::move(shard)){};

View File

@ -11,13 +11,11 @@
#pragma once
#include <atomic>
#include <limits>
#include <list>
#include <memory>
#include "utils/skip_list.hpp"
#include "coordinator/hybrid_logical_clock.hpp"
#include "storage/v3/delta.hpp"
#include "storage/v3/edge.hpp"
#include "storage/v3/isolation_level.hpp"
@ -27,24 +25,27 @@
namespace memgraph::storage::v3 {
const uint64_t kTimestampInitialId = 0;
const uint64_t kTransactionInitialId = 1ULL << 63U;
struct CommitInfo {
bool is_locally_committed{false};
coordinator::Hlc start_or_commit_timestamp;
};
struct Transaction {
Transaction(uint64_t transaction_id, uint64_t start_timestamp, IsolationLevel isolation_level)
: transaction_id(transaction_id),
start_timestamp(start_timestamp),
Transaction(coordinator::Hlc start_timestamp, IsolationLevel isolation_level)
: start_timestamp(start_timestamp),
commit_info(std::make_unique<CommitInfo>(CommitInfo{false, {start_timestamp}})),
command_id(0),
must_abort(false),
is_aborted(false),
isolation_level(isolation_level) {}
Transaction(Transaction &&other) noexcept
: transaction_id(other.transaction_id),
start_timestamp(other.start_timestamp),
commit_timestamp(std::move(other.commit_timestamp)),
: start_timestamp(other.start_timestamp),
commit_info(std::move(other.commit_info)),
command_id(other.command_id),
deltas(std::move(other.deltas)),
must_abort(other.must_abort),
is_aborted(other.is_aborted),
isolation_level(other.isolation_level) {}
Transaction(const Transaction &) = delete;
@ -53,32 +54,23 @@ struct Transaction {
~Transaction() {}
/// @throw std::bad_alloc if failed to create the `commit_timestamp`
void EnsureCommitTimestampExists() {
if (commit_timestamp != nullptr) return;
commit_timestamp = std::make_unique<std::atomic<uint64_t>>(transaction_id);
}
uint64_t transaction_id;
uint64_t start_timestamp;
// The `Transaction` object is stack allocated, but the `commit_timestamp`
// must be heap allocated because `Delta`s have a pointer to it, and that
// pointer must stay valid after the `Transaction` is moved into
// `commited_transactions_` list for GC.
std::unique_ptr<std::atomic<uint64_t>> commit_timestamp;
coordinator::Hlc start_timestamp;
std::unique_ptr<CommitInfo> commit_info;
uint64_t command_id;
std::list<Delta> deltas;
bool must_abort;
bool is_aborted;
IsolationLevel isolation_level;
};
// Relies on start timestamps are unique
inline bool operator==(const Transaction &first, const Transaction &second) {
return first.transaction_id == second.transaction_id;
return first.start_timestamp == second.start_timestamp;
}
inline bool operator<(const Transaction &first, const Transaction &second) {
return first.transaction_id < second.transaction_id;
return first.start_timestamp < second.start_timestamp;
}
inline bool operator==(const Transaction &first, const uint64_t &second) { return first.transaction_id == second; }
inline bool operator<(const Transaction &first, const uint64_t &second) { return first.transaction_id < second; }
inline bool operator==(const Transaction &first, const uint64_t second) { return first.start_timestamp == second; }
inline bool operator<(const Transaction &first, const uint64_t second) { return first.start_timestamp < second; }
} // namespace memgraph::storage::v3

View File

@ -64,13 +64,13 @@ std::pair<bool, bool> IsVisible(Vertex *vertex, Transaction *transaction, View v
} // namespace detail
std::optional<VertexAccessor> VertexAccessor::Create(Vertex *vertex, Transaction *transaction, Indices *indices,
Constraints *constraints, Config::Items config,
const VertexValidator &vertex_validator, View view) {
Config::Items config, const VertexValidator &vertex_validator,
View view) {
if (const auto [exists, deleted] = detail::IsVisible(vertex, transaction, view); !exists || deleted) {
return std::nullopt;
}
return VertexAccessor{vertex, transaction, indices, constraints, config, vertex_validator};
return VertexAccessor{vertex, transaction, indices, config, vertex_validator};
}
bool VertexAccessor::IsVisible(View view) const {
@ -544,7 +544,7 @@ Result<std::vector<EdgeAccessor>> VertexAccessor::InEdges(View view, const std::
const auto id = VertexId{vertex_validator_->primary_label_, vertex_->keys.Keys()};
for (const auto &item : in_edges) {
const auto &[edge_type, from_vertex, edge] = item;
ret.emplace_back(edge, edge_type, from_vertex, id, transaction_, indices_, constraints_, config_);
ret.emplace_back(edge, edge_type, from_vertex, id, transaction_, indices_, config_);
}
return ret;
}
@ -624,7 +624,7 @@ Result<std::vector<EdgeAccessor>> VertexAccessor::OutEdges(View view, const std:
const auto id = VertexId{vertex_validator_->primary_label_, vertex_->keys.Keys()};
for (const auto &item : out_edges) {
const auto &[edge_type, to_vertex, edge] = item;
ret.emplace_back(edge, edge_type, id, to_vertex, transaction_, indices_, constraints_, config_);
ret.emplace_back(edge, edge_type, id, to_vertex, transaction_, indices_, config_);
}
return ret;
}

View File

@ -28,7 +28,6 @@ namespace memgraph::storage::v3 {
class EdgeAccessor;
class Shard;
struct Indices;
struct Constraints;
class VertexAccessor final {
private:
@ -37,19 +36,17 @@ class VertexAccessor final {
public:
// Be careful when using VertexAccessor since it can be instantiated with
// nullptr values
VertexAccessor(Vertex *vertex, Transaction *transaction, Indices *indices, Constraints *constraints,
Config::Items config, const VertexValidator &vertex_validator, bool for_deleted = false)
VertexAccessor(Vertex *vertex, Transaction *transaction, Indices *indices, Config::Items config,
const VertexValidator &vertex_validator, bool for_deleted = false)
: vertex_(vertex),
transaction_(transaction),
indices_(indices),
constraints_(constraints),
config_(config),
vertex_validator_{&vertex_validator},
for_deleted_(for_deleted) {}
static std::optional<VertexAccessor> Create(Vertex *vertex, Transaction *transaction, Indices *indices,
Constraints *constraints, Config::Items config,
const VertexValidator &vertex_validator, View view);
Config::Items config, const VertexValidator &vertex_validator, View view);
/// @return true if the object is visible from the current transaction
bool IsVisible(View view) const;
@ -141,7 +138,6 @@ class VertexAccessor final {
Vertex *vertex_;
Transaction *transaction_;
Indices *indices_;
Constraints *constraints_;
Config::Items config_;
const VertexValidator *vertex_validator_;

View File

@ -37,7 +37,7 @@ namespace memgraph::logging {
// compilers
inline void AssertFailed(const char *file_name, int line_num, const char *expr, const std::string &message) {
spdlog::critical(
"\nAssertion failed in file {} at line {}."
"\nAssertion failed at {}:{}."
"\n\tExpression: '{}'"
"{}",
file_name, line_num, expr, !message.empty() ? fmt::format("\n\tMessage: '{}'", message) : "");

View File

@ -9,6 +9,7 @@
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include <chrono>
#include <iostream>
#include <optional>
#include <thread>
@ -126,17 +127,15 @@ bool AttemtpToCreateVertex(ShardClient &client, int64_t value) {
create_req.new_vertices = {vertex};
create_req.transaction_id.logical_id = GetTransactionId();
while (true) {
auto write_res = client.SendWriteRequest(create_req);
if (write_res.HasError()) {
continue;
}
auto write_res = client.SendWriteRequest(create_req);
MG_ASSERT(write_res.HasValue() && std::get<msgs::CreateVerticesResponse>(write_res.GetValue()).success,
"Unexpected failure");
auto write_response_result = write_res.GetValue();
auto write_response = std::get<msgs::CreateVerticesResponse>(write_response_result);
return write_response.success;
}
auto commit_req = msgs::CommitRequest{create_req.transaction_id, msgs::Hlc{.logical_id = GetTransactionId()}};
auto commit_res = client.SendWriteRequest(commit_req);
MG_ASSERT(commit_res.HasValue() && std::get<msgs::CommitResponse>(commit_res.GetValue()).success,
"Unexpected failure");
return true;
}
bool AttemptToAddEdge(ShardClient &client, int64_t value_of_vertex_1, int64_t value_of_vertex_2, int64_t edge_gid,
@ -161,17 +160,15 @@ bool AttemptToAddEdge(ShardClient &client, int64_t value_of_vertex_1, int64_t va
create_req.edges = {edge};
create_req.transaction_id.logical_id = GetTransactionId();
while (true) {
auto write_res = client.SendWriteRequest(create_req);
if (write_res.HasError()) {
continue;
}
auto write_res = client.SendWriteRequest(create_req);
MG_ASSERT(write_res.HasValue() && std::get<msgs::CreateEdgesResponse>(write_res.GetValue()).success,
"Unexpected failure");
auto write_response_result = write_res.GetValue();
auto write_response = std::get<msgs::CreateEdgesResponse>(write_response_result);
return write_response.success;
}
auto commit_req = msgs::CommitRequest{create_req.transaction_id, msgs::Hlc{.logical_id = GetTransactionId()}};
auto commit_res = client.SendWriteRequest(commit_req);
MG_ASSERT(commit_res.HasValue() && std::get<msgs::CommitResponse>(commit_res.GetValue()).success,
"Unexpected failure");
return true;
}
std::tuple<size_t, std::optional<msgs::VertexId>> AttemptToScanAllWithBatchLimit(ShardClient &client,
@ -298,12 +295,16 @@ int TestMessages() {
auto simulator = Simulator(config);
Io<SimulatorTransport> shard_server_io_1 = simulator.RegisterNew();
shard_server_io_1.SetDefaultTimeout(std::chrono::seconds(1));
const auto shard_server_1_address = shard_server_io_1.GetAddress();
Io<SimulatorTransport> shard_server_io_2 = simulator.RegisterNew();
shard_server_io_2.SetDefaultTimeout(std::chrono::seconds(1));
const auto shard_server_2_address = shard_server_io_2.GetAddress();
Io<SimulatorTransport> shard_server_io_3 = simulator.RegisterNew();
shard_server_io_3.SetDefaultTimeout(std::chrono::seconds(1));
const auto shard_server_3_address = shard_server_io_3.GetAddress();
Io<SimulatorTransport> shard_client_io = simulator.RegisterNew();
shard_client_io.SetDefaultTimeout(std::chrono::seconds(1));
PropertyValue min_pk(static_cast<int64_t>(0));
std::vector<PropertyValue> min_prim_key = {min_pk};

View File

@ -346,6 +346,9 @@ target_link_libraries(${test_prefix}storage_v3_vertex_accessors mg-storage-v3)
add_unit_test(storage_v3_edge.cpp)
target_link_libraries(${test_prefix}storage_v3_edge mg-storage-v3)
add_unit_test(storage_v3_isolation_level.cpp)
target_link_libraries(${test_prefix}storage_v3_isolation_level mg-storage-v3)
# Test mg-query-v2
add_unit_test(query_v2_interpreter.cpp ${CMAKE_SOURCE_DIR}/src/glue/v2/communication.cpp)
target_link_libraries(${test_prefix}query_v2_interpreter mg-storage-v3 mg-query-v2 mg-communication)

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

View File

@ -11,6 +11,7 @@
#include <gmock/gmock.h>
#include <gtest/gtest.h>
#include <chrono>
#include <exception>
#include <optional>
#include <vector>
@ -134,7 +135,7 @@ class ExpressionEvaluatorTest : public ::testing::Test {
PrimaryKey min_pk{PropertyValue(0)};
Shard db{primary_label, min_pk, std::nullopt};
Shard::Accessor storage_dba{db.Access()};
Shard::Accessor storage_dba{db.Access(GetNextHlc())};
DbAccessor dba{&storage_dba};
AstStorage storage;
@ -145,6 +146,8 @@ class ExpressionEvaluatorTest : public ::testing::Test {
Frame frame{128};
ExpressionEvaluator eval{&frame, symbol_table, ctx, &dba, View::OLD};
coordinator::Hlc last_hlc{0, io::Time{}};
void SetUp() override {
db.StoreMapping({{1, "label"}, {2, "property"}});
ASSERT_TRUE(
@ -186,6 +189,12 @@ class ExpressionEvaluatorTest : public ::testing::Test {
"EvaluationContext for allocations!";
return value;
}
coordinator::Hlc GetNextHlc() {
++last_hlc.logical_id;
last_hlc.coordinator_wall_clock += std::chrono::seconds(1);
return last_hlc;
}
};
TEST_F(ExpressionEvaluatorTest, OrOperator) {

View File

@ -10,6 +10,7 @@
// licenses/APL.txt.
#include <gmock/gmock-matchers.h>
#include <gmock/gmock-more-matchers.h>
#include <gmock/gmock.h>
#include <gtest/gtest.h>
#include <cstdint>
@ -19,6 +20,7 @@
#include "storage/v3/property_value.hpp"
#include "storage/v3/storage.hpp"
#include "storage/v3/temporal.hpp"
#include "storage/v3/view.hpp"
// NOLINTNEXTLINE(google-build-using-namespace)
@ -44,19 +46,23 @@ class IndexTest : public testing::Test {
Shard storage{primary_label, pk, std::nullopt};
const PropertyId primary_property{PropertyId::FromUint(2)};
const LabelId label1{LabelId::FromUint(3)};
const LabelId label2{LabelId::FromUint(4)};
const PropertyId prop_id{PropertyId::FromUint(5)};
const PropertyId prop_val{PropertyId::FromUint(6)};
const LabelId label1{LabelId::FromUint(3)};
const LabelId label2{LabelId::FromUint(4)};
static constexpr std::chrono::seconds wall_clock_increment{10};
static constexpr std::chrono::seconds reclamation_interval{wall_clock_increment / 2};
static constexpr io::Duration one_time_unit{1};
int primary_key_id{0};
int vertex_id{0};
coordinator::Hlc last_hlc{0, io::Time{}};
LabelId NameToLabelId(std::string_view label_name) { return storage.NameToLabel(label_name); }
PropertyId NameToPropertyId(std::string_view property_name) { return storage.NameToProperty(property_name); }
VertexAccessor CreateVertex(Shard::Accessor *accessor) {
auto vertex = *accessor->CreateVertexAndValidate(
VertexAccessor CreateVertex(Shard::Accessor &accessor) {
auto vertex = *accessor.CreateVertexAndValidate(
primary_label, {},
{{primary_property, PropertyValue(primary_key_id++)}, {prop_id, PropertyValue(vertex_id++)}});
return vertex;
@ -82,37 +88,47 @@ class IndexTest : public testing::Test {
}
return ret;
}
coordinator::Hlc GetNextHlc() {
++last_hlc.logical_id;
last_hlc.coordinator_wall_clock += wall_clock_increment;
return last_hlc;
}
void CleanupHlc(const coordinator::Hlc hlc) {
storage.CollectGarbage(hlc.coordinator_wall_clock + reclamation_interval + one_time_unit);
}
};
// NOLINTNEXTLINE(hicpp-special-member-functions)
TEST_F(IndexTest, LabelIndexCreate) {
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_FALSE(acc.LabelIndexExists(label1));
}
EXPECT_EQ(storage.ListAllIndices().label.size(), 0);
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
for (int i = 0; i < 10; ++i) {
auto vertex = CreateVertex(&acc);
auto vertex = CreateVertex(acc);
ASSERT_NO_ERROR(vertex.AddLabelAndValidate(i % 2 ? label1 : label2));
}
ASSERT_NO_ERROR(acc.Commit());
acc.Commit(GetNextHlc());
}
EXPECT_TRUE(storage.CreateIndex(label1));
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_THAT(GetIds(acc.Vertices(label1, View::OLD), View::OLD), UnorderedElementsAre(1, 3, 5, 7, 9));
EXPECT_THAT(GetIds(acc.Vertices(label1, View::NEW), View::NEW), UnorderedElementsAre(1, 3, 5, 7, 9));
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
for (int i = 10; i < 20; ++i) {
auto vertex = CreateVertex(&acc);
auto vertex = CreateVertex(acc);
ASSERT_NO_ERROR(vertex.AddLabelAndValidate(i % 2 ? label1 : label2));
}
@ -131,9 +147,9 @@ TEST_F(IndexTest, LabelIndexCreate) {
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
for (int i = 10; i < 20; ++i) {
auto vertex = CreateVertex(&acc);
auto vertex = CreateVertex(acc);
ASSERT_NO_ERROR(vertex.AddLabelAndValidate(i % 2 ? label1 : label2));
}
@ -148,11 +164,11 @@ TEST_F(IndexTest, LabelIndexCreate) {
EXPECT_THAT(GetIds(acc.Vertices(label1, View::NEW), View::NEW),
UnorderedElementsAre(1, 3, 5, 7, 9, 21, 23, 25, 27, 29));
ASSERT_NO_ERROR(acc.Commit());
acc.Commit(GetNextHlc());
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_THAT(GetIds(acc.Vertices(label1, View::OLD), View::OLD),
UnorderedElementsAre(1, 3, 5, 7, 9, 21, 23, 25, 27, 29));
EXPECT_THAT(GetIds(acc.Vertices(label1, View::NEW), View::NEW),
@ -165,67 +181,67 @@ TEST_F(IndexTest, LabelIndexCreate) {
EXPECT_THAT(GetIds(acc.Vertices(label1, View::NEW), View::NEW),
UnorderedElementsAre(1, 3, 5, 7, 9, 21, 23, 25, 27, 29));
ASSERT_NO_ERROR(acc.Commit());
acc.Commit(GetNextHlc());
}
}
// NOLINTNEXTLINE(hicpp-special-member-functions)
TEST_F(IndexTest, LabelIndexDrop) {
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_FALSE(acc.LabelIndexExists(label1));
}
EXPECT_EQ(storage.ListAllIndices().label.size(), 0);
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
for (int i = 0; i < 10; ++i) {
auto vertex = CreateVertex(&acc);
auto vertex = CreateVertex(acc);
ASSERT_NO_ERROR(vertex.AddLabelAndValidate(i % 2 ? label1 : label2));
}
ASSERT_NO_ERROR(acc.Commit());
acc.Commit(GetNextHlc());
}
EXPECT_TRUE(storage.CreateIndex(label1));
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_THAT(GetIds(acc.Vertices(label1, View::OLD), View::OLD), UnorderedElementsAre(1, 3, 5, 7, 9));
EXPECT_THAT(GetIds(acc.Vertices(label1, View::NEW), View::NEW), UnorderedElementsAre(1, 3, 5, 7, 9));
}
EXPECT_TRUE(storage.DropIndex(label1));
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_FALSE(acc.LabelIndexExists(label1));
}
EXPECT_EQ(storage.ListAllIndices().label.size(), 0);
EXPECT_FALSE(storage.DropIndex(label1));
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_FALSE(acc.LabelIndexExists(label1));
}
EXPECT_EQ(storage.ListAllIndices().label.size(), 0);
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
for (int i = 10; i < 20; ++i) {
auto vertex = CreateVertex(&acc);
auto vertex = CreateVertex(acc);
ASSERT_NO_ERROR(vertex.AddLabelAndValidate(i % 2 ? label1 : label2));
}
ASSERT_NO_ERROR(acc.Commit());
acc.Commit(GetNextHlc());
}
EXPECT_TRUE(storage.CreateIndex(label1));
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_TRUE(acc.LabelIndexExists(label1));
}
EXPECT_THAT(storage.ListAllIndices().label, UnorderedElementsAre(label1));
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_THAT(GetIds(acc.Vertices(label1, View::OLD), View::OLD),
UnorderedElementsAre(1, 3, 5, 7, 9, 11, 13, 15, 17, 19));
@ -253,7 +269,7 @@ TEST_F(IndexTest, LabelIndexBasic) {
EXPECT_TRUE(storage.CreateIndex(label1));
EXPECT_TRUE(storage.CreateIndex(label2));
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_THAT(storage.ListAllIndices().label, UnorderedElementsAre(label1, label2));
EXPECT_THAT(GetIds(acc.Vertices(label1, View::OLD), View::OLD), IsEmpty());
EXPECT_THAT(GetIds(acc.Vertices(label2, View::OLD), View::OLD), IsEmpty());
@ -261,7 +277,7 @@ TEST_F(IndexTest, LabelIndexBasic) {
EXPECT_THAT(GetIds(acc.Vertices(label2, View::NEW), View::NEW), IsEmpty());
for (int i = 0; i < 10; ++i) {
auto vertex = CreateVertex(&acc);
auto vertex = CreateVertex(acc);
ASSERT_NO_ERROR(vertex.AddLabelAndValidate(i % 2 ? label1 : label2));
}
@ -319,19 +335,19 @@ TEST_F(IndexTest, LabelIndexDuplicateVersions) {
EXPECT_TRUE(storage.CreateIndex(label2));
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
for (int i = 0; i < 5; ++i) {
auto vertex = CreateVertex(&acc);
auto vertex = CreateVertex(acc);
ASSERT_NO_ERROR(vertex.AddLabelAndValidate(label1));
}
EXPECT_THAT(GetIds(acc.Vertices(label1, View::NEW), View::NEW), UnorderedElementsAre(0, 1, 2, 3, 4));
ASSERT_NO_ERROR(acc.Commit());
acc.Commit(GetNextHlc());
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_THAT(GetIds(acc.Vertices(label1, View::OLD), View::OLD), UnorderedElementsAre(0, 1, 2, 3, 4));
for (auto vertex : acc.Vertices(View::OLD)) {
@ -355,12 +371,12 @@ TEST_F(IndexTest, LabelIndexTransactionalIsolation) {
EXPECT_TRUE(storage.CreateIndex(label1));
EXPECT_TRUE(storage.CreateIndex(label2));
auto acc_before = storage.Access();
auto acc = storage.Access();
auto acc_after = storage.Access();
auto acc_before = storage.Access(GetNextHlc());
auto acc = storage.Access(GetNextHlc());
auto acc_after = storage.Access(GetNextHlc());
for (int i = 0; i < 5; ++i) {
auto vertex = CreateVertex(&acc);
auto vertex = CreateVertex(acc);
ASSERT_NO_ERROR(vertex.AddLabelAndValidate(label1));
}
@ -368,9 +384,9 @@ TEST_F(IndexTest, LabelIndexTransactionalIsolation) {
EXPECT_THAT(GetIds(acc_before.Vertices(label1, View::NEW), View::NEW), IsEmpty());
EXPECT_THAT(GetIds(acc_after.Vertices(label1, View::NEW), View::NEW), IsEmpty());
ASSERT_NO_ERROR(acc.Commit());
acc.Commit(GetNextHlc());
auto acc_after_commit = storage.Access();
auto acc_after_commit = storage.Access(GetNextHlc());
EXPECT_THAT(GetIds(acc_before.Vertices(label1, View::NEW), View::NEW), IsEmpty());
EXPECT_THAT(GetIds(acc_after.Vertices(label1, View::NEW), View::NEW), IsEmpty());
@ -382,9 +398,9 @@ TEST_F(IndexTest, LabelIndexCountEstimate) {
EXPECT_TRUE(storage.CreateIndex(label1));
EXPECT_TRUE(storage.CreateIndex(label2));
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
for (int i = 0; i < 20; ++i) {
auto vertex = CreateVertex(&acc);
auto vertex = CreateVertex(acc);
ASSERT_NO_ERROR(vertex.AddLabelAndValidate(i % 3 ? label1 : label2));
}
@ -397,12 +413,12 @@ TEST_F(IndexTest, LabelPropertyIndexCreateAndDrop) {
EXPECT_EQ(storage.ListAllIndices().label_property.size(), 0);
EXPECT_TRUE(storage.CreateIndex(label1, prop_id));
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_TRUE(acc.LabelPropertyIndexExists(label1, prop_id));
}
EXPECT_THAT(storage.ListAllIndices().label_property, UnorderedElementsAre(std::make_pair(label1, prop_id)));
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_FALSE(acc.LabelPropertyIndexExists(label2, prop_id));
}
EXPECT_FALSE(storage.CreateIndex(label1, prop_id));
@ -410,7 +426,7 @@ TEST_F(IndexTest, LabelPropertyIndexCreateAndDrop) {
EXPECT_TRUE(storage.CreateIndex(label2, prop_id));
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_TRUE(acc.LabelPropertyIndexExists(label2, prop_id));
}
EXPECT_THAT(storage.ListAllIndices().label_property,
@ -418,7 +434,7 @@ TEST_F(IndexTest, LabelPropertyIndexCreateAndDrop) {
EXPECT_TRUE(storage.DropIndex(label1, prop_id));
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_FALSE(acc.LabelPropertyIndexExists(label1, prop_id));
}
EXPECT_THAT(storage.ListAllIndices().label_property, UnorderedElementsAre(std::make_pair(label2, prop_id)));
@ -426,7 +442,7 @@ TEST_F(IndexTest, LabelPropertyIndexCreateAndDrop) {
EXPECT_TRUE(storage.DropIndex(label2, prop_id));
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_FALSE(acc.LabelPropertyIndexExists(label2, prop_id));
}
EXPECT_EQ(storage.ListAllIndices().label_property.size(), 0);
@ -442,11 +458,11 @@ TEST_F(IndexTest, LabelPropertyIndexBasic) {
storage.CreateIndex(label1, prop_val);
storage.CreateIndex(label2, prop_val);
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_THAT(GetIds(acc.Vertices(label1, prop_val, View::OLD), View::OLD), IsEmpty());
for (int i = 0; i < 10; ++i) {
auto vertex = CreateVertex(&acc);
auto vertex = CreateVertex(acc);
ASSERT_NO_ERROR(vertex.AddLabelAndValidate(i % 2 ? label1 : label2));
ASSERT_NO_ERROR(vertex.SetPropertyAndValidate(prop_val, PropertyValue(i)));
}
@ -501,20 +517,20 @@ TEST_F(IndexTest, LabelPropertyIndexBasic) {
TEST_F(IndexTest, LabelPropertyIndexDuplicateVersions) {
storage.CreateIndex(label1, prop_val);
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
for (int i = 0; i < 5; ++i) {
auto vertex = CreateVertex(&acc);
auto vertex = CreateVertex(acc);
ASSERT_NO_ERROR(vertex.AddLabelAndValidate(label1));
ASSERT_NO_ERROR(vertex.SetPropertyAndValidate(prop_val, PropertyValue(i)));
}
EXPECT_THAT(GetIds(acc.Vertices(label1, prop_val, View::NEW), View::NEW), UnorderedElementsAre(0, 1, 2, 3, 4));
ASSERT_NO_ERROR(acc.Commit());
acc.Commit(GetNextHlc());
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_THAT(GetIds(acc.Vertices(label1, prop_val, View::OLD), View::OLD), UnorderedElementsAre(0, 1, 2, 3, 4));
for (auto vertex : acc.Vertices(View::OLD)) {
@ -536,12 +552,12 @@ TEST_F(IndexTest, LabelPropertyIndexDuplicateVersions) {
TEST_F(IndexTest, LabelPropertyIndexTransactionalIsolation) {
storage.CreateIndex(label1, prop_val);
auto acc_before = storage.Access();
auto acc = storage.Access();
auto acc_after = storage.Access();
auto acc_before = storage.Access(GetNextHlc());
auto acc = storage.Access(GetNextHlc());
auto acc_after = storage.Access(GetNextHlc());
for (int i = 0; i < 5; ++i) {
auto vertex = CreateVertex(&acc);
auto vertex = CreateVertex(acc);
ASSERT_NO_ERROR(vertex.AddLabelAndValidate(label1));
ASSERT_NO_ERROR(vertex.SetPropertyAndValidate(prop_val, PropertyValue(i)));
}
@ -550,9 +566,9 @@ TEST_F(IndexTest, LabelPropertyIndexTransactionalIsolation) {
EXPECT_THAT(GetIds(acc_before.Vertices(label1, prop_val, View::NEW), View::NEW), IsEmpty());
EXPECT_THAT(GetIds(acc_after.Vertices(label1, prop_val, View::NEW), View::NEW), IsEmpty());
ASSERT_NO_ERROR(acc.Commit());
acc.Commit(GetNextHlc());
auto acc_after_commit = storage.Access();
auto acc_after_commit = storage.Access(GetNextHlc());
EXPECT_THAT(GetIds(acc_before.Vertices(label1, prop_val, View::NEW), View::NEW), IsEmpty());
EXPECT_THAT(GetIds(acc_after.Vertices(label1, prop_val, View::NEW), View::NEW), IsEmpty());
@ -571,17 +587,17 @@ TEST_F(IndexTest, LabelPropertyIndexFiltering) {
storage.CreateIndex(label1, prop_val);
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
for (int i = 0; i < 10; ++i) {
auto vertex = CreateVertex(&acc);
auto vertex = CreateVertex(acc);
ASSERT_NO_ERROR(vertex.AddLabelAndValidate(label1));
ASSERT_NO_ERROR(vertex.SetPropertyAndValidate(prop_val, i % 2 ? PropertyValue(i / 2) : PropertyValue(i / 2.0)));
}
ASSERT_NO_ERROR(acc.Commit());
acc.Commit(GetNextHlc());
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
for (int i = 0; i < 5; ++i) {
EXPECT_THAT(GetIds(acc.Vertices(label1, prop_val, PropertyValue(i), View::OLD)),
UnorderedElementsAre(2 * i, 2 * i + 1));
@ -628,10 +644,10 @@ TEST_F(IndexTest, LabelPropertyIndexFiltering) {
TEST_F(IndexTest, LabelPropertyIndexCountEstimate) {
storage.CreateIndex(label1, prop_val);
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
for (int i = 1; i <= 10; ++i) {
for (int j = 0; j < i; ++j) {
auto vertex = CreateVertex(&acc);
auto vertex = CreateVertex(acc);
ASSERT_NO_ERROR(vertex.AddLabelAndValidate(label1));
ASSERT_NO_ERROR(vertex.SetPropertyAndValidate(prop_val, PropertyValue(i)));
}
@ -684,18 +700,18 @@ TEST_F(IndexTest, LabelPropertyIndexMixedIteration) {
// Create vertices, each with one of the values above.
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
for (const auto &value : values) {
auto v = acc.CreateVertexAndValidate(primary_label, {}, {{primary_property, PropertyValue(primary_key_id++)}});
ASSERT_TRUE(v->AddLabelAndValidate(label1).HasValue());
ASSERT_TRUE(v->SetPropertyAndValidate(prop_val, value).HasValue());
}
ASSERT_FALSE(acc.Commit().HasError());
acc.Commit(GetNextHlc());
}
// Verify that all nodes are in the index.
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
auto iterable = acc.Vertices(label1, prop_val, View::OLD);
auto it = iterable.begin();
for (const auto &value : values) {
@ -712,7 +728,7 @@ TEST_F(IndexTest, LabelPropertyIndexMixedIteration) {
auto verify = [&](const std::optional<memgraph::utils::Bound<PropertyValue>> &from,
const std::optional<memgraph::utils::Bound<PropertyValue>> &to,
const std::vector<PropertyValue> &expected) {
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
auto iterable = acc.Vertices(label1, prop_val, from, to, View::OLD);
size_t i = 0;
for (auto it = iterable.begin(); it != iterable.end(); ++it, ++i) {
@ -866,7 +882,7 @@ TEST_F(IndexTest, LabelPropertyIndexCreateWithExistingPrimaryKey) {
EXPECT_EQ(storage.ListAllIndices().label_property.size(), 1);
EXPECT_EQ(storage.ListAllIndices().label.size(), 0);
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_TRUE(acc.LabelPropertyIndexExists(primary_label, prop_id));
}
EXPECT_THAT(storage.ListAllIndices().label_property, UnorderedElementsAre(Pair(primary_label, prop_id)));
@ -884,12 +900,11 @@ TEST_F(IndexTest, LabelPropertyIndexCreateWithExistingPrimaryKey) {
TEST_F(IndexTest, LabelIndexCreateVertexAndValidate) {
{
auto acc = storage.Access();
EXPECT_EQ(storage.ListAllIndices().label.size(), 0);
EXPECT_EQ(storage.ListAllIndices().label_property.size(), 0);
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
// Create vertices with CreateVertexAndValidate
for (int i = 0; i < 5; ++i) {
@ -897,25 +912,25 @@ TEST_F(IndexTest, LabelIndexCreateVertexAndValidate) {
acc.CreateVertexAndValidate(primary_label, {label1}, {{primary_property, PropertyValue(primary_key_id++)}});
ASSERT_TRUE(vertex.HasValue());
}
ASSERT_NO_ERROR(acc.Commit());
acc.Commit(GetNextHlc());
}
{
EXPECT_TRUE(storage.CreateIndex(label1));
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_THAT(GetPrimaryKeyIds(acc.Vertices(label1, View::OLD), View::OLD), UnorderedElementsAre(0, 1, 2, 3, 4));
}
}
{
EXPECT_TRUE(storage.DropIndex(label1));
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_FALSE(acc.LabelIndexExists(label1));
}
EXPECT_EQ(storage.ListAllIndices().label.size(), 0);
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_TRUE(storage.CreateIndex(label1));
EXPECT_THAT(GetPrimaryKeyIds(acc.Vertices(label1, View::OLD), View::OLD), UnorderedElementsAre(0, 1, 2, 3, 4));
@ -933,12 +948,11 @@ TEST_F(IndexTest, LabelIndexCreateVertexAndValidate) {
TEST_F(IndexTest, LabelPropertyIndexCreateVertexAndValidate) {
{
auto acc = storage.Access();
EXPECT_EQ(storage.ListAllIndices().label.size(), 0);
EXPECT_EQ(storage.ListAllIndices().label_property.size(), 0);
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
// Create vertices with CreateVertexAndValidate
for (int i = 0; i < 5; ++i) {
@ -947,12 +961,12 @@ TEST_F(IndexTest, LabelPropertyIndexCreateVertexAndValidate) {
{{primary_property, PropertyValue(primary_key_id++)}, {prop_id, PropertyValue(vertex_id++)}});
ASSERT_TRUE(vertex.HasValue());
}
ASSERT_NO_ERROR(acc.Commit());
acc.Commit(GetNextHlc());
}
{
EXPECT_TRUE(storage.CreateIndex(label1, prop_id));
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_THAT(GetPrimaryKeyIds(acc.Vertices(label1, prop_id, View::OLD), View::OLD),
UnorderedElementsAre(0, 1, 2, 3, 4));
}
@ -960,13 +974,13 @@ TEST_F(IndexTest, LabelPropertyIndexCreateVertexAndValidate) {
{
EXPECT_TRUE(storage.DropIndex(label1, prop_id));
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_FALSE(acc.LabelPropertyIndexExists(label1, prop_id));
}
EXPECT_EQ(storage.ListAllIndices().label_property.size(), 0);
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
EXPECT_TRUE(storage.CreateIndex(label1, prop_id));
EXPECT_THAT(GetPrimaryKeyIds(acc.Vertices(label1, prop_id, View::OLD), View::OLD),
UnorderedElementsAre(0, 1, 2, 3, 4));
@ -984,4 +998,187 @@ TEST_F(IndexTest, LabelPropertyIndexCreateVertexAndValidate) {
UnorderedElementsAre(0, 1, 2, 3, 4));
}
}
TEST_F(IndexTest, CollectGarbageDeleteVertex) {
// First part
// T1 (start 1, commit 3) Creates the vertex, adds label and property
// T2 (start 2, no commit) reads nothing
// T3 (start 4, commit 6) deletes label and property
// T4 (start 5, no commit) reads the vertex
// T5 (start 7, no commit) reads nothing
auto t1_start = GetNextHlc();
auto t2_start = GetNextHlc();
auto t1_commit = GetNextHlc();
auto t3_start = GetNextHlc();
auto t4_start = GetNextHlc();
auto t3_commit = GetNextHlc();
auto t5_start = GetNextHlc();
ASSERT_TRUE(storage.CreateIndex(label1, prop_val));
auto acc1 = storage.Access(t1_start);
{
auto vertex = CreateVertex(acc1);
ASSERT_NO_ERROR(vertex.AddLabelAndValidate(label1));
ASSERT_NO_ERROR(vertex.SetPropertyAndValidate(prop_val, PropertyValue(42)));
}
acc1.Commit(t1_commit);
EXPECT_THAT(GetIds(acc1.Vertices(label1, prop_val, View::OLD), View::OLD), IsEmpty());
EXPECT_THAT(GetIds(acc1.Vertices(label1, prop_val, View::NEW), View::NEW), UnorderedElementsAre(0));
auto acc2 = storage.Access(t2_start);
EXPECT_THAT(GetIds(acc2.Vertices(View::OLD), View::OLD), IsEmpty());
EXPECT_THAT(GetIds(acc2.Vertices(View::NEW), View::NEW), IsEmpty());
EXPECT_THAT(GetIds(acc2.Vertices(label1, prop_val, View::OLD), View::OLD), IsEmpty());
EXPECT_THAT(GetIds(acc2.Vertices(label1, prop_val, View::NEW), View::NEW), IsEmpty());
auto acc3 = storage.Access(t3_start);
{
auto vertices = acc3.Vertices(label1, prop_val, View::OLD);
EXPECT_THAT(GetIds(acc3.Vertices(label1, prop_val, View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc3.Vertices(label1, prop_val, View::NEW), View::NEW), UnorderedElementsAre(0));
auto vertex = *vertices.begin();
ASSERT_NO_ERROR(acc3.DeleteVertex(&vertex));
EXPECT_THAT(GetIds(acc3.Vertices(label1, prop_val, View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc3.Vertices(label1, prop_val, View::NEW), View::NEW), IsEmpty());
}
acc3.Commit(t3_commit);
auto check_t3 = [this, &acc3]() {
EXPECT_THAT(GetIds(acc3.Vertices(label1, prop_val, View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc3.Vertices(label1, prop_val, View::NEW), View::NEW), IsEmpty());
};
check_t3();
auto acc4 = storage.Access(t4_start);
EXPECT_THAT(GetIds(acc4.Vertices(View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc4.Vertices(View::NEW), View::NEW), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc4.Vertices(label1, prop_val, View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc4.Vertices(label1, prop_val, View::NEW), View::NEW), UnorderedElementsAre(0));
auto acc5 = storage.Access(t5_start);
auto check_t5 = [this, &acc5]() {
EXPECT_THAT(GetIds(acc5.Vertices(View::OLD), View::OLD), IsEmpty());
EXPECT_THAT(GetIds(acc5.Vertices(View::NEW), View::NEW), IsEmpty());
EXPECT_THAT(GetIds(acc5.Vertices(label1, prop_val, View::OLD), View::OLD), IsEmpty());
EXPECT_THAT(GetIds(acc5.Vertices(label1, prop_val, View::NEW), View::NEW), IsEmpty());
};
check_t5();
// Second part
// Start to clean up things.
CleanupHlc(t1_start);
// As the deltas of T1 is cleaned up, T2 will see the vertex as an existing one
EXPECT_THAT(GetIds(acc2.Vertices(View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc2.Vertices(View::NEW), View::NEW), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc2.Vertices(label1, prop_val, View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc2.Vertices(label1, prop_val, View::NEW), View::NEW), UnorderedElementsAre(0));
check_t3();
EXPECT_THAT(GetIds(acc4.Vertices(View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc4.Vertices(View::NEW), View::NEW), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc4.Vertices(label1, prop_val, View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc4.Vertices(label1, prop_val, View::NEW), View::NEW), UnorderedElementsAre(0));
check_t5();
CleanupHlc(t3_start);
// As T3 got cleaned up, it will delete the vertex from the actual storage and the index
EXPECT_THAT(GetIds(acc4.Vertices(View::OLD), View::OLD), IsEmpty());
EXPECT_THAT(GetIds(acc4.Vertices(View::NEW), View::NEW), IsEmpty());
EXPECT_THAT(GetIds(acc4.Vertices(label1, prop_val, View::OLD), View::OLD), IsEmpty());
EXPECT_THAT(GetIds(acc4.Vertices(label1, prop_val, View::NEW), View::NEW), IsEmpty());
check_t5();
}
TEST_F(IndexTest, CollectGarbageRemoveLabel) {
// First part
// T1 (start 1, commit 3) Creates the vertex, adds label and property
// T2 (start 2, no commit) reads nothing
// T3 (start 4, commit 6) removes label
// T4 (start 5, no commit) reads the vertex
// T5 (start 7, no commit) reads nothing
auto t1_start = GetNextHlc();
auto t2_start = GetNextHlc();
auto t1_commit = GetNextHlc();
auto t3_start = GetNextHlc();
auto t4_start = GetNextHlc();
auto t3_commit = GetNextHlc();
auto t5_start = GetNextHlc();
ASSERT_TRUE(storage.CreateIndex(label1, prop_val));
auto acc1 = storage.Access(t1_start);
{
auto vertex = CreateVertex(acc1);
ASSERT_NO_ERROR(vertex.AddLabelAndValidate(label1));
ASSERT_NO_ERROR(vertex.SetPropertyAndValidate(prop_val, PropertyValue(42)));
}
acc1.Commit(t1_commit);
EXPECT_THAT(GetIds(acc1.Vertices(label1, prop_val, View::OLD), View::OLD), IsEmpty());
EXPECT_THAT(GetIds(acc1.Vertices(label1, prop_val, View::NEW), View::NEW), UnorderedElementsAre(0));
auto acc2 = storage.Access(t2_start);
EXPECT_THAT(GetIds(acc2.Vertices(View::OLD), View::OLD), IsEmpty());
EXPECT_THAT(GetIds(acc2.Vertices(View::NEW), View::NEW), IsEmpty());
EXPECT_THAT(GetIds(acc2.Vertices(label1, prop_val, View::OLD), View::OLD), IsEmpty());
EXPECT_THAT(GetIds(acc2.Vertices(label1, prop_val, View::NEW), View::NEW), IsEmpty());
auto acc3 = storage.Access(t3_start);
{
auto vertices = acc3.Vertices(label1, prop_val, View::OLD);
EXPECT_THAT(GetIds(acc3.Vertices(label1, prop_val, View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc3.Vertices(label1, prop_val, View::NEW), View::NEW), UnorderedElementsAre(0));
auto vertex = *vertices.begin();
ASSERT_NO_ERROR(vertex.RemoveLabelAndValidate(label1));
EXPECT_THAT(GetIds(acc3.Vertices(label1, prop_val, View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc3.Vertices(label1, prop_val, View::NEW), View::NEW), IsEmpty());
}
// We clean up T1 here to make sure the garbace collection doesn't remove the vertex from the indices until the vertex
// has a version that is reachable from an active transaction, even though the latest version of the vertex doesn't
// belong to the index
CleanupHlc(t1_start);
acc3.Commit(t3_commit);
auto check_t3 = [this, &acc3]() {
EXPECT_THAT(GetIds(acc3.Vertices(label1, prop_val, View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc3.Vertices(label1, prop_val, View::NEW), View::NEW), IsEmpty());
};
check_t3();
auto acc4 = storage.Access(t4_start);
EXPECT_THAT(GetIds(acc4.Vertices(View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc4.Vertices(View::NEW), View::NEW), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc4.Vertices(label1, prop_val, View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc4.Vertices(label1, prop_val, View::NEW), View::NEW), UnorderedElementsAre(0));
auto acc5 = storage.Access(t5_start);
auto check_t5 = [this, &acc5]() {
EXPECT_THAT(GetIds(acc5.Vertices(View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc5.Vertices(View::NEW), View::NEW), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc5.Vertices(label1, prop_val, View::OLD), View::OLD), IsEmpty());
EXPECT_THAT(GetIds(acc5.Vertices(label1, prop_val, View::NEW), View::NEW), IsEmpty());
};
check_t5();
// As the deltas of T1 is cleaned up, T2 will see the vertex as an existing one
EXPECT_THAT(GetIds(acc2.Vertices(View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc2.Vertices(View::NEW), View::NEW), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc2.Vertices(label1, prop_val, View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc2.Vertices(label1, prop_val, View::NEW), View::NEW), UnorderedElementsAre(0));
check_t3();
EXPECT_THAT(GetIds(acc4.Vertices(View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc4.Vertices(View::NEW), View::NEW), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc4.Vertices(label1, prop_val, View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc4.Vertices(label1, prop_val, View::NEW), View::NEW), UnorderedElementsAre(0));
check_t5();
// Second part
// Clean up T3 and check the changes
CleanupHlc(t3_start);
// As T3 got cleaned up, it will delete the vertex from the index but not from the actual storage
EXPECT_THAT(GetIds(acc4.Vertices(View::OLD), View::OLD), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc4.Vertices(View::NEW), View::NEW), UnorderedElementsAre(0));
EXPECT_THAT(GetIds(acc4.Vertices(label1, prop_val, View::OLD), View::OLD), IsEmpty());
EXPECT_THAT(GetIds(acc4.Vertices(label1, prop_val, View::NEW), View::NEW), IsEmpty());
check_t5();
}
} // namespace memgraph::storage::v3::tests

View File

@ -0,0 +1,142 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include <gtest/gtest.h>
#include "storage/v3/isolation_level.hpp"
#include "storage/v3/property_value.hpp"
#include "storage/v3/storage.hpp"
namespace memgraph::storage::v3::tests {
int64_t VerticesCount(Shard::Accessor &accessor) {
int64_t count{0};
for ([[maybe_unused]] const auto &vertex : accessor.Vertices(View::NEW)) {
++count;
}
return count;
}
inline constexpr std::array isolation_levels{IsolationLevel::SNAPSHOT_ISOLATION, IsolationLevel::READ_COMMITTED,
IsolationLevel::READ_UNCOMMITTED};
std::string_view IsolationLevelToString(const IsolationLevel isolation_level) {
switch (isolation_level) {
case IsolationLevel::SNAPSHOT_ISOLATION:
return "SNAPSHOT_ISOLATION";
case IsolationLevel::READ_COMMITTED:
return "READ_COMMITTED";
case IsolationLevel::READ_UNCOMMITTED:
return "READ_UNCOMMITTED";
}
}
class StorageIsolationLevelTest : public ::testing::TestWithParam<IsolationLevel> {
protected:
[[nodiscard]] LabelId NameToLabelId(std::string_view label_name) {
return LabelId::FromUint(id_mapper.NameToId(label_name));
}
[[nodiscard]] PropertyId NameToPropertyId(std::string_view property_name) {
return PropertyId::FromUint(id_mapper.NameToId(property_name));
}
[[nodiscard]] EdgeTypeId NameToEdgeTypeId(std::string_view edge_type_name) {
return EdgeTypeId::FromUint(id_mapper.NameToId(edge_type_name));
}
[[nodiscard]] coordinator::Hlc GetNextHlc() {
++last_hlc.logical_id;
last_hlc.coordinator_wall_clock += std::chrono::seconds(10);
return last_hlc;
}
NameIdMapper id_mapper;
static constexpr int64_t min_primary_key_value{0};
static constexpr int64_t max_primary_key_value{10000};
const LabelId primary_label{NameToLabelId("label")};
const PropertyId primary_property{NameToPropertyId("property")};
coordinator::Hlc last_hlc{0, io::Time{}};
public:
struct PrintToStringParamName {
std::string operator()(const testing::TestParamInfo<IsolationLevel> &info) {
return std::string(IsolationLevelToString(static_cast<IsolationLevel>(info.param)));
}
};
};
TEST_P(StorageIsolationLevelTest, Visibility) {
const auto default_isolation_level = GetParam();
for (auto override_isolation_level_index{0U}; override_isolation_level_index < isolation_levels.size();
++override_isolation_level_index) {
Shard store{primary_label,
{PropertyValue{min_primary_key_value}},
std::vector{PropertyValue{max_primary_key_value}},
Config{.transaction = {.isolation_level = default_isolation_level}}};
ASSERT_TRUE(
store.CreateSchema(primary_label, {storage::v3::SchemaProperty{primary_property, common::SchemaType::INT}}));
const auto override_isolation_level = isolation_levels[override_isolation_level_index];
auto creator = store.Access(GetNextHlc());
auto default_isolation_level_reader = store.Access(GetNextHlc());
auto override_isolation_level_reader = store.Access(GetNextHlc(), override_isolation_level);
ASSERT_EQ(VerticesCount(default_isolation_level_reader), 0);
ASSERT_EQ(VerticesCount(override_isolation_level_reader), 0);
static constexpr auto iteration_count = 10;
{
SCOPED_TRACE(fmt::format(
"Visibility while the creator transaction is active "
"(default isolation level = {}, override isolation level = {})",
IsolationLevelToString(default_isolation_level), IsolationLevelToString(override_isolation_level)));
for (auto i{1}; i <= iteration_count; ++i) {
ASSERT_TRUE(
creator.CreateVertexAndValidate(primary_label, {}, {{primary_property, PropertyValue{i}}}).HasValue());
const auto check_vertices_count = [i](auto &accessor, const auto isolation_level) {
const auto expected_count = isolation_level == IsolationLevel::READ_UNCOMMITTED ? i : 0;
EXPECT_EQ(VerticesCount(accessor), expected_count);
};
check_vertices_count(default_isolation_level_reader, default_isolation_level);
check_vertices_count(override_isolation_level_reader, override_isolation_level);
}
}
creator.Commit(GetNextHlc());
{
SCOPED_TRACE(fmt::format(
"Visibility after the creator transaction is committed "
"(default isolation level = {}, override isolation level = {})",
IsolationLevelToString(default_isolation_level), IsolationLevelToString(override_isolation_level)));
const auto check_vertices_count = [](auto &accessor, const auto isolation_level) {
const auto expected_count = isolation_level == IsolationLevel::SNAPSHOT_ISOLATION ? 0 : iteration_count;
ASSERT_EQ(VerticesCount(accessor), expected_count);
};
check_vertices_count(default_isolation_level_reader, default_isolation_level);
check_vertices_count(override_isolation_level_reader, override_isolation_level);
}
default_isolation_level_reader.Commit(GetNextHlc());
override_isolation_level_reader.Commit(GetNextHlc());
SCOPED_TRACE("Visibility after a new transaction is started");
auto verifier = store.Access(GetNextHlc());
ASSERT_EQ(VerticesCount(verifier), iteration_count);
verifier.Commit(GetNextHlc());
}
}
INSTANTIATE_TEST_CASE_P(ParameterizedStorageIsolationLevelTests, StorageIsolationLevelTest,
::testing::ValuesIn(isolation_levels), StorageIsolationLevelTest::PrintToStringParamName());
} // namespace memgraph::storage::v3::tests

View File

@ -48,15 +48,22 @@ class StorageV3Accessor : public ::testing::Test {
PropertyId NameToPropertyId(std::string_view property_name) { return storage.NameToProperty(property_name); }
coordinator::Hlc GetNextHlc() {
++last_hlc.logical_id;
last_hlc.coordinator_wall_clock += std::chrono::seconds(10);
return last_hlc;
}
const std::vector<PropertyValue> pk{PropertyValue{0}};
const LabelId primary_label{LabelId::FromUint(1)};
const PropertyId primary_property{PropertyId::FromUint(2)};
Shard storage{primary_label, pk, std::nullopt};
coordinator::Hlc last_hlc{0, io::Time{}};
};
TEST_F(StorageV3Accessor, TestPrimaryLabel) {
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
const auto vertex = CreateVertexAndValidate(acc, primary_label, {}, {{primary_property, PropertyValue(0)}});
ASSERT_TRUE(vertex.PrimaryLabel(View::NEW).HasValue());
const auto vertex_primary_label = vertex.PrimaryLabel(View::NEW).GetValue();
@ -64,7 +71,7 @@ TEST_F(StorageV3Accessor, TestPrimaryLabel) {
EXPECT_EQ(vertex_primary_label, primary_label);
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
const auto vertex = CreateVertexAndValidate(acc, primary_label, {}, {{primary_property, PropertyValue(1)}});
ASSERT_TRUE(vertex.PrimaryLabel(View::OLD).HasError());
const auto error_primary_label = vertex.PrimaryLabel(View::OLD).GetError();
@ -72,12 +79,12 @@ TEST_F(StorageV3Accessor, TestPrimaryLabel) {
EXPECT_EQ(error_primary_label, Error::NONEXISTENT_OBJECT);
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
CreateVertexAndValidate(acc, primary_label, {}, {{primary_property, PropertyValue(2)}});
ASSERT_FALSE(acc.Commit().HasError());
acc.Commit(GetNextHlc());
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
const auto vertex = acc.FindVertex({PropertyValue{2}}, View::OLD);
ASSERT_TRUE(vertex.has_value());
ASSERT_TRUE(acc.FindVertex({PropertyValue{2}}, View::NEW).has_value());
@ -91,7 +98,7 @@ TEST_F(StorageV3Accessor, TestPrimaryLabel) {
TEST_F(StorageV3Accessor, TestAddLabels) {
storage.StoreMapping({{1, "label"}, {2, "property"}, {3, "label1"}, {4, "label2"}, {5, "label3"}});
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
const auto label1 = NameToLabelId("label1");
const auto label2 = NameToLabelId("label2");
const auto label3 = NameToLabelId("label3");
@ -102,7 +109,7 @@ TEST_F(StorageV3Accessor, TestAddLabels) {
EXPECT_THAT(vertex.Labels(View::NEW).GetValue(), UnorderedElementsAre(label1, label2, label3));
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
const auto label1 = NameToLabelId("label1");
const auto label2 = NameToLabelId("label2");
const auto label3 = NameToLabelId("label3");
@ -117,7 +124,7 @@ TEST_F(StorageV3Accessor, TestAddLabels) {
EXPECT_THAT(vertex.Labels(View::NEW).GetValue(), UnorderedElementsAre(label1, label2, label3));
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
const auto label1 = NameToLabelId("label");
auto vertex = acc.CreateVertexAndValidate(primary_label, {label1}, {{primary_property, PropertyValue(2)}});
ASSERT_TRUE(vertex.HasError());
@ -126,7 +133,7 @@ TEST_F(StorageV3Accessor, TestAddLabels) {
SchemaViolation(SchemaViolation::ValidationStatus::VERTEX_SECONDARY_LABEL_IS_PRIMARY, label1));
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
const auto label1 = NameToLabelId("label");
auto vertex = acc.CreateVertexAndValidate(primary_label, {}, {{primary_property, PropertyValue(3)}});
ASSERT_TRUE(vertex.HasValue());
@ -142,7 +149,7 @@ TEST_F(StorageV3Accessor, TestRemoveLabels) {
storage.StoreMapping({{1, "label"}, {2, "property"}, {3, "label1"}, {4, "label2"}, {5, "label3"}});
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
const auto label1 = NameToLabelId("label1");
const auto label2 = NameToLabelId("label2");
const auto label3 = NameToLabelId("label3");
@ -165,7 +172,7 @@ TEST_F(StorageV3Accessor, TestRemoveLabels) {
EXPECT_TRUE(vertex.Labels(View::NEW).GetValue().empty());
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
const auto label1 = NameToLabelId("label1");
auto vertex = CreateVertexAndValidate(acc, primary_label, {}, {{primary_property, PropertyValue(1)}});
ASSERT_TRUE(vertex.Labels(View::NEW).HasValue());
@ -175,7 +182,7 @@ TEST_F(StorageV3Accessor, TestRemoveLabels) {
EXPECT_FALSE(res1.GetValue());
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
auto vertex = CreateVertexAndValidate(acc, primary_label, {}, {{primary_property, PropertyValue(2)}});
const auto res1 = vertex.RemoveLabelAndValidate(primary_label);
ASSERT_TRUE(res1.HasError());
@ -189,14 +196,14 @@ TEST_F(StorageV3Accessor, TestSetKeysAndProperties) {
storage.StoreMapping({{1, "label"}, {2, "property"}, {3, "prop1"}});
storage.StoreMapping({{1, "label"}, {2, "property"}, {3, "prop1"}});
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
const PropertyId prop1{NameToPropertyId("prop1")};
auto vertex = CreateVertexAndValidate(acc, primary_label, {}, {{primary_property, PropertyValue(0)}});
const auto res = vertex.SetPropertyAndValidate(prop1, PropertyValue(1));
ASSERT_TRUE(res.HasValue());
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
auto vertex = CreateVertexAndValidate(acc, primary_label, {}, {{primary_property, PropertyValue(1)}});
const auto res = vertex.SetPropertyAndValidate(primary_property, PropertyValue(1));
ASSERT_TRUE(res.HasError());
@ -206,7 +213,7 @@ TEST_F(StorageV3Accessor, TestSetKeysAndProperties) {
SchemaProperty{primary_property, common::SchemaType::INT}));
}
{
auto acc = storage.Access();
auto acc = storage.Access(GetNextHlc());
auto vertex = CreateVertexAndValidate(acc, primary_label, {}, {{primary_property, PropertyValue(2)}});
const auto res = vertex.SetPropertyAndValidate(primary_property, PropertyValue());
ASSERT_TRUE(res.HasError());