tursom/memgraph
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge branch 'master' into fix-valuetype-function-on-all-data-types

Browse Source
This commit is contained in:
Josipmrden 2024-03-11 11:24:29 +01:00 committed by GitHub
commit 3c4faa8663
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
203 changed files with 11121 additions and 1857 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

60
.github/workflows/release_build_test.yaml vendored
View File

@ -56,6 +56,15 @@ jobs:
- name: "Build package"
run: |
./release/package/run.sh package debian-10 $BUILD_TYPE
- name: Upload to S3
uses: jakejarvis/s3-sync-action@v0.5.1
env:
AWS_S3_BUCKET: "deps.memgraph.io"
AWS_ACCESS_KEY_ID: ${{ secrets.S3_AWS_ACCESS_KEY_ID }}
AWS_SECRET_ACCESS_KEY: ${{ secrets.S3_AWS_SECRET_ACCESS_KEY }}
AWS_REGION: "eu-west-1"
SOURCE_DIR: "build/output"
DEST_DIR: "memgraph-unofficial/${{ github.ref_name }}/"
- name: "Upload package"
uses: actions/upload-artifact@v4
with:
@ -75,6 +84,15 @@ jobs:
- name: "Build package"
run: |
./release/package/run.sh package ubuntu-22.04 $BUILD_TYPE
- name: Upload to S3
uses: jakejarvis/s3-sync-action@v0.5.1
env:
AWS_S3_BUCKET: "deps.memgraph.io"
AWS_ACCESS_KEY_ID: ${{ secrets.S3_AWS_ACCESS_KEY_ID }}
AWS_SECRET_ACCESS_KEY: ${{ secrets.S3_AWS_SECRET_ACCESS_KEY }}
AWS_REGION: "eu-west-1"
SOURCE_DIR: "build/output"
DEST_DIR: "memgraph-unofficial/${{ github.ref_name }}/"
- name: "Upload package"
uses: actions/upload-artifact@v4
with:
@ -86,7 +104,7 @@ jobs:
needs: [Ubuntu20_04]
runs-on: [self-hosted, DockerMgBuild, ARM64]
# M1 Mac mini is sometimes slower
timeout-minutes: 90
timeout-minutes: 150
steps:
- name: "Set up repository"
uses: actions/checkout@v4
@ -101,6 +119,26 @@ jobs:
name: ubuntu-22.04-aarch64
path: build/output/ubuntu-22.04-arm/memgraph*.deb
PushToS3Ubuntu20_04_ARM:
if: github.ref_type == 'tag'
needs: [PackageUbuntu20_04_ARM]
runs-on: ubuntu-latest
steps:
- name: Download package
uses: actions/download-artifact@v4
with:
name: ubuntu-22.04-aarch64
path: build/output/release
- name: Upload to S3
uses: jakejarvis/s3-sync-action@v0.5.1
env:
AWS_S3_BUCKET: "deps.memgraph.io"
AWS_ACCESS_KEY_ID: ${{ secrets.S3_AWS_ACCESS_KEY_ID }}
AWS_SECRET_ACCESS_KEY: ${{ secrets.S3_AWS_SECRET_ACCESS_KEY }}
AWS_REGION: "eu-west-1"
SOURCE_DIR: "build/output/release"
DEST_DIR: "memgraph-unofficial/${{ github.ref_name }}/"
PackageDebian11:
if: github.ref_type == 'tag'
needs: [Debian10, Ubuntu20_04]
@ -114,6 +152,15 @@ jobs:
- name: "Build package"
run: |
./release/package/run.sh package debian-11 $BUILD_TYPE
- name: Upload to S3
uses: jakejarvis/s3-sync-action@v0.5.1
env:
AWS_S3_BUCKET: "deps.memgraph.io"
AWS_ACCESS_KEY_ID: ${{ secrets.S3_AWS_ACCESS_KEY_ID }}
AWS_SECRET_ACCESS_KEY: ${{ secrets.S3_AWS_SECRET_ACCESS_KEY }}
AWS_REGION: "eu-west-1"
SOURCE_DIR: "build/output"
DEST_DIR: "memgraph-unofficial/${{ github.ref_name }}/"
- name: "Upload package"
uses: actions/upload-artifact@v4
with:
@ -125,7 +172,7 @@ jobs:
needs: [Debian10, Ubuntu20_04]
runs-on: [self-hosted, DockerMgBuild, ARM64]
# M1 Mac mini is sometimes slower
timeout-minutes: 90
timeout-minutes: 150
steps:
- name: "Set up repository"
uses: actions/checkout@v4
@ -140,16 +187,15 @@ jobs:
name: debian-11-aarch64
path: build/output/debian-11-arm/memgraph*.deb
PushToS3:
PushToS3Debian11_ARM:
if: github.ref_type == 'tag'
needs: [PackageDebian10, PackageDebian11, PackageDebian11_ARM, PackageUbuntu20_04, PackageUbuntu20_04_ARM]
needs: [PackageDebian11_ARM]
runs-on: ubuntu-latest
steps:
- name: Download artifacts
- name: Download package
uses: actions/download-artifact@v4
with:
# name: # if name input parameter is not provided, all artifacts are downloaded
# and put in directories named after each one.
name: debian-11-aarch64
path: build/output/release
- name: Upload to S3
uses: jakejarvis/s3-sync-action@v0.5.1

1
environment/os/amzn-2.sh
View File

@ -45,6 +45,7 @@ MEMGRAPH_BUILD_DEPS=(
readline-devel # for memgraph console
python3-devel # for query modules
openssl-devel
openssl
libseccomp-devel
python3 python3-pip nmap-ncat # for tests
#

1
environment/os/centos-7.sh
View File

@ -43,6 +43,7 @@ MEMGRAPH_BUILD_DEPS=(
readline-devel # for memgraph console
python3-devel # for query modules
openssl-devel
openssl
libseccomp-devel
python3 python-virtualenv python3-pip nmap-ncat # for qa, macro_benchmark and stress tests
#

8
import/n2mg_cypherl.sh
View File

@ -20,14 +20,18 @@ if [ ! -f "$INPUT" ]; then
fi
echo -e "${COLOR_ORANGE}NOTE:${COLOR_NULL} BEGIN and COMMIT are required because variables share the same name (e.g. row)"
echo -e "${COLOR_ORANGE}NOTE:${COLOR_NULL} CONSTRAINTS are just skipped -> ${COLOR_RED}please create consraints manually if needed${COLOR_NULL}"
echo -e "${COLOR_ORANGE}NOTE:${COLOR_NULL} CONSTRAINTS are just skipped -> ${COLOR_RED}please create constraints manually if needed${COLOR_NULL}"
echo 'CREATE INDEX ON :`UNIQUE IMPORT LABEL`(`UNIQUE IMPORT ID`);' > "$OUTPUT"
sed -e 's/^:begin/BEGIN/g; s/^BEGIN$/BEGIN;/g;' \
-e 's/^:commit/COMMIT/g; s/^COMMIT$/COMMIT;/g;' \
-e '/^CALL/d; /^SCHEMA AWAIT/d;' \
-e 's/CREATE RANGE INDEX FOR (n:/CREATE INDEX ON :/g;' \
-e 's/) ON (n./(/g;' \
-e '/^CREATE CONSTRAINT/d; /^DROP CONSTRAINT/d;' "$INPUT" > "$OUTPUT"
-e '/^CREATE CONSTRAINT/d; /^DROP CONSTRAINT/d;' "$INPUT" >> "$OUTPUT"
echo 'DROP INDEX ON :`UNIQUE IMPORT LABEL`(`UNIQUE IMPORT ID`);' >> "$OUTPUT"
echo ""
echo -e "${COLOR_GREEN}DONE!${COLOR_NULL} Please find Memgraph compatible cypherl|.cypher file under $OUTPUT"

61
import/n2mg_separate_files_cypherl.sh Executable file
View File

@ -0,0 +1,61 @@
#!/bin/bash -e
COLOR_ORANGE="\e[38;5;208m"
COLOR_GREEN="\e[38;5;35m"
COLOR_RED="\e[0;31m"
COLOR_NULL="\e[0m"
print_help() {
echo -e "${COLOR_ORANGE}HOW TO RUN:${COLOR_NULL} $0 input_file_schema_path input_file_nodes_path input_file_relationships_path input_file_cleanup_path output_file_path"
exit 1
}
if [ "$#" -ne 5 ]; then
print_help
fi
INPUT_SCHEMA="$1"
INPUT_NODES="$2"
INPUT_RELATIONSHIPS="$3"
INPUT_CLEANUP="$4"
OUTPUT="$5"
if [ ! -f "$INPUT_SCHEMA" ]; then
echo -e "${COLOR_RED}ERROR:${COLOR_NULL} input_file_path is not a file!"
print_help
fi
if [ ! -f "$INPUT_NODES" ]; then
echo -e "${COLOR_RED}ERROR:${COLOR_NULL} input_file_path is not a file!"
print_help
fi
if [ ! -f "$INPUT_RELATIONSHIPS" ]; then
echo -e "${COLOR_RED}ERROR:${COLOR_NULL} input_file_path is not a file!"
print_help
fi
if [ ! -f "$INPUT_CLEANUP" ]; then
echo -e "${COLOR_RED}ERROR:${COLOR_NULL} input_file_path is not a file!"
print_help
fi
echo -e "${COLOR_ORANGE}NOTE:${COLOR_NULL} BEGIN and COMMIT are required because variables share the same name (e.g. row)"
echo -e "${COLOR_ORANGE}NOTE:${COLOR_NULL} CONSTRAINTS are just skipped -> ${COLOR_RED}please create constraints manually if needed${COLOR_NULL}"
echo 'CREATE INDEX ON :`UNIQUE IMPORT LABEL`(`UNIQUE IMPORT ID`);' > "$OUTPUT"
sed -e 's/CREATE RANGE INDEX FOR (n:/CREATE INDEX ON :/g;' \
-e 's/) ON (n./(/g;' \
-e '/^CREATE CONSTRAINT/d' $INPUT_SCHEMA >> "$OUTPUT"
cat "$INPUT_NODES" >> "$OUTPUT"
cat "$INPUT_RELATIONSHIPS" >> "$OUTPUT"
sed -e '/^DROP CONSTRAINT/d' "$INPUT_CLEANUP" >> "$OUTPUT"
echo 'DROP INDEX ON :`UNIQUE IMPORT LABEL`(`UNIQUE IMPORT ID`);' >> "$OUTPUT"
echo ""
echo -e "${COLOR_GREEN}DONE!${COLOR_NULL} Please find Memgraph compatible cypherl|.cypher file under $OUTPUT"
echo ""
echo "Please import data by executing => \`cat $OUTPUT | mgconsole\`"

64
import/n2mg_separate_files_cypherls.sh Executable file
View File

@ -0,0 +1,64 @@
#!/bin/bash -e
COLOR_ORANGE="\e[38;5;208m"
COLOR_GREEN="\e[38;5;35m"
COLOR_RED="\e[0;31m"
COLOR_NULL="\e[0m"
print_help() {
echo -e "${COLOR_ORANGE}HOW TO RUN:${COLOR_NULL} $0 input_file_schema_path input_file_nodes_path input_file_relationships_path input_file_cleanup_path output_file_schema_path output_file_nodes_path output_file_relationships_path output_file_cleanup_path"
exit 1
}
if [ "$#" -ne 8 ]; then
print_help
fi
INPUT_SCHEMA="$1"
INPUT_NODES="$2"
INPUT_RELATIONSHIPS="$3"
INPUT_CLEANUP="$4"
OUTPUT_SCHEMA="$5"
OUTPUT_NODES="$6"
OUTPUT_RELATIONSHIPS="$7"
OUTPUT_CLEANUP="$8"
if [ ! -f "$INPUT_SCHEMA" ]; then
echo -e "${COLOR_RED}ERROR:${COLOR_NULL} input_file_path is not a file!"
print_help
fi
if [ ! -f "$INPUT_NODES" ]; then
echo -e "${COLOR_RED}ERROR:${COLOR_NULL} input_file_path is not a file!"
print_help
fi
if [ ! -f "$INPUT_RELATIONSHIPS" ]; then
echo -e "${COLOR_RED}ERROR:${COLOR_NULL} input_file_path is not a file!"
print_help
fi
if [ ! -f "$INPUT_CLEANUP" ]; then
echo -e "${COLOR_RED}ERROR:${COLOR_NULL} input_file_path is not a file!"
print_help
fi
echo -e "${COLOR_ORANGE}NOTE:${COLOR_NULL} BEGIN and COMMIT are required because variables share the same name (e.g. row)"
echo -e "${COLOR_ORANGE}NOTE:${COLOR_NULL} CONSTRAINTS are just skipped -> ${COLOR_RED}please create constraints manually if needed${COLOR_NULL}"
echo 'CREATE INDEX ON :`UNIQUE IMPORT LABEL`(`UNIQUE IMPORT ID`);' > "$OUTPUT_SCHEMA"
sed -e 's/CREATE RANGE INDEX FOR (n:/CREATE INDEX ON :/g;' \
-e 's/) ON (n./(/g;' \
-e '/^CREATE CONSTRAINT/d' $INPUT_SCHEMA >> "$OUTPUT_SCHEMA"
cat "$INPUT_NODES" > "$OUTPUT_NODES"
cat "$INPUT_RELATIONSHIPS" > "$OUTPUT_RELATIONSHIPS"
sed -e '/^DROP CONSTRAINT/d' "$INPUT_CLEANUP" >> "$OUTPUT_CLEANUP"
echo 'DROP INDEX ON :`UNIQUE IMPORT LABEL`(`UNIQUE IMPORT ID`);' >> "$OUTPUT_CLEANUP"
echo ""
echo -e "${COLOR_GREEN}DONE!${COLOR_NULL} Please find Memgraph compatible cypherl|.cypher files under $OUTPUT_SCHEMA, $OUTPUT_NODES, $OUTPUT_RELATIONSHIPS and $OUTPUT_CLEANUP"
echo ""
echo "Please import data by executing => \`cat $OUTPUT_SCHEMA | mgconsole\`, \`cat $OUTPUT_NODES | mgconsole\`, \`cat $OUTPUT_RELATIONSHIPS | mgconsole\` and \`cat $OUTPUT_CLEANUP | mgconsole\`"

125
query_modules/schema.cpp
View File

@ -1,4 +1,4 @@
// Copyright 2023 Memgraph Ltd.
// Copyright 2024 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
@ -9,10 +9,11 @@
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include <boost/functional/hash.hpp>
#include <mgp.hpp>
#include "utils/string.hpp"
#include <optional>
#include <unordered_set>
namespace Schema {
@ -37,6 +38,7 @@ constexpr std::string_view kParameterIndices = "indices";
constexpr std::string_view kParameterUniqueConstraints = "unique_constraints";
constexpr std::string_view kParameterExistenceConstraints = "existence_constraints";
constexpr std::string_view kParameterDropExisting = "drop_existing";
constexpr int kInitialNumberOfPropertyOccurances = 1;
std::string TypeOf(const mgp::Type &type);
@ -108,83 +110,79 @@ void Schema::ProcessPropertiesRel(mgp::Record &record, const std::string_view &t
record.Insert(std::string(kReturnMandatory).c_str(), mandatory);
}
struct Property {
std::string name;
mgp::Value value;
struct PropertyInfo {
std::unordered_set<std::string> property_types; // property types
int64_t number_of_property_occurrences = 0;
Property(const std::string &name, mgp::Value &&value) : name(name), value(std::move(value)) {}
PropertyInfo() = default;
explicit PropertyInfo(std::string &&property_type)
: property_types({std::move(property_type)}),
number_of_property_occurrences(Schema::kInitialNumberOfPropertyOccurances) {}
};
struct LabelsInfo {
std::unordered_map<std::string, PropertyInfo> properties; // key is a property name
int64_t number_of_label_occurrences = 0;
};
struct LabelsHash {
std::size_t operator()(const std::set<std::string> &set) const {
std::size_t seed = set.size();
for (const auto &i : set) {
seed ^= std::hash<std::string>{}(i) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
}
return seed;
}
std::size_t operator()(const std::set<std::string> &s) const { return boost::hash_range(s.begin(), s.end()); }
};
struct LabelsComparator {
bool operator()(const std::set<std::string> &lhs, const std::set<std::string> &rhs) const { return lhs == rhs; }
};
struct PropertyComparator {
bool operator()(const Property &lhs, const Property &rhs) const { return lhs.name < rhs.name; }
};
struct PropertyInfo {
std::set<Property, PropertyComparator> properties;
bool mandatory;
};
void Schema::NodeTypeProperties(mgp_list * /*args*/, mgp_graph *memgraph_graph, mgp_result *result,
mgp_memory *memory) {
mgp::MemoryDispatcherGuard guard{memory};
const auto record_factory = mgp::RecordFactory(result);
try {
std::unordered_map<std::set<std::string>, PropertyInfo, LabelsHash, LabelsComparator> node_types_properties;
std::unordered_map<std::set<std::string>, LabelsInfo, LabelsHash, LabelsComparator> node_types_properties;
for (auto node : mgp::Graph(memgraph_graph).Nodes()) {
for (const auto node : mgp::Graph(memgraph_graph).Nodes()) {
std::set<std::string> labels_set = {};
for (auto label : node.Labels()) {
for (const auto label : node.Labels()) {
labels_set.emplace(label);
}
if (node_types_properties.find(labels_set) == node_types_properties.end()) {
node_types_properties[labels_set] = PropertyInfo{std::set<Property, PropertyComparator>(), true};
}
node_types_properties[labels_set].number_of_label_occurrences++;
if (node.Properties().empty()) {
node_types_properties[labels_set].mandatory = false; // if there is node with no property, it is not mandatory
continue;
}
auto &property_info = node_types_properties.at(labels_set);
for (auto &[key, prop] : node.Properties()) {
property_info.properties.emplace(key, std::move(prop));
if (property_info.mandatory) {
property_info.mandatory =
property_info.properties.size() == 1; // if there is only one property, it is mandatory
auto &labels_info = node_types_properties.at(labels_set);
for (const auto &[key, prop] : node.Properties()) {
auto prop_type = TypeOf(prop.Type());
if (labels_info.properties.find(key) == labels_info.properties.end()) {
labels_info.properties[key] = PropertyInfo{std::move(prop_type)};
} else {
labels_info.properties[key].property_types.emplace(prop_type);
labels_info.properties[key].number_of_property_occurrences++;
}
}
}
for (auto &[labels, property_info] : node_types_properties) {
for (auto &[node_type, labels_info] : node_types_properties) { // node type is a set of labels
std::string label_type;
mgp::List labels_list = mgp::List();
for (auto const &label : labels) {
auto labels_list = mgp::List();
for (const auto &label : node_type) {
label_type += ":`" + std::string(label) + "`";
labels_list.AppendExtend(mgp::Value(label));
}
for (auto const &prop : property_info.properties) {
for (const auto &prop : labels_info.properties) {
auto prop_types = mgp::List();
for (const auto &prop_type : prop.second.property_types) {
prop_types.AppendExtend(mgp::Value(prop_type));
}
bool mandatory = prop.second.number_of_property_occurrences == labels_info.number_of_label_occurrences;
auto record = record_factory.NewRecord();
ProcessPropertiesNode(record, label_type, labels_list, prop.name, TypeOf(prop.value.Type()),
property_info.mandatory);
ProcessPropertiesNode(record, label_type, labels_list, prop.first, prop_types, mandatory);
}
if (property_info.properties.empty()) {
if (labels_info.properties.empty()) {
auto record = record_factory.NewRecord();
ProcessPropertiesNode<std::string>(record, label_type, labels_list, "", "", false);
ProcessPropertiesNode<mgp::List>(record, label_type, labels_list, "", mgp::List(), false);
}
}
@ -197,40 +195,45 @@ void Schema::NodeTypeProperties(mgp_list * /*args*/, mgp_graph *memgraph_graph,
void Schema::RelTypeProperties(mgp_list * /*args*/, mgp_graph *memgraph_graph, mgp_result *result, mgp_memory *memory) {
mgp::MemoryDispatcherGuard guard{memory};
std::unordered_map<std::string, PropertyInfo> rel_types_properties;
std::unordered_map<std::string, LabelsInfo> rel_types_properties;
const auto record_factory = mgp::RecordFactory(result);
try {
const mgp::Graph graph = mgp::Graph(memgraph_graph);
for (auto rel : graph.Relationships()) {
const auto graph = mgp::Graph(memgraph_graph);
for (const auto rel : graph.Relationships()) {
std::string rel_type = std::string(rel.Type());
if (rel_types_properties.find(rel_type) == rel_types_properties.end()) {
rel_types_properties[rel_type] = PropertyInfo{std::set<Property, PropertyComparator>(), true};
}
rel_types_properties[rel_type].number_of_label_occurrences++;
if (rel.Properties().empty()) {
rel_types_properties[rel_type].mandatory = false; // if there is rel with no property, it is not mandatory
continue;
}
auto &property_info = rel_types_properties.at(rel_type);
auto &labels_info = rel_types_properties.at(rel_type);
for (auto &[key, prop] : rel.Properties()) {
property_info.properties.emplace(key, std::move(prop));
if (property_info.mandatory) {
property_info.mandatory =
property_info.properties.size() == 1; // if there is only one property, it is mandatory
auto prop_type = TypeOf(prop.Type());
if (labels_info.properties.find(key) == labels_info.properties.end()) {
labels_info.properties[key] = PropertyInfo{std::move(prop_type)};
} else {
labels_info.properties[key].property_types.emplace(prop_type);
labels_info.properties[key].number_of_property_occurrences++;
}
}
}
for (auto &[type, property_info] : rel_types_properties) {
std::string type_str = ":`" + std::string(type) + "`";
for (auto const &prop : property_info.properties) {
for (auto &[rel_type, labels_info] : rel_types_properties) {
std::string type_str = ":`" + std::string(rel_type) + "`";
for (const auto &prop : labels_info.properties) {
auto prop_types = mgp::List();
for (const auto &prop_type : prop.second.property_types) {
prop_types.AppendExtend(mgp::Value(prop_type));
}
bool mandatory = prop.second.number_of_property_occurrences == labels_info.number_of_label_occurrences;
auto record = record_factory.NewRecord();
ProcessPropertiesRel(record, type_str, prop.name, TypeOf(prop.value.Type()), property_info.mandatory);
ProcessPropertiesRel(record, type_str, prop.first, prop_types, mandatory);
}
if (property_info.properties.empty()) {
if (labels_info.properties.empty()) {
auto record = record_factory.NewRecord();
ProcessPropertiesRel<std::string>(record, type_str, "", "", false);
ProcessPropertiesRel<mgp::List>(record, type_str, "", mgp::List(), false);
}
}

2
src/communication/result_stream_faker.hpp
View File

@ -1,4 +1,4 @@
// Copyright 2023 Memgraph Ltd.
// Copyright 2024 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

4
src/coordination/CMakeLists.txt
View File

@ -16,11 +16,14 @@ target_sources(mg-coordination
include/coordination/raft_state.hpp
include/coordination/rpc_errors.hpp
include/nuraft/raft_log_action.hpp
include/nuraft/coordinator_cluster_state.hpp
include/nuraft/coordinator_log_store.hpp
include/nuraft/coordinator_state_machine.hpp
include/nuraft/coordinator_state_manager.hpp
PRIVATE
coordinator_config.cpp
coordinator_client.cpp
coordinator_state.cpp
coordinator_rpc.cpp
@ -33,6 +36,7 @@ target_sources(mg-coordination
coordinator_log_store.cpp
coordinator_state_machine.cpp
coordinator_state_manager.cpp
coordinator_cluster_state.cpp
)
target_include_directories(mg-coordination PUBLIC include)

34
src/coordination/coordinator_client.cpp
View File

@ -16,6 +16,7 @@
#include "coordination/coordinator_config.hpp"
#include "coordination/coordinator_rpc.hpp"
#include "replication_coordination_glue/common.hpp"
#include "replication_coordination_glue/messages.hpp"
#include "utils/result.hpp"
@ -30,7 +31,7 @@ auto CreateClientContext(memgraph::coordination::CoordinatorClientConfig const &
} // namespace
CoordinatorClient::CoordinatorClient(CoordinatorInstance *coord_instance, CoordinatorClientConfig config,
HealthCheckCallback succ_cb, HealthCheckCallback fail_cb)
HealthCheckClientCallback succ_cb, HealthCheckClientCallback fail_cb)
: rpc_context_{CreateClientContext(config)},
rpc_client_{io::network::Endpoint(io::network::Endpoint::needs_resolving, config.ip_address, config.port),
&rpc_context_},
@ -40,7 +41,9 @@ CoordinatorClient::CoordinatorClient(CoordinatorInstance *coord_instance, Coordi
fail_cb_{std::move(fail_cb)} {}
auto CoordinatorClient::InstanceName() const -> std::string { return config_.instance_name; }
auto CoordinatorClient::SocketAddress() const -> std::string { return rpc_client_.Endpoint().SocketAddress(); }
auto CoordinatorClient::CoordinatorSocketAddress() const -> std::string { return config_.CoordinatorSocketAddress(); }
auto CoordinatorClient::ReplicationSocketAddress() const -> std::string { return config_.ReplicationSocketAddress(); }
auto CoordinatorClient::InstanceDownTimeoutSec() const -> std::chrono::seconds {
return config_.instance_down_timeout_sec;
@ -63,11 +66,15 @@ void CoordinatorClient::StartFrequentCheck() {
[this, instance_name = config_.instance_name] {
try {
spdlog::trace("Sending frequent heartbeat to machine {} on {}", instance_name,
rpc_client_.Endpoint().SocketAddress());
config_.CoordinatorSocketAddress());
{ // NOTE: This is intentionally scoped so that stream lock could get released.
auto stream{rpc_client_.Stream<memgraph::replication_coordination_glue::FrequentHeartbeatRpc>()};
stream.AwaitResponse();
}
// Subtle race condition:
// acquiring of lock needs to happen before function call, as function callback can be changed
// for instance after lock is already acquired
// (failover case when instance is promoted to MAIN)
succ_cb_(coord_instance_, instance_name);
} catch (rpc::RpcFailedException const &) {
fail_cb_(coord_instance_, instance_name);
@ -79,11 +86,6 @@ void CoordinatorClient::StopFrequentCheck() { instance_checker_.Stop(); }
void CoordinatorClient::PauseFrequentCheck() { instance_checker_.Pause(); }
void CoordinatorClient::ResumeFrequentCheck() { instance_checker_.Resume(); }
auto CoordinatorClient::SetCallbacks(HealthCheckCallback succ_cb, HealthCheckCallback fail_cb) -> void {
succ_cb_ = std::move(succ_cb);
fail_cb_ = std::move(fail_cb);
}
auto CoordinatorClient::ReplicationClientInfo() const -> ReplClientInfo { return config_.replication_client_info; }
auto CoordinatorClient::SendPromoteReplicaToMainRpc(const utils::UUID &uuid,
@ -117,7 +119,7 @@ auto CoordinatorClient::DemoteToReplica() const -> bool {
return false;
}
auto CoordinatorClient::SendSwapMainUUIDRpc(const utils::UUID &uuid) const -> bool {
auto CoordinatorClient::SendSwapMainUUIDRpc(utils::UUID const &uuid) const -> bool {
try {
auto stream{rpc_client_.Stream<replication_coordination_glue::SwapMainUUIDRpc>(uuid)};
if (!stream.AwaitResponse().success) {
@ -131,7 +133,7 @@ auto CoordinatorClient::SendSwapMainUUIDRpc(const utils::UUID &uuid) const -> bo
return false;
}
auto CoordinatorClient::SendUnregisterReplicaRpc(std::string const &instance_name) const -> bool {
auto CoordinatorClient::SendUnregisterReplicaRpc(std::string_view instance_name) const -> bool {
try {
auto stream{rpc_client_.Stream<UnregisterReplicaRpc>(instance_name)};
if (!stream.AwaitResponse().success) {
@ -171,5 +173,17 @@ auto CoordinatorClient::SendEnableWritingOnMainRpc() const -> bool {
return false;
}
auto CoordinatorClient::SendGetInstanceTimestampsRpc() const
-> utils::BasicResult<GetInstanceUUIDError, replication_coordination_glue::DatabaseHistories> {
try {
auto stream{rpc_client_.Stream<coordination::GetDatabaseHistoriesRpc>()};
return stream.AwaitResponse().database_histories;
} catch (const rpc::RpcFailedException &) {
spdlog::error("RPC error occured while sending GetInstance UUID RPC");
return GetInstanceUUIDError::RPC_EXCEPTION;
}
}
} // namespace memgraph::coordination
#endif

147
src/coordination/coordinator_cluster_state.cpp Normal file
View File

@ -0,0 +1,147 @@
// Copyright 2024 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.
#ifdef MG_ENTERPRISE
#include "nuraft/coordinator_cluster_state.hpp"
#include "utils/logging.hpp"
#include <shared_mutex>
namespace memgraph::coordination {
void to_json(nlohmann::json &j, InstanceState const &instance_state) {
j = nlohmann::json{{"config", instance_state.config}, {"status", instance_state.status}};
}
void from_json(nlohmann::json const &j, InstanceState &instance_state) {
j.at("config").get_to(instance_state.config);
j.at("status").get_to(instance_state.status);
}
CoordinatorClusterState::CoordinatorClusterState(std::map<std::string, InstanceState, std::less<>> instances)
: instances_{std::move(instances)} {}
CoordinatorClusterState::CoordinatorClusterState(CoordinatorClusterState const &other) : instances_{other.instances_} {}
CoordinatorClusterState &CoordinatorClusterState::operator=(CoordinatorClusterState const &other) {
if (this == &other) {
return *this;
}
instances_ = other.instances_;
return *this;
}
CoordinatorClusterState::CoordinatorClusterState(CoordinatorClusterState &&other) noexcept
: instances_{std::move(other.instances_)} {}
CoordinatorClusterState &CoordinatorClusterState::operator=(CoordinatorClusterState &&other) noexcept {
if (this == &other) {
return *this;
}
instances_ = std::move(other.instances_);
return *this;
}
auto CoordinatorClusterState::MainExists() const -> bool {
auto lock = std::shared_lock{log_lock_};
return std::ranges::any_of(instances_,
[](auto const &entry) { return entry.second.status == ReplicationRole::MAIN; });
}
auto CoordinatorClusterState::IsMain(std::string_view instance_name) const -> bool {
auto lock = std::shared_lock{log_lock_};
auto const it = instances_.find(instance_name);
return it != instances_.end() && it->second.status == ReplicationRole::MAIN;
}
auto CoordinatorClusterState::IsReplica(std::string_view instance_name) const -> bool {
auto lock = std::shared_lock{log_lock_};
auto const it = instances_.find(instance_name);
return it != instances_.end() && it->second.status == ReplicationRole::REPLICA;
}
auto CoordinatorClusterState::InsertInstance(std::string instance_name, InstanceState instance_state) -> void {
auto lock = std::lock_guard{log_lock_};
instances_.insert_or_assign(std::move(instance_name), std::move(instance_state));
}
auto CoordinatorClusterState::DoAction(TRaftLog log_entry, RaftLogAction log_action) -> void {
auto lock = std::lock_guard{log_lock_};
switch (log_action) {
case RaftLogAction::REGISTER_REPLICATION_INSTANCE: {
auto const &config = std::get<CoordinatorClientConfig>(log_entry);
instances_[config.instance_name] = InstanceState{config, ReplicationRole::REPLICA};
break;
}
case RaftLogAction::UNREGISTER_REPLICATION_INSTANCE: {
auto const instance_name = std::get<std::string>(log_entry);
instances_.erase(instance_name);
break;
}
case RaftLogAction::SET_INSTANCE_AS_MAIN: {
auto const instance_name = std::get<std::string>(log_entry);
auto it = instances_.find(instance_name);
MG_ASSERT(it != instances_.end(), "Instance does not exist as part of raft state!");
it->second.status = ReplicationRole::MAIN;
break;
}
case RaftLogAction::SET_INSTANCE_AS_REPLICA: {
auto const instance_name = std::get<std::string>(log_entry);
auto it = instances_.find(instance_name);
MG_ASSERT(it != instances_.end(), "Instance does not exist as part of raft state!");
it->second.status = ReplicationRole::REPLICA;
break;
}
case RaftLogAction::UPDATE_UUID: {
uuid_ = std::get<utils::UUID>(log_entry);
break;
}
}
}
auto CoordinatorClusterState::Serialize(ptr<buffer> &data) -> void {
auto lock = std::shared_lock{log_lock_};
auto const log = nlohmann::json(instances_).dump();
data = buffer::alloc(sizeof(uint32_t) + log.size());
buffer_serializer bs(data);
bs.put_str(log);
}
auto CoordinatorClusterState::Deserialize(buffer &data) -> CoordinatorClusterState {
buffer_serializer bs(data);
auto const j = nlohmann::json::parse(bs.get_str());
auto instances = j.get<std::map<std::string, InstanceState, std::less<>>>();
return CoordinatorClusterState{std::move(instances)};
}
auto CoordinatorClusterState::GetInstances() const -> std::vector<InstanceState> {
auto lock = std::shared_lock{log_lock_};
return instances_ | ranges::views::values | ranges::to<std::vector<InstanceState>>;
}
auto CoordinatorClusterState::GetUUID() const -> utils::UUID { return uuid_; }
auto CoordinatorClusterState::FindCurrentMainInstanceName() const -> std::optional<std::string> {
auto lock = std::shared_lock{log_lock_};
auto const it =
std::ranges::find_if(instances_, [](auto const &entry) { return entry.second.status == ReplicationRole::MAIN; });
if (it == instances_.end()) {
return {};
}
return it->first;
}
} // namespace memgraph::coordination
#endif

54
src/coordination/coordinator_config.cpp Normal file
View File

@ -0,0 +1,54 @@
// Copyright 2024 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.
#ifdef MG_ENTERPRISE
#include "coordination/coordinator_config.hpp"
namespace memgraph::coordination {
void to_json(nlohmann::json &j, ReplClientInfo const &config) {
j = nlohmann::json{{"instance_name", config.instance_name},
{"replication_mode", config.replication_mode},
{"replication_ip_address", config.replication_ip_address},
{"replication_port", config.replication_port}};
}
void from_json(nlohmann::json const &j, ReplClientInfo &config) {
config.instance_name = j.at("instance_name").get<std::string>();
config.replication_mode = j.at("replication_mode").get<replication_coordination_glue::ReplicationMode>();
config.replication_ip_address = j.at("replication_ip_address").get<std::string>();
config.replication_port = j.at("replication_port").get<uint16_t>();
}
void to_json(nlohmann::json &j, CoordinatorClientConfig const &config) {
j = nlohmann::json{{"instance_name", config.instance_name},
{"ip_address", config.ip_address},
{"port", config.port},
{"instance_health_check_frequency_sec", config.instance_health_check_frequency_sec.count()},
{"instance_down_timeout_sec", config.instance_down_timeout_sec.count()},
{"instance_get_uuid_frequency_sec", config.instance_get_uuid_frequency_sec.count()},
{"replication_client_info", config.replication_client_info}};
}
void from_json(nlohmann::json const &j, CoordinatorClientConfig &config) {
config.instance_name = j.at("instance_name").get<std::string>();
config.ip_address = j.at("ip_address").get<std::string>();
config.port = j.at("port").get<uint16_t>();
config.instance_health_check_frequency_sec =
std::chrono::seconds{j.at("instance_health_check_frequency_sec").get<int>()};
config.instance_down_timeout_sec = std::chrono::seconds{j.at("instance_down_timeout_sec").get<int>()};
config.instance_get_uuid_frequency_sec = std::chrono::seconds{j.at("instance_get_uuid_frequency_sec").get<int>()};
config.replication_client_info = j.at("replication_client_info").get<ReplClientInfo>();
}
} // namespace memgraph::coordination
#endif

11
src/coordination/coordinator_handlers.cpp
View File

@ -57,6 +57,17 @@ void CoordinatorHandlers::Register(memgraph::coordination::CoordinatorServer &se
spdlog::info("Received GetInstanceUUIDRpc on coordinator server");
CoordinatorHandlers::GetInstanceUUIDHandler(replication_handler, req_reader, res_builder);
});
server.Register<coordination::GetDatabaseHistoriesRpc>(
[&replication_handler](slk::Reader *req_reader, slk::Builder *res_builder) -> void {
spdlog::info("Received GetDatabasesHistoryRpc on coordinator server");
CoordinatorHandlers::GetDatabaseHistoriesHandler(replication_handler, req_reader, res_builder);
});
}
void CoordinatorHandlers::GetDatabaseHistoriesHandler(replication::ReplicationHandler &replication_handler,
slk::Reader * /*req_reader*/, slk::Builder *res_builder) {
slk::Save(coordination::GetDatabaseHistoriesRes{replication_handler.GetDatabasesHistories()}, res_builder);
}
void CoordinatorHandlers::SwapMainUUIDHandler(replication::ReplicationHandler &replication_handler,

578
src/coordination/coordinator_instance.cpp
View File

@ -15,10 +15,12 @@
#include "coordination/coordinator_exceptions.hpp"
#include "coordination/fmt.hpp"
#include "dbms/constants.hpp"
#include "nuraft/coordinator_state_machine.hpp"
#include "nuraft/coordinator_state_manager.hpp"
#include "utils/counter.hpp"
#include "utils/functional.hpp"
#include "utils/resource_lock.hpp"
#include <range/v3/view.hpp>
#include <shared_mutex>
@ -30,144 +32,156 @@ using nuraft::srv_config;
CoordinatorInstance::CoordinatorInstance()
: raft_state_(RaftState::MakeRaftState(
[this] { std::ranges::for_each(repl_instances_, &ReplicationInstance::StartFrequentCheck); },
[this] { std::ranges::for_each(repl_instances_, &ReplicationInstance::StopFrequentCheck); })) {
auto find_repl_instance = [](CoordinatorInstance *self,
std::string_view repl_instance_name) -> ReplicationInstance & {
auto repl_instance =
std::ranges::find_if(self->repl_instances_, [repl_instance_name](ReplicationInstance const &instance) {
return instance.InstanceName() == repl_instance_name;
});
[this]() {
spdlog::info("Leader changed, starting all replication instances!");
auto const instances = raft_state_.GetInstances();
auto replicas = instances | ranges::views::filter([](auto const &instance) {
return instance.status == ReplicationRole::REPLICA;
});
MG_ASSERT(repl_instance != self->repl_instances_.end(), "Instance {} not found during callback!",
repl_instance_name);
return *repl_instance;
std::ranges::for_each(replicas, [this](auto &replica) {
spdlog::info("Started pinging replication instance {}", replica.config.instance_name);
repl_instances_.emplace_back(this, replica.config, client_succ_cb_, client_fail_cb_,
&CoordinatorInstance::ReplicaSuccessCallback,
&CoordinatorInstance::ReplicaFailCallback);
});
auto main = instances | ranges::views::filter(
[](auto const &instance) { return instance.status == ReplicationRole::MAIN; });
std::ranges::for_each(main, [this](auto &main_instance) {
spdlog::info("Started pinging main instance {}", main_instance.config.instance_name);
repl_instances_.emplace_back(this, main_instance.config, client_succ_cb_, client_fail_cb_,
&CoordinatorInstance::MainSuccessCallback,
&CoordinatorInstance::MainFailCallback);
});
std::ranges::for_each(repl_instances_, [this](auto &instance) {
instance.SetNewMainUUID(raft_state_.GetUUID());
instance.StartFrequentCheck();
});
},
[this]() {
spdlog::info("Leader changed, stopping all replication instances!");
repl_instances_.clear();
})) {
client_succ_cb_ = [](CoordinatorInstance *self, std::string_view repl_instance_name) -> void {
auto lock = std::lock_guard{self->coord_instance_lock_};
auto &repl_instance = self->FindReplicationInstance(repl_instance_name);
std::invoke(repl_instance.GetSuccessCallback(), self, repl_instance_name);
};
replica_succ_cb_ = [find_repl_instance](CoordinatorInstance *self, std::string_view repl_instance_name) -> void {
client_fail_cb_ = [](CoordinatorInstance *self, std::string_view repl_instance_name) -> void {
auto lock = std::lock_guard{self->coord_instance_lock_};
spdlog::trace("Instance {} performing replica successful callback", repl_instance_name);
auto &repl_instance = find_repl_instance(self, repl_instance_name);
// We need to get replicas UUID from time to time to ensure replica is listening to correct main
// and that it didn't go down for less time than we could notice
// We need to get id of main replica is listening to
// and swap if necessary
if (!repl_instance.EnsureReplicaHasCorrectMainUUID(self->GetMainUUID())) {
spdlog::error("Failed to swap uuid for replica instance {} which is alive", repl_instance.InstanceName());
return;
}
repl_instance.OnSuccessPing();
};
replica_fail_cb_ = [find_repl_instance](CoordinatorInstance *self, std::string_view repl_instance_name) -> void {
auto lock = std::lock_guard{self->coord_instance_lock_};
spdlog::trace("Instance {} performing replica failure callback", repl_instance_name);
auto &repl_instance = find_repl_instance(self, repl_instance_name);
repl_instance.OnFailPing();
};
main_succ_cb_ = [find_repl_instance](CoordinatorInstance *self, std::string_view repl_instance_name) -> void {
auto lock = std::lock_guard{self->coord_instance_lock_};
spdlog::trace("Instance {} performing main successful callback", repl_instance_name);
auto &repl_instance = find_repl_instance(self, repl_instance_name);
if (repl_instance.IsAlive()) {
repl_instance.OnSuccessPing();
return;
}
const auto &repl_instance_uuid = repl_instance.GetMainUUID();
MG_ASSERT(repl_instance_uuid.has_value(), "Instance must have uuid set.");
auto const curr_main_uuid = self->GetMainUUID();
if (curr_main_uuid == repl_instance_uuid.value()) {
if (!repl_instance.EnableWritingOnMain()) {
spdlog::error("Failed to enable writing on main instance {}", repl_instance_name);
return;
}
repl_instance.OnSuccessPing();
return;
}
// TODO(antoniof) make demoteToReplica idempotent since main can be demoted to replica but
// swapUUID can fail
if (repl_instance.DemoteToReplica(self->replica_succ_cb_, self->replica_fail_cb_)) {
repl_instance.OnSuccessPing();
spdlog::info("Instance {} demoted to replica", repl_instance_name);
} else {
spdlog::error("Instance {} failed to become replica", repl_instance_name);
return;
}
if (!repl_instance.SendSwapAndUpdateUUID(curr_main_uuid)) {
spdlog::error(fmt::format("Failed to swap uuid for demoted main instance {}", repl_instance.InstanceName()));
return;
}
};
main_fail_cb_ = [find_repl_instance](CoordinatorInstance *self, std::string_view repl_instance_name) -> void {
auto lock = std::lock_guard{self->coord_instance_lock_};
spdlog::trace("Instance {} performing main failure callback", repl_instance_name);
auto &repl_instance = find_repl_instance(self, repl_instance_name);
repl_instance.OnFailPing();
const auto &repl_instance_uuid = repl_instance.GetMainUUID();
MG_ASSERT(repl_instance_uuid.has_value(), "Instance must have uuid set");
if (!repl_instance.IsAlive() && self->GetMainUUID() == repl_instance_uuid.value()) {
spdlog::info("Cluster without main instance, trying automatic failover");
self->TryFailover(); // TODO: (andi) Initiate failover
}
auto &repl_instance = self->FindReplicationInstance(repl_instance_name);
std::invoke(repl_instance.GetFailCallback(), self, repl_instance_name);
};
}
auto CoordinatorInstance::FindReplicationInstance(std::string_view replication_instance_name) -> ReplicationInstance & {
auto repl_instance =
std::ranges::find_if(repl_instances_, [replication_instance_name](ReplicationInstance const &instance) {
return instance.InstanceName() == replication_instance_name;
});
MG_ASSERT(repl_instance != repl_instances_.end(), "Instance {} not found during callback!",
replication_instance_name);
return *repl_instance;
}
auto CoordinatorInstance::ShowInstances() const -> std::vector<InstanceStatus> {
auto const coord_instances = raft_state_.GetAllCoordinators();
auto const stringify_repl_role = [](ReplicationInstance const &instance) -> std::string {
if (!instance.IsAlive()) return "unknown";
if (instance.IsMain()) return "main";
return "replica";
};
auto const repl_instance_to_status = [&stringify_repl_role](ReplicationInstance const &instance) -> InstanceStatus {
return {.instance_name = instance.InstanceName(),
.coord_socket_address = instance.SocketAddress(),
.cluster_role = stringify_repl_role(instance),
.is_alive = instance.IsAlive()};
};
auto const coord_instance_to_status = [](ptr<srv_config> const &instance) -> InstanceStatus {
return {.instance_name = "coordinator_" + std::to_string(instance->get_id()),
.raft_socket_address = instance->get_endpoint(),
.cluster_role = "coordinator",
.is_alive = true}; // TODO: (andi) Get this info from RAFT and test it or when we will move
// CoordinatorState to every instance, we can be smarter about this using our RPC.
.health = "unknown"}; // TODO: (andi) Get this info from RAFT and test it or when we will move
};
auto instances_status = utils::fmap(raft_state_.GetAllCoordinators(), coord_instance_to_status);
auto instances_status = utils::fmap(coord_instance_to_status, coord_instances);
{
auto lock = std::shared_lock{coord_instance_lock_};
std::ranges::transform(repl_instances_, std::back_inserter(instances_status), repl_instance_to_status);
if (raft_state_.IsLeader()) {
auto const stringify_repl_role = [this](ReplicationInstance const &instance) -> std::string {
if (!instance.IsAlive()) return "unknown";
if (raft_state_.IsMain(instance.InstanceName())) return "main";
return "replica";
};
auto const stringify_repl_health = [](ReplicationInstance const &instance) -> std::string {
return instance.IsAlive() ? "up" : "down";
};
auto process_repl_instance_as_leader =
[&stringify_repl_role, &stringify_repl_health](ReplicationInstance const &instance) -> InstanceStatus {
return {.instance_name = instance.InstanceName(),
.coord_socket_address = instance.CoordinatorSocketAddress(),
.cluster_role = stringify_repl_role(instance),
.health = stringify_repl_health(instance)};
};
{
auto lock = std::shared_lock{coord_instance_lock_};
std::ranges::transform(repl_instances_, std::back_inserter(instances_status), process_repl_instance_as_leader);
}
} else {
auto const stringify_inst_status = [](ReplicationRole status) -> std::string {
return status == ReplicationRole::MAIN ? "main" : "replica";
};
// TODO: (andi) Add capability that followers can also return socket addresses
auto process_repl_instance_as_follower = [&stringify_inst_status](auto const &instance) -> InstanceStatus {
return {.instance_name = instance.config.instance_name,
.cluster_role = stringify_inst_status(instance.status),
.health = "unknown"};
};
std::ranges::transform(raft_state_.GetInstances(), std::back_inserter(instances_status),
process_repl_instance_as_follower);
}
return instances_status;
}
auto CoordinatorInstance::TryFailover() -> void {
auto alive_replicas = repl_instances_ | ranges::views::filter(&ReplicationInstance::IsReplica) |
ranges::views::filter(&ReplicationInstance::IsAlive);
auto const is_replica = [this](ReplicationInstance const &instance) { return IsReplica(instance.InstanceName()); };
auto alive_replicas =
repl_instances_ | ranges::views::filter(is_replica) | ranges::views::filter(&ReplicationInstance::IsAlive);
if (ranges::empty(alive_replicas)) {
spdlog::warn("Failover failed since all replicas are down!");
return;
}
// TODO: Smarter choice
auto new_main = ranges::begin(alive_replicas);
if (!raft_state_.RequestLeadership()) {
spdlog::error("Failover failed since the instance is not the leader!");
return;
}
auto const get_ts = [](ReplicationInstance &replica) { return replica.GetClient().SendGetInstanceTimestampsRpc(); };
auto maybe_instance_db_histories = alive_replicas | ranges::views::transform(get_ts) | ranges::to<std::vector>();
auto const ts_has_error = [](auto const &res) -> bool { return res.HasError(); };
if (std::ranges::any_of(maybe_instance_db_histories, ts_has_error)) {
spdlog::error("Aborting failover as at least one instance didn't provide per database history.");
return;
}
auto transform_to_pairs = ranges::views::transform([](auto const &zipped) {
auto &[replica, res] = zipped;
return std::make_pair(replica.InstanceName(), res.GetValue());
});
auto instance_db_histories =
ranges::views::zip(alive_replicas, maybe_instance_db_histories) | transform_to_pairs | ranges::to<std::vector>();
auto [most_up_to_date_instance, latest_epoch, latest_commit_timestamp] =
ChooseMostUpToDateInstance(instance_db_histories);
spdlog::trace("The most up to date instance is {} with epoch {} and {} latest commit timestamp",
most_up_to_date_instance, latest_epoch, latest_commit_timestamp); // NOLINT
auto *new_main = &FindReplicationInstance(most_up_to_date_instance);
new_main->PauseFrequentCheck();
utils::OnScopeExit scope_exit{[&new_main] { new_main->ResumeFrequentCheck(); }};
@ -177,41 +191,56 @@ auto CoordinatorInstance::TryFailover() -> void {
};
auto const new_main_uuid = utils::UUID{};
auto const failed_to_swap = [&new_main_uuid](ReplicationInstance &instance) {
return !instance.SendSwapAndUpdateUUID(new_main_uuid);
};
// If for some replicas swap fails, for others on successful ping we will revert back on next change
// or we will do failover first again and then it will be consistent again
for (auto &other_replica_instance : alive_replicas | ranges::views::filter(is_not_new_main)) {
if (!other_replica_instance.SendSwapAndUpdateUUID(new_main_uuid)) {
spdlog::error(fmt::format("Failed to swap uuid for instance {} which is alive, aborting failover",
other_replica_instance.InstanceName()));
return;
}
if (std::ranges::any_of(alive_replicas | ranges::views::filter(is_not_new_main), failed_to_swap)) {
spdlog::error("Failed to swap uuid for all instances");
return;
}
auto repl_clients_info = repl_instances_ | ranges::views::filter(is_not_new_main) |
ranges::views::transform(&ReplicationInstance::ReplicationClientInfo) |
ranges::to<ReplicationClientsInfo>();
if (!new_main->PromoteToMain(new_main_uuid, std::move(repl_clients_info), main_succ_cb_, main_fail_cb_)) {
if (!new_main->PromoteToMain(new_main_uuid, std::move(repl_clients_info), &CoordinatorInstance::MainSuccessCallback,
&CoordinatorInstance::MainFailCallback)) {
spdlog::warn("Failover failed since promoting replica to main failed!");
return;
}
// TODO: (andi) This should be replicated across all coordinator instances with Raft log
SetMainUUID(new_main_uuid);
if (!raft_state_.AppendUpdateUUIDLog(new_main_uuid)) {
return;
}
auto const new_main_instance_name = new_main->InstanceName();
if (!raft_state_.AppendSetInstanceAsMainLog(new_main_instance_name)) {
return;
}
spdlog::info("Failover successful! Instance {} promoted to main.", new_main->InstanceName());
}
// TODO: (andi) Make sure you cannot put coordinator instance to the main
auto CoordinatorInstance::SetReplicationInstanceToMain(std::string instance_name)
auto CoordinatorInstance::SetReplicationInstanceToMain(std::string_view instance_name)
-> SetInstanceToMainCoordinatorStatus {
auto lock = std::lock_guard{coord_instance_lock_};
if (std::ranges::any_of(repl_instances_, &ReplicationInstance::IsMain)) {
if (raft_state_.MainExists()) {
return SetInstanceToMainCoordinatorStatus::MAIN_ALREADY_EXISTS;
}
if (!raft_state_.RequestLeadership()) {
return SetInstanceToMainCoordinatorStatus::NOT_LEADER;
}
auto const is_new_main = [&instance_name](ReplicationInstance const &instance) {
return instance.InstanceName() == instance_name;
};
auto new_main = std::ranges::find_if(repl_instances_, is_new_main);
if (new_main == repl_instances_.end()) {
@ -229,85 +258,93 @@ auto CoordinatorInstance::SetReplicationInstanceToMain(std::string instance_name
auto const new_main_uuid = utils::UUID{};
for (auto &other_instance : repl_instances_ | ranges::views::filter(is_not_new_main)) {
if (!other_instance.SendSwapAndUpdateUUID(new_main_uuid)) {
spdlog::error(
fmt::format("Failed to swap uuid for instance {}, aborting failover", other_instance.InstanceName()));
return SetInstanceToMainCoordinatorStatus::SWAP_UUID_FAILED;
}
auto const failed_to_swap = [&new_main_uuid](ReplicationInstance &instance) {
return !instance.SendSwapAndUpdateUUID(new_main_uuid);
};
if (std::ranges::any_of(repl_instances_ | ranges::views::filter(is_not_new_main), failed_to_swap)) {
spdlog::error("Failed to swap uuid for all instances");
return SetInstanceToMainCoordinatorStatus::SWAP_UUID_FAILED;
}
ReplicationClientsInfo repl_clients_info;
repl_clients_info.reserve(repl_instances_.size() - 1);
std::ranges::transform(repl_instances_ | ranges::views::filter(is_not_new_main),
std::back_inserter(repl_clients_info), &ReplicationInstance::ReplicationClientInfo);
auto repl_clients_info = repl_instances_ | ranges::views::filter(is_not_new_main) |
ranges::views::transform(&ReplicationInstance::ReplicationClientInfo) |
ranges::to<ReplicationClientsInfo>();
if (!new_main->PromoteToMain(new_main_uuid, std::move(repl_clients_info), main_succ_cb_, main_fail_cb_)) {
if (!new_main->PromoteToMain(new_main_uuid, std::move(repl_clients_info), &CoordinatorInstance::MainSuccessCallback,
&CoordinatorInstance::MainFailCallback)) {
return SetInstanceToMainCoordinatorStatus::COULD_NOT_PROMOTE_TO_MAIN;
}
// TODO: (andi) This should be replicated across all coordinator instances with Raft log
SetMainUUID(new_main_uuid);
spdlog::info("Instance {} promoted to main", instance_name);
if (!raft_state_.AppendUpdateUUIDLog(new_main_uuid)) {
return SetInstanceToMainCoordinatorStatus::RAFT_LOG_ERROR;
}
if (!raft_state_.AppendSetInstanceAsMainLog(instance_name)) {
return SetInstanceToMainCoordinatorStatus::RAFT_LOG_ERROR;
}
spdlog::info("Instance {} promoted to main on leader", instance_name);
return SetInstanceToMainCoordinatorStatus::SUCCESS;
}
auto CoordinatorInstance::RegisterReplicationInstance(CoordinatorClientConfig config)
auto CoordinatorInstance::RegisterReplicationInstance(CoordinatorClientConfig const &config)
-> RegisterInstanceCoordinatorStatus {
auto lock = std::lock_guard{coord_instance_lock_};
auto instance_name = config.instance_name;
auto const name_matches = [&instance_name](ReplicationInstance const &instance) {
return instance.InstanceName() == instance_name;
};
if (std::ranges::any_of(repl_instances_, name_matches)) {
if (std::ranges::any_of(repl_instances_, [instance_name = config.instance_name](ReplicationInstance const &instance) {
return instance.InstanceName() == instance_name;
})) {
return RegisterInstanceCoordinatorStatus::NAME_EXISTS;
}
auto const socket_address_matches = [&config](ReplicationInstance const &instance) {
return instance.SocketAddress() == config.SocketAddress();
};
if (std::ranges::any_of(repl_instances_, [&config](ReplicationInstance const &instance) {
return instance.CoordinatorSocketAddress() == config.CoordinatorSocketAddress();
})) {
return RegisterInstanceCoordinatorStatus::COORD_ENDPOINT_EXISTS;
}
if (std::ranges::any_of(repl_instances_, socket_address_matches)) {
return RegisterInstanceCoordinatorStatus::ENDPOINT_EXISTS;
if (std::ranges::any_of(repl_instances_, [&config](ReplicationInstance const &instance) {
return instance.ReplicationSocketAddress() == config.ReplicationSocketAddress();
})) {
return RegisterInstanceCoordinatorStatus::REPL_ENDPOINT_EXISTS;
}
if (!raft_state_.RequestLeadership()) {
return RegisterInstanceCoordinatorStatus::NOT_LEADER;
}
auto const res = raft_state_.AppendRegisterReplicationInstance(instance_name);
if (!res->get_accepted()) {
spdlog::error(
"Failed to accept request for registering instance {}. Most likely the reason is that the instance is not "
"the "
"leader.",
config.instance_name);
return RegisterInstanceCoordinatorStatus::RAFT_COULD_NOT_ACCEPT;
}
auto const undo_action_ = [this]() { repl_instances_.pop_back(); };
spdlog::info("Request for registering instance {} accepted", instance_name);
try {
repl_instances_.emplace_back(this, std::move(config), replica_succ_cb_, replica_fail_cb_);
} catch (CoordinatorRegisterInstanceException const &) {
auto *new_instance = &repl_instances_.emplace_back(this, config, client_succ_cb_, client_fail_cb_,
&CoordinatorInstance::ReplicaSuccessCallback,
&CoordinatorInstance::ReplicaFailCallback);
if (!new_instance->SendDemoteToReplicaRpc()) {
spdlog::error("Failed to send demote to replica rpc for instance {}", config.instance_name);
undo_action_();
return RegisterInstanceCoordinatorStatus::RPC_FAILED;
}
if (res->get_result_code() != nuraft::cmd_result_code::OK) {
spdlog::error("Failed to register instance {} with error code {}", instance_name, res->get_result_code());
return RegisterInstanceCoordinatorStatus::RAFT_COULD_NOT_APPEND;
if (!raft_state_.AppendRegisterReplicationInstanceLog(config)) {
undo_action_();
return RegisterInstanceCoordinatorStatus::RAFT_LOG_ERROR;
}
spdlog::info("Instance {} registered", instance_name);
new_instance->StartFrequentCheck();
spdlog::info("Instance {} registered", config.instance_name);
return RegisterInstanceCoordinatorStatus::SUCCESS;
}
auto CoordinatorInstance::UnregisterReplicationInstance(std::string instance_name)
auto CoordinatorInstance::UnregisterReplicationInstance(std::string_view instance_name)
-> UnregisterInstanceCoordinatorStatus {
auto lock = std::lock_guard{coord_instance_lock_};
if (!raft_state_.RequestLeadership()) {
return UnregisterInstanceCoordinatorStatus::NOT_LEADER;
}
auto const name_matches = [&instance_name](ReplicationInstance const &instance) {
return instance.InstanceName() == instance_name;
};
@ -317,31 +354,208 @@ auto CoordinatorInstance::UnregisterReplicationInstance(std::string instance_nam
return UnregisterInstanceCoordinatorStatus::NO_INSTANCE_WITH_NAME;
}
if (inst_to_remove->IsMain() && inst_to_remove->IsAlive()) {
auto const is_main = [this](ReplicationInstance const &instance) {
return IsMain(instance.InstanceName()) && instance.GetMainUUID() == raft_state_.GetUUID() && instance.IsAlive();
};
if (is_main(*inst_to_remove)) {
return UnregisterInstanceCoordinatorStatus::IS_MAIN;
}
inst_to_remove->StopFrequentCheck();
auto curr_main = std::ranges::find_if(repl_instances_, &ReplicationInstance::IsMain);
MG_ASSERT(curr_main != repl_instances_.end(), "There must be a main instance when unregistering a replica");
if (!curr_main->SendUnregisterReplicaRpc(instance_name)) {
inst_to_remove->StartFrequentCheck();
return UnregisterInstanceCoordinatorStatus::RPC_FAILED;
auto curr_main = std::ranges::find_if(repl_instances_, is_main);
if (curr_main != repl_instances_.end() && curr_main->IsAlive()) {
if (!curr_main->SendUnregisterReplicaRpc(instance_name)) {
inst_to_remove->StartFrequentCheck();
return UnregisterInstanceCoordinatorStatus::RPC_FAILED;
}
}
std::erase_if(repl_instances_, name_matches);
if (!raft_state_.AppendUnregisterReplicationInstanceLog(instance_name)) {
return UnregisterInstanceCoordinatorStatus::RAFT_LOG_ERROR;
}
return UnregisterInstanceCoordinatorStatus::SUCCESS;
}
auto CoordinatorInstance::AddCoordinatorInstance(uint32_t raft_server_id, uint32_t raft_port, std::string raft_address)
-> void {
raft_state_.AddCoordinatorInstance(raft_server_id, raft_port, std::move(raft_address));
auto CoordinatorInstance::AddCoordinatorInstance(uint32_t raft_server_id, uint32_t raft_port,
std::string_view raft_address) -> void {
raft_state_.AddCoordinatorInstance(raft_server_id, raft_port, raft_address);
}
auto CoordinatorInstance::GetMainUUID() const -> utils::UUID { return main_uuid_; }
void CoordinatorInstance::MainFailCallback(std::string_view repl_instance_name) {
spdlog::trace("Instance {} performing main fail callback", repl_instance_name);
auto &repl_instance = FindReplicationInstance(repl_instance_name);
repl_instance.OnFailPing();
const auto &repl_instance_uuid = repl_instance.GetMainUUID();
MG_ASSERT(repl_instance_uuid.has_value(), "Replication instance must have uuid set");
// TODO: (andi) Add to the RAFT log.
auto CoordinatorInstance::SetMainUUID(utils::UUID new_uuid) -> void { main_uuid_ = new_uuid; }
// NOLINTNEXTLINE
if (!repl_instance.IsAlive() && raft_state_.GetUUID() == repl_instance_uuid.value()) {
spdlog::info("Cluster without main instance, trying automatic failover");
TryFailover();
}
}
void CoordinatorInstance::MainSuccessCallback(std::string_view repl_instance_name) {
spdlog::trace("Instance {} performing main successful callback", repl_instance_name);
auto &repl_instance = FindReplicationInstance(repl_instance_name);
if (repl_instance.IsAlive()) {
repl_instance.OnSuccessPing();
return;
}
const auto &repl_instance_uuid = repl_instance.GetMainUUID();
MG_ASSERT(repl_instance_uuid.has_value(), "Instance must have uuid set.");
// NOLINTNEXTLINE
if (raft_state_.GetUUID() == repl_instance_uuid.value()) {
if (!repl_instance.EnableWritingOnMain()) {
spdlog::error("Failed to enable writing on main instance {}", repl_instance_name);
return;
}
repl_instance.OnSuccessPing();
return;
}
if (!raft_state_.RequestLeadership()) {
spdlog::error("Demoting main instance {} to replica failed since the instance is not the leader!",
repl_instance_name);
return;
}
if (repl_instance.DemoteToReplica(&CoordinatorInstance::ReplicaSuccessCallback,
&CoordinatorInstance::ReplicaFailCallback)) {
repl_instance.OnSuccessPing();
spdlog::info("Instance {} demoted to replica", repl_instance_name);
} else {
spdlog::error("Instance {} failed to become replica", repl_instance_name);
return;
}
if (!repl_instance.SendSwapAndUpdateUUID(raft_state_.GetUUID())) {
spdlog::error("Failed to swap uuid for demoted main instance {}", repl_instance_name);
return;
}
if (!raft_state_.AppendSetInstanceAsReplicaLog(repl_instance_name)) {
return;
}
}
void CoordinatorInstance::ReplicaSuccessCallback(std::string_view repl_instance_name) {
spdlog::trace("Instance {} performing replica successful callback", repl_instance_name);
auto &repl_instance = FindReplicationInstance(repl_instance_name);
if (!IsReplica(repl_instance_name)) {
spdlog::error("Aborting replica callback since instance {} is not replica anymore", repl_instance_name);
return;
}
// We need to get replicas UUID from time to time to ensure replica is listening to correct main
// and that it didn't go down for less time than we could notice
// We need to get id of main replica is listening to
// and swap if necessary
if (!repl_instance.EnsureReplicaHasCorrectMainUUID(raft_state_.GetUUID())) {
spdlog::error("Failed to swap uuid for replica instance {} which is alive", repl_instance.InstanceName());
return;
}
repl_instance.OnSuccessPing();
}
void CoordinatorInstance::ReplicaFailCallback(std::string_view repl_instance_name) {
spdlog::trace("Instance {} performing replica failure callback", repl_instance_name);
auto &repl_instance = FindReplicationInstance(repl_instance_name);
if (!IsReplica(repl_instance_name)) {
spdlog::error("Aborting replica fail callback since instance {} is not replica anymore", repl_instance_name);
return;
}
repl_instance.OnFailPing();
}
auto CoordinatorInstance::ChooseMostUpToDateInstance(std::span<InstanceNameDbHistories> instance_database_histories)
-> NewMainRes {
std::optional<NewMainRes> new_main_res;
std::for_each(
instance_database_histories.begin(), instance_database_histories.end(),
[&new_main_res](const InstanceNameDbHistories &instance_res_pair) {
const auto &[instance_name, instance_db_histories] = instance_res_pair;
// Find default db for instance and its history
auto default_db_history_data = std::ranges::find_if(
instance_db_histories, [default_db = memgraph::dbms::kDefaultDB](
const replication_coordination_glue::DatabaseHistory &db_timestamps) {
return db_timestamps.name == default_db;
});
std::ranges::for_each(
instance_db_histories,
[&instance_name = instance_name](const replication_coordination_glue::DatabaseHistory &db_history) {
spdlog::debug("Instance {}: name {}, default db {}", instance_name, db_history.name,
memgraph::dbms::kDefaultDB);
});
MG_ASSERT(default_db_history_data != instance_db_histories.end(), "No history for instance");
const auto &instance_default_db_history = default_db_history_data->history;
std::ranges::for_each(instance_default_db_history | ranges::views::reverse,
[&instance_name = instance_name](const auto &epoch_history_it) {
spdlog::debug("Instance {}: epoch {}, last_commit_timestamp: {}", instance_name,
std::get<0>(epoch_history_it), std::get<1>(epoch_history_it));
});
// get latest epoch
// get latest timestamp
if (!new_main_res) {
const auto &[epoch, timestamp] = *instance_default_db_history.crbegin();
new_main_res = std::make_optional<NewMainRes>({instance_name, epoch, timestamp});
spdlog::debug("Currently the most up to date instance is {} with epoch {} and {} latest commit timestamp",
instance_name, epoch, timestamp);
return;
}
bool found_same_point{false};
std::string last_most_up_to_date_epoch{new_main_res->latest_epoch};
for (auto [epoch, timestamp] : ranges::reverse_view(instance_default_db_history)) {
if (new_main_res->latest_commit_timestamp < timestamp) {
new_main_res = std::make_optional<NewMainRes>({instance_name, epoch, timestamp});
spdlog::trace("Found the new most up to date instance {} with epoch {} and {} latest commit timestamp",
instance_name, epoch, timestamp);
}
// we found point at which they were same
if (epoch == last_most_up_to_date_epoch) {
found_same_point = true;
break;
}
}
if (!found_same_point) {
spdlog::error("Didn't find same history epoch {} for instance {} and instance {}", last_most_up_to_date_epoch,
new_main_res->most_up_to_date_instance, instance_name);
}
});
return std::move(*new_main_res);
}
auto CoordinatorInstance::IsMain(std::string_view instance_name) const -> bool {
return raft_state_.IsMain(instance_name);
}
auto CoordinatorInstance::IsReplica(std::string_view instance_name) const -> bool {
return raft_state_.IsReplica(instance_name);
}
} // namespace memgraph::coordination
#endif

42
src/coordination/coordinator_log_store.cpp
View File

@ -62,34 +62,33 @@ ptr<log_entry> CoordinatorLogStore::last_entry() const {
uint64_t CoordinatorLogStore::append(ptr<log_entry> &entry) {
ptr<log_entry> clone = MakeClone(entry);
uint64_t next_slot{0};
{
auto lock = std::lock_guard{logs_lock_};
next_slot = start_idx_ + logs_.size() - 1;
logs_[next_slot] = clone;
}
auto lock = std::lock_guard{logs_lock_};
uint64_t next_slot = start_idx_ + logs_.size() - 1;
logs_[next_slot] = clone;
return next_slot;
}
// TODO: (andi) I think this is used for resolving conflicts inside NuRaft, check...
// different compared to in_memory_log_store.cxx
void CoordinatorLogStore::write_at(uint64_t index, ptr<log_entry> &entry) {
ptr<log_entry> clone = MakeClone(entry);
// Discard all logs equal to or greater than `index.
{
auto lock = std::lock_guard{logs_lock_};
auto itr = logs_.lower_bound(index);
while (itr != logs_.end()) {
itr = logs_.erase(itr);
}
logs_[index] = clone;
auto lock = std::lock_guard{logs_lock_};
auto itr = logs_.lower_bound(index);
while (itr != logs_.end()) {
itr = logs_.erase(itr);
}
logs_[index] = clone;
}
ptr<std::vector<ptr<log_entry>>> CoordinatorLogStore::log_entries(uint64_t start, uint64_t end) {
auto ret = cs_new<std::vector<ptr<log_entry>>>();
ret->resize(end - start);
for (uint64_t i = start, curr_index = 0; i < end; ++i, ++curr_index) {
for (uint64_t i = start, curr_index = 0; i < end; i++, curr_index++) {
ptr<log_entry> src = nullptr;
{
auto lock = std::lock_guard{logs_lock_};
@ -105,21 +104,14 @@ ptr<std::vector<ptr<log_entry>>> CoordinatorLogStore::log_entries(uint64_t start
}
ptr<log_entry> CoordinatorLogStore::entry_at(uint64_t index) {
ptr<log_entry> src = nullptr;
{
auto lock = std::lock_guard{logs_lock_};
src = FindOrDefault_(index);
}
auto lock = std::lock_guard{logs_lock_};
ptr<log_entry> src = FindOrDefault_(index);
return MakeClone(src);
}
uint64_t CoordinatorLogStore::term_at(uint64_t index) {
uint64_t term = 0;
{
auto lock = std::lock_guard{logs_lock_};
term = FindOrDefault_(index)->get_term();
}
return term;
auto lock = std::lock_guard{logs_lock_};
return FindOrDefault_(index)->get_term();
}
ptr<buffer> CoordinatorLogStore::pack(uint64_t index, int32 cnt) {

38
src/coordination/coordinator_rpc.cpp
View File

@ -76,9 +76,9 @@ void EnableWritingOnMainRes::Load(EnableWritingOnMainRes *self, memgraph::slk::R
memgraph::slk::Load(self, reader);
}
void EnableWritingOnMainReq::Save(EnableWritingOnMainReq const &self, memgraph::slk::Builder *builder) {}
void EnableWritingOnMainReq::Save(EnableWritingOnMainReq const & /*self*/, memgraph::slk::Builder * /*builder*/) {}
void EnableWritingOnMainReq::Load(EnableWritingOnMainReq *self, memgraph::slk::Reader *reader) {}
void EnableWritingOnMainReq::Load(EnableWritingOnMainReq * /*self*/, memgraph::slk::Reader * /*reader*/) {}
// GetInstanceUUID
void GetInstanceUUIDReq::Save(const GetInstanceUUIDReq &self, memgraph::slk::Builder *builder) {
@ -97,6 +97,24 @@ void GetInstanceUUIDRes::Load(GetInstanceUUIDRes *self, memgraph::slk::Reader *r
memgraph::slk::Load(self, reader);
}
// GetDatabaseHistoriesRpc
void GetDatabaseHistoriesReq::Save(const GetDatabaseHistoriesReq & /*self*/, memgraph::slk::Builder * /*builder*/) {
/* nothing to serialize */
}
void GetDatabaseHistoriesReq::Load(GetDatabaseHistoriesReq * /*self*/, memgraph::slk::Reader * /*reader*/) {
/* nothing to serialize */
}
void GetDatabaseHistoriesRes::Save(const GetDatabaseHistoriesRes &self, memgraph::slk::Builder *builder) {
memgraph::slk::Save(self, builder);
}
void GetDatabaseHistoriesRes::Load(GetDatabaseHistoriesRes *self, memgraph::slk::Reader *reader) {
memgraph::slk::Load(self, reader);
}
} // namespace coordination
constexpr utils::TypeInfo coordination::PromoteReplicaToMainReq::kType{utils::TypeId::COORD_FAILOVER_REQ,
@ -130,6 +148,12 @@ constexpr utils::TypeInfo coordination::GetInstanceUUIDReq::kType{utils::TypeId:
constexpr utils::TypeInfo coordination::GetInstanceUUIDRes::kType{utils::TypeId::COORD_GET_UUID_RES, "CoordGetUUIDRes",
nullptr};
constexpr utils::TypeInfo coordination::GetDatabaseHistoriesReq::kType{utils::TypeId::COORD_GET_INSTANCE_DATABASES_REQ,
"GetInstanceDatabasesReq", nullptr};
constexpr utils::TypeInfo coordination::GetDatabaseHistoriesRes::kType{utils::TypeId::COORD_GET_INSTANCE_DATABASES_RES,
"GetInstanceDatabasesRes", nullptr};
namespace slk {
// PromoteReplicaToMainRpc
@ -213,6 +237,16 @@ void Load(memgraph::coordination::GetInstanceUUIDRes *self, memgraph::slk::Reade
memgraph::slk::Load(&self->uuid, reader);
}
// GetInstanceTimestampsReq
void Save(const memgraph::coordination::GetDatabaseHistoriesRes &self, memgraph::slk::Builder *builder) {
memgraph::slk::Save(self.database_histories, builder);
}
void Load(memgraph::coordination::GetDatabaseHistoriesRes *self, memgraph::slk::Reader *reader) {
memgraph::slk::Load(&self->database_histories, reader);
}
} // namespace slk
} // namespace memgraph

12
src/coordination/coordinator_state.cpp
View File

@ -41,7 +41,7 @@ CoordinatorState::CoordinatorState() {
}
}
auto CoordinatorState::RegisterReplicationInstance(CoordinatorClientConfig config)
auto CoordinatorState::RegisterReplicationInstance(CoordinatorClientConfig const &config)
-> RegisterInstanceCoordinatorStatus {
MG_ASSERT(std::holds_alternative<CoordinatorInstance>(data_),
"Coordinator cannot register replica since variant holds wrong alternative");
@ -56,7 +56,8 @@ auto CoordinatorState::RegisterReplicationInstance(CoordinatorClientConfig confi
data_);
}
auto CoordinatorState::UnregisterReplicationInstance(std::string instance_name) -> UnregisterInstanceCoordinatorStatus {
auto CoordinatorState::UnregisterReplicationInstance(std::string_view instance_name)
-> UnregisterInstanceCoordinatorStatus {
MG_ASSERT(std::holds_alternative<CoordinatorInstance>(data_),
"Coordinator cannot unregister instance since variant holds wrong alternative");
@ -70,7 +71,8 @@ auto CoordinatorState::UnregisterReplicationInstance(std::string instance_name)
data_);
}
auto CoordinatorState::SetReplicationInstanceToMain(std::string instance_name) -> SetInstanceToMainCoordinatorStatus {
auto CoordinatorState::SetReplicationInstanceToMain(std::string_view instance_name)
-> SetInstanceToMainCoordinatorStatus {
MG_ASSERT(std::holds_alternative<CoordinatorInstance>(data_),
"Coordinator cannot register replica since variant holds wrong alternative");
@ -96,8 +98,8 @@ auto CoordinatorState::GetCoordinatorServer() const -> CoordinatorServer & {
return *std::get<CoordinatorMainReplicaData>(data_).coordinator_server_;
}
auto CoordinatorState::AddCoordinatorInstance(uint32_t raft_server_id, uint32_t raft_port, std::string raft_address)
-> void {
auto CoordinatorState::AddCoordinatorInstance(uint32_t raft_server_id, uint32_t raft_port,
std::string_view raft_address) -> void {
MG_ASSERT(std::holds_alternative<CoordinatorInstance>(data_),
"Coordinator cannot register replica since variant holds wrong alternative");
return std::get<CoordinatorInstance>(data_).AddCoordinatorInstance(raft_server_id, raft_port, raft_address);

206
src/coordination/coordinator_state_machine.cpp
View File

@ -12,100 +12,204 @@
#ifdef MG_ENTERPRISE
#include "nuraft/coordinator_state_machine.hpp"
#include "utils/logging.hpp"
namespace {
constexpr int MAX_SNAPSHOTS = 3;
} // namespace
namespace memgraph::coordination {
auto CoordinatorStateMachine::EncodeRegisterReplicationInstance(const std::string &name) -> ptr<buffer> {
std::string str_log = name + "_replica";
ptr<buffer> log = buffer::alloc(sizeof(uint32_t) + str_log.size());
buffer_serializer bs(log);
bs.put_str(str_log);
return log;
auto CoordinatorStateMachine::FindCurrentMainInstanceName() const -> std::optional<std::string> {
return cluster_state_.FindCurrentMainInstanceName();
}
auto CoordinatorStateMachine::DecodeRegisterReplicationInstance(buffer &data) -> std::string {
auto CoordinatorStateMachine::MainExists() const -> bool { return cluster_state_.MainExists(); }
auto CoordinatorStateMachine::IsMain(std::string_view instance_name) const -> bool {
return cluster_state_.IsMain(instance_name);
}
auto CoordinatorStateMachine::IsReplica(std::string_view instance_name) const -> bool {
return cluster_state_.IsReplica(instance_name);
}
auto CoordinatorStateMachine::CreateLog(nlohmann::json &&log) -> ptr<buffer> {
auto const log_dump = log.dump();
ptr<buffer> log_buf = buffer::alloc(sizeof(uint32_t) + log_dump.size());
buffer_serializer bs(log_buf);
bs.put_str(log_dump);
return log_buf;
}
auto CoordinatorStateMachine::SerializeRegisterInstance(CoordinatorClientConfig const &config) -> ptr<buffer> {
return CreateLog({{"action", RaftLogAction::REGISTER_REPLICATION_INSTANCE}, {"info", config}});
}
auto CoordinatorStateMachine::SerializeUnregisterInstance(std::string_view instance_name) -> ptr<buffer> {
return CreateLog({{"action", RaftLogAction::UNREGISTER_REPLICATION_INSTANCE}, {"info", instance_name}});
}
auto CoordinatorStateMachine::SerializeSetInstanceAsMain(std::string_view instance_name) -> ptr<buffer> {
return CreateLog({{"action", RaftLogAction::SET_INSTANCE_AS_MAIN}, {"info", instance_name}});
}
auto CoordinatorStateMachine::SerializeSetInstanceAsReplica(std::string_view instance_name) -> ptr<buffer> {
return CreateLog({{"action", RaftLogAction::SET_INSTANCE_AS_REPLICA}, {"info", instance_name}});
}
auto CoordinatorStateMachine::SerializeUpdateUUID(utils::UUID const &uuid) -> ptr<buffer> {
return CreateLog({{"action", RaftLogAction::UPDATE_UUID}, {"info", uuid}});
}
auto CoordinatorStateMachine::DecodeLog(buffer &data) -> std::pair<TRaftLog, RaftLogAction> {
buffer_serializer bs(data);
return bs.get_str();
auto const json = nlohmann::json::parse(bs.get_str());
auto const action = json["action"].get<RaftLogAction>();
auto const &info = json["info"];
switch (action) {
case RaftLogAction::REGISTER_REPLICATION_INSTANCE:
return {info.get<CoordinatorClientConfig>(), action};
case RaftLogAction::UPDATE_UUID:
return {info.get<utils::UUID>(), action};
case RaftLogAction::UNREGISTER_REPLICATION_INSTANCE:
case RaftLogAction::SET_INSTANCE_AS_MAIN:
[[fallthrough]];
case RaftLogAction::SET_INSTANCE_AS_REPLICA:
return {info.get<std::string>(), action};
}
throw std::runtime_error("Unknown action");
}
auto CoordinatorStateMachine::pre_commit(ulong const log_idx, buffer &data) -> ptr<buffer> {
buffer_serializer bs(data);
std::string str = bs.get_str();
spdlog::info("pre_commit {} : {}", log_idx, str);
return nullptr;
}
auto CoordinatorStateMachine::pre_commit(ulong const /*log_idx*/, buffer & /*data*/) -> ptr<buffer> { return nullptr; }
auto CoordinatorStateMachine::commit(ulong const log_idx, buffer &data) -> ptr<buffer> {
buffer_serializer bs(data);
std::string str = bs.get_str();
spdlog::info("commit {} : {}", log_idx, str);
spdlog::debug("Commit: log_idx={}, data.size()={}", log_idx, data.size());
auto const [parsed_data, log_action] = DecodeLog(data);
cluster_state_.DoAction(parsed_data, log_action);
last_committed_idx_ = log_idx;
return nullptr;
// Return raft log number
ptr<buffer> ret = buffer::alloc(sizeof(log_idx));
buffer_serializer bs_ret(ret);
bs_ret.put_u64(log_idx);
return ret;
}
auto CoordinatorStateMachine::commit_config(ulong const log_idx, ptr<cluster_config> & /*new_conf*/) -> void {
last_committed_idx_ = log_idx;
spdlog::debug("Commit config: log_idx={}", log_idx);
}
auto CoordinatorStateMachine::rollback(ulong const log_idx, buffer &data) -> void {
buffer_serializer bs(data);
std::string str = bs.get_str();
spdlog::info("rollback {} : {}", log_idx, str);
// NOTE: Nothing since we don't do anything in pre_commit
spdlog::debug("Rollback: log_idx={}, data.size()={}", log_idx, data.size());
}
auto CoordinatorStateMachine::read_logical_snp_obj(snapshot & /*snapshot*/, void *& /*user_snp_ctx*/, ulong /*obj_id*/,
auto CoordinatorStateMachine::read_logical_snp_obj(snapshot &snapshot, void *& /*user_snp_ctx*/, ulong obj_id,
ptr<buffer> &data_out, bool &is_last_obj) -> int {
// Put dummy data.
data_out = buffer::alloc(sizeof(int32));
buffer_serializer bs(data_out);
bs.put_i32(0);
spdlog::debug("read logical snapshot object, obj_id: {}", obj_id);
ptr<SnapshotCtx> ctx = nullptr;
{
auto ll = std::lock_guard{snapshots_lock_};
auto entry = snapshots_.find(snapshot.get_last_log_idx());
if (entry == snapshots_.end()) {
data_out = nullptr;
is_last_obj = true;
return 0;
}
ctx = entry->second;
}
if (obj_id == 0) {
// Object ID == 0: first object, put dummy data.
data_out = buffer::alloc(sizeof(int32));
buffer_serializer bs(data_out);
bs.put_i32(0);
is_last_obj = false;
} else {
// Object ID > 0: second object, put actual value.
ctx->cluster_state_.Serialize(data_out);
}
is_last_obj = true;
return 0;
}
auto CoordinatorStateMachine::save_logical_snp_obj(snapshot &s, ulong &obj_id, buffer & /*data*/, bool /*is_first_obj*/,
bool /*is_last_obj*/) -> void {
spdlog::info("save snapshot {} term {} object ID", s.get_last_log_idx(), s.get_last_log_term(), obj_id);
// Request next object.
obj_id++;
auto CoordinatorStateMachine::save_logical_snp_obj(snapshot &snapshot, ulong &obj_id, buffer &data, bool is_first_obj,
bool is_last_obj) -> void {
spdlog::debug("save logical snapshot object, obj_id: {}, is_first_obj: {}, is_last_obj: {}", obj_id, is_first_obj,
is_last_obj);
if (obj_id == 0) {
ptr<buffer> snp_buf = snapshot.serialize();
auto ss = snapshot::deserialize(*snp_buf);
create_snapshot_internal(ss);
} else {
auto cluster_state = CoordinatorClusterState::Deserialize(data);
auto ll = std::lock_guard{snapshots_lock_};
auto entry = snapshots_.find(snapshot.get_last_log_idx());
DMG_ASSERT(entry != snapshots_.end());
entry->second->cluster_state_ = cluster_state;
}
}
auto CoordinatorStateMachine::apply_snapshot(snapshot &s) -> bool {
spdlog::info("apply snapshot {} term {}", s.get_last_log_idx(), s.get_last_log_term());
{
auto lock = std::lock_guard{last_snapshot_lock_};
ptr<buffer> snp_buf = s.serialize();
last_snapshot_ = snapshot::deserialize(*snp_buf);
}
auto ll = std::lock_guard{snapshots_lock_};
spdlog::debug("apply snapshot, last_log_idx: {}", s.get_last_log_idx());
auto entry = snapshots_.find(s.get_last_log_idx());
if (entry == snapshots_.end()) return false;
cluster_state_ = entry->second->cluster_state_;
return true;
}
auto CoordinatorStateMachine::free_user_snp_ctx(void *&user_snp_ctx) -> void {}
auto CoordinatorStateMachine::last_snapshot() -> ptr<snapshot> {
auto lock = std::lock_guard{last_snapshot_lock_};
return last_snapshot_;
auto ll = std::lock_guard{snapshots_lock_};
spdlog::debug("last_snapshot");
auto entry = snapshots_.rbegin();
if (entry == snapshots_.rend()) return nullptr;
ptr<SnapshotCtx> ctx = entry->second;
return ctx->snapshot_;
}
auto CoordinatorStateMachine::last_commit_index() -> ulong { return last_committed_idx_; }
auto CoordinatorStateMachine::create_snapshot(snapshot &s, async_result<bool>::handler_type &when_done) -> void {
spdlog::info("create snapshot {} term {}", s.get_last_log_idx(), s.get_last_log_term());
// Clone snapshot from `s`.
{
auto lock = std::lock_guard{last_snapshot_lock_};
ptr<buffer> snp_buf = s.serialize();
last_snapshot_ = snapshot::deserialize(*snp_buf);
}
spdlog::debug("create_snapshot, last_log_idx: {}", s.get_last_log_idx());
ptr<buffer> snp_buf = s.serialize();
ptr<snapshot> ss = snapshot::deserialize(*snp_buf);
create_snapshot_internal(ss);
ptr<std::exception> except(nullptr);
bool ret = true;
when_done(ret, except);
}
auto CoordinatorStateMachine::create_snapshot_internal(ptr<snapshot> snapshot) -> void {
auto ll = std::lock_guard{snapshots_lock_};
spdlog::debug("create_snapshot_internal, last_log_idx: {}", snapshot->get_last_log_idx());
auto ctx = cs_new<SnapshotCtx>(snapshot, cluster_state_);
snapshots_[snapshot->get_last_log_idx()] = ctx;
while (snapshots_.size() > MAX_SNAPSHOTS) {
snapshots_.erase(snapshots_.begin());
}
}
auto CoordinatorStateMachine::GetInstances() const -> std::vector<InstanceState> {
return cluster_state_.GetInstances();
}
auto CoordinatorStateMachine::GetUUID() const -> utils::UUID { return cluster_state_.GetUUID(); }
} // namespace memgraph::coordination
#endif

21
src/coordination/include/coordination/coordinator_client.hpp
View File

@ -14,6 +14,7 @@
#ifdef MG_ENTERPRISE
#include "coordination/coordinator_config.hpp"
#include "replication_coordination_glue/common.hpp"
#include "rpc/client.hpp"
#include "rpc_errors.hpp"
#include "utils/result.hpp"
@ -23,13 +24,13 @@
namespace memgraph::coordination {
class CoordinatorInstance;
using HealthCheckCallback = std::function<void(CoordinatorInstance *, std::string_view)>;
using HealthCheckClientCallback = std::function<void(CoordinatorInstance *, std::string_view)>;
using ReplicationClientsInfo = std::vector<ReplClientInfo>;
class CoordinatorClient {
public:
explicit CoordinatorClient(CoordinatorInstance *coord_instance, CoordinatorClientConfig config,
HealthCheckCallback succ_cb, HealthCheckCallback fail_cb);
HealthCheckClientCallback succ_cb, HealthCheckClientCallback fail_cb);
~CoordinatorClient() = default;
@ -45,16 +46,17 @@ class CoordinatorClient {
void ResumeFrequentCheck();
auto InstanceName() const -> std::string;
auto SocketAddress() const -> std::string;
auto CoordinatorSocketAddress() const -> std::string;
auto ReplicationSocketAddress() const -> std::string;
[[nodiscard]] auto DemoteToReplica() const -> bool;
auto SendPromoteReplicaToMainRpc(const utils::UUID &uuid, ReplicationClientsInfo replication_clients_info) const
auto SendPromoteReplicaToMainRpc(utils::UUID const &uuid, ReplicationClientsInfo replication_clients_info) const
-> bool;
auto SendSwapMainUUIDRpc(const utils::UUID &uuid) const -> bool;
auto SendSwapMainUUIDRpc(utils::UUID const &uuid) const -> bool;
auto SendUnregisterReplicaRpc(std::string const &instance_name) const -> bool;
auto SendUnregisterReplicaRpc(std::string_view instance_name) const -> bool;
auto SendEnableWritingOnMainRpc() const -> bool;
@ -62,7 +64,8 @@ class CoordinatorClient {
auto ReplicationClientInfo() const -> ReplClientInfo;
auto SetCallbacks(HealthCheckCallback succ_cb, HealthCheckCallback fail_cb) -> void;
auto SendGetInstanceTimestampsRpc() const
-> utils::BasicResult<GetInstanceUUIDError, replication_coordination_glue::DatabaseHistories>;
auto RpcClient() -> rpc::Client & { return rpc_client_; }
@ -82,8 +85,8 @@ class CoordinatorClient {
CoordinatorClientConfig config_;
CoordinatorInstance *coord_instance_;
HealthCheckCallback succ_cb_;
HealthCheckCallback fail_cb_;
HealthCheckClientCallback succ_cb_;
HealthCheckClientCallback fail_cb_;
};
} // namespace memgraph::coordination

16
src/coordination/include/coordination/coordinator_config.hpp
View File

@ -14,12 +14,16 @@
#ifdef MG_ENTERPRISE
#include "replication_coordination_glue/mode.hpp"
#include "utils/string.hpp"
#include <chrono>
#include <cstdint>
#include <optional>
#include <string>
#include <fmt/format.h>
#include "json/json.hpp"
namespace memgraph::coordination {
inline constexpr auto *kDefaultReplicationServerIp = "0.0.0.0";
@ -32,7 +36,11 @@ struct CoordinatorClientConfig {
std::chrono::seconds instance_down_timeout_sec{5};
std::chrono::seconds instance_get_uuid_frequency_sec{10};
auto SocketAddress() const -> std::string { return ip_address + ":" + std::to_string(port); }
auto CoordinatorSocketAddress() const -> std::string { return fmt::format("{}:{}", ip_address, port); }
auto ReplicationSocketAddress() const -> std::string {
return fmt::format("{}:{}", replication_client_info.replication_ip_address,
replication_client_info.replication_port);
}
struct ReplicationClientInfo {
std::string instance_name;
@ -75,5 +83,11 @@ struct CoordinatorServerConfig {
friend bool operator==(CoordinatorServerConfig const &, CoordinatorServerConfig const &) = default;
};
void to_json(nlohmann::json &j, CoordinatorClientConfig const &config);
void from_json(nlohmann::json const &j, CoordinatorClientConfig &config);
void to_json(nlohmann::json &j, ReplClientInfo const &config);
void from_json(nlohmann::json const &j, ReplClientInfo &config);
} // namespace memgraph::coordination
#endif

11
src/coordination/include/coordination/coordinator_exceptions.hpp
View File

@ -83,5 +83,16 @@ class RaftCouldNotParseFlagsException final : public utils::BasicException {
SPECIALIZE_GET_EXCEPTION_NAME(RaftCouldNotParseFlagsException)
};
class InvalidRaftLogActionException final : public utils::BasicException {
public:
explicit InvalidRaftLogActionException(std::string_view what) noexcept : BasicException(what) {}
template <class... Args>
explicit InvalidRaftLogActionException(fmt::format_string<Args...> fmt, Args &&...args) noexcept
: InvalidRaftLogActionException(fmt::format(fmt, std::forward<Args>(args)...)) {}
SPECIALIZE_GET_EXCEPTION_NAME(InvalidRaftLogActionException)
};
} // namespace memgraph::coordination
#endif

3
src/coordination/include/coordination/coordinator_handlers.hpp
View File

@ -41,6 +41,9 @@ class CoordinatorHandlers {
static void GetInstanceUUIDHandler(replication::ReplicationHandler &replication_handler, slk::Reader *req_reader,
slk::Builder *res_builder);
static void GetDatabaseHistoriesHandler(replication::ReplicationHandler &replication_handler, slk::Reader *req_reader,
slk::Builder *res_builder);
};
} // namespace memgraph::dbms

46
src/coordination/include/coordination/coordinator_instance.hpp
View File

@ -18,6 +18,7 @@
#include "coordination/raft_state.hpp"
#include "coordination/register_main_replica_coordinator_status.hpp"
#include "coordination/replication_instance.hpp"
#include "utils/resource_lock.hpp"
#include "utils/rw_lock.hpp"
#include "utils/thread_pool.hpp"
@ -25,33 +26,54 @@
namespace memgraph::coordination {
struct NewMainRes {
std::string most_up_to_date_instance;
std::string latest_epoch;
uint64_t latest_commit_timestamp;
};
using InstanceNameDbHistories = std::pair<std::string, replication_coordination_glue::DatabaseHistories>;
class CoordinatorInstance {
public:
CoordinatorInstance();
[[nodiscard]] auto RegisterReplicationInstance(CoordinatorClientConfig config) -> RegisterInstanceCoordinatorStatus;
[[nodiscard]] auto UnregisterReplicationInstance(std::string instance_name) -> UnregisterInstanceCoordinatorStatus;
[[nodiscard]] auto RegisterReplicationInstance(CoordinatorClientConfig const &config)
-> RegisterInstanceCoordinatorStatus;
[[nodiscard]] auto UnregisterReplicationInstance(std::string_view instance_name)
-> UnregisterInstanceCoordinatorStatus;
[[nodiscard]] auto SetReplicationInstanceToMain(std::string instance_name) -> SetInstanceToMainCoordinatorStatus;
[[nodiscard]] auto SetReplicationInstanceToMain(std::string_view instance_name) -> SetInstanceToMainCoordinatorStatus;
auto ShowInstances() const -> std::vector<InstanceStatus>;
auto TryFailover() -> void;
auto AddCoordinatorInstance(uint32_t raft_server_id, uint32_t raft_port, std::string raft_address) -> void;
auto AddCoordinatorInstance(uint32_t raft_server_id, uint32_t raft_port, std::string_view raft_address) -> void;
auto GetMainUUID() const -> utils::UUID;
auto SetMainUUID(utils::UUID new_uuid) -> void;
static auto ChooseMostUpToDateInstance(std::span<InstanceNameDbHistories> histories) -> NewMainRes;
private:
HealthCheckCallback main_succ_cb_, main_fail_cb_, replica_succ_cb_, replica_fail_cb_;
HealthCheckClientCallback client_succ_cb_, client_fail_cb_;
// NOTE: Must be std::list because we rely on pointer stability
auto OnRaftCommitCallback(TRaftLog const &log_entry, RaftLogAction log_action) -> void;
auto FindReplicationInstance(std::string_view replication_instance_name) -> ReplicationInstance &;
void MainFailCallback(std::string_view);
void MainSuccessCallback(std::string_view);
void ReplicaSuccessCallback(std::string_view);
void ReplicaFailCallback(std::string_view);
auto IsMain(std::string_view instance_name) const -> bool;
auto IsReplica(std::string_view instance_name) const -> bool;
// NOTE: Must be std::list because we rely on pointer stability.
// Leader and followers should both have same view on repl_instances_
std::list<ReplicationInstance> repl_instances_;
mutable utils::RWLock coord_instance_lock_{utils::RWLock::Priority::READ};
utils::UUID main_uuid_;
mutable utils::ResourceLock coord_instance_lock_{};
RaftState raft_state_;
};

35
src/coordination/include/coordination/coordinator_rpc.hpp
View File

@ -15,6 +15,7 @@
#ifdef MG_ENTERPRISE
#include "coordination/coordinator_config.hpp"
#include "replication_coordination_glue/common.hpp"
#include "rpc/messages.hpp"
#include "slk/serialization.hpp"
@ -89,7 +90,7 @@ struct UnregisterReplicaReq {
static void Load(UnregisterReplicaReq *self, memgraph::slk::Reader *reader);
static void Save(UnregisterReplicaReq const &self, memgraph::slk::Builder *builder);
explicit UnregisterReplicaReq(std::string instance_name) : instance_name(std::move(instance_name)) {}
explicit UnregisterReplicaReq(std::string_view inst_name) : instance_name(inst_name) {}
UnregisterReplicaReq() = default;
@ -161,6 +162,32 @@ struct GetInstanceUUIDRes {
using GetInstanceUUIDRpc = rpc::RequestResponse<GetInstanceUUIDReq, GetInstanceUUIDRes>;
struct GetDatabaseHistoriesReq {
static const utils::TypeInfo kType;
static const utils::TypeInfo &GetTypeInfo() { return kType; }
static void Load(GetDatabaseHistoriesReq *self, memgraph::slk::Reader *reader);
static void Save(const GetDatabaseHistoriesReq &self, memgraph::slk::Builder *builder);
GetDatabaseHistoriesReq() = default;
};
struct GetDatabaseHistoriesRes {
static const utils::TypeInfo kType;
static const utils::TypeInfo &GetTypeInfo() { return kType; }
static void Load(GetDatabaseHistoriesRes *self, memgraph::slk::Reader *reader);
static void Save(const GetDatabaseHistoriesRes &self, memgraph::slk::Builder *builder);
explicit GetDatabaseHistoriesRes(const replication_coordination_glue::DatabaseHistories &database_histories)
: database_histories(database_histories) {}
GetDatabaseHistoriesRes() = default;
replication_coordination_glue::DatabaseHistories database_histories;
};
using GetDatabaseHistoriesRpc = rpc::RequestResponse<GetDatabaseHistoriesReq, GetDatabaseHistoriesRes>;
} // namespace memgraph::coordination
// SLK serialization declarations
@ -183,15 +210,21 @@ void Save(const memgraph::coordination::GetInstanceUUIDReq &self, memgraph::slk:
void Load(memgraph::coordination::GetInstanceUUIDReq *self, memgraph::slk::Reader *reader);
void Save(const memgraph::coordination::GetInstanceUUIDRes &self, memgraph::slk::Builder *builder);
void Load(memgraph::coordination::GetInstanceUUIDRes *self, memgraph::slk::Reader *reader);
// UnregisterReplicaRpc
void Save(memgraph::coordination::UnregisterReplicaRes const &self, memgraph::slk::Builder *builder);
void Load(memgraph::coordination::UnregisterReplicaRes *self, memgraph::slk::Reader *reader);
void Save(memgraph::coordination::UnregisterReplicaReq const &self, memgraph::slk::Builder *builder);
void Load(memgraph::coordination::UnregisterReplicaReq *self, memgraph::slk::Reader *reader);
// EnableWritingOnMainRpc
void Save(memgraph::coordination::EnableWritingOnMainRes const &self, memgraph::slk::Builder *builder);
void Load(memgraph::coordination::EnableWritingOnMainRes *self, memgraph::slk::Reader *reader);
// GetDatabaseHistoriesRpc
void Save(const memgraph::coordination::GetDatabaseHistoriesRes &self, memgraph::slk::Builder *builder);
void Load(memgraph::coordination::GetDatabaseHistoriesRes *self, memgraph::slk::Reader *reader);
} // namespace memgraph::slk
#endif

14
src/coordination/include/coordination/coordinator_slk.hpp
View File

@ -14,6 +14,7 @@
#ifdef MG_ENTERPRISE
#include "coordination/coordinator_config.hpp"
#include "replication_coordination_glue/common.hpp"
#include "slk/serialization.hpp"
#include "slk/streams.hpp"
@ -34,5 +35,18 @@ inline void Load(ReplicationClientInfo *obj, Reader *reader) {
Load(&obj->replication_ip_address, reader);
Load(&obj->replication_port, reader);
}
inline void Save(const replication_coordination_glue::DatabaseHistory &obj, Builder *builder) {
Save(obj.db_uuid, builder);
Save(obj.history, builder);
Save(obj.name, builder);
}
inline void Load(replication_coordination_glue::DatabaseHistory *obj, Reader *reader) {
Load(&obj->db_uuid, reader);
Load(&obj->history, reader);
Load(&obj->name, reader);
}
} // namespace memgraph::slk
#endif

10
src/coordination/include/coordination/coordinator_state.hpp
View File

@ -33,14 +33,16 @@ class CoordinatorState {
CoordinatorState(CoordinatorState &&) noexcept = delete;
CoordinatorState &operator=(CoordinatorState &&) noexcept = delete;
[[nodiscard]] auto RegisterReplicationInstance(CoordinatorClientConfig config) -> RegisterInstanceCoordinatorStatus;
[[nodiscard]] auto UnregisterReplicationInstance(std::string instance_name) -> UnregisterInstanceCoordinatorStatus;
[[nodiscard]] auto RegisterReplicationInstance(CoordinatorClientConfig const &config)
-> RegisterInstanceCoordinatorStatus;
[[nodiscard]] auto UnregisterReplicationInstance(std::string_view instance_name)
-> UnregisterInstanceCoordinatorStatus;
[[nodiscard]] auto SetReplicationInstanceToMain(std::string instance_name) -> SetInstanceToMainCoordinatorStatus;
[[nodiscard]] auto SetReplicationInstanceToMain(std::string_view instance_name) -> SetInstanceToMainCoordinatorStatus;
auto ShowInstances() const -> std::vector<InstanceStatus>;
auto AddCoordinatorInstance(uint32_t raft_server_id, uint32_t raft_port, std::string raft_address) -> void;
auto AddCoordinatorInstance(uint32_t raft_server_id, uint32_t raft_port, std::string_view raft_address) -> void;
// NOTE: The client code must check that the server exists before calling this method.
auto GetCoordinatorServer() const -> CoordinatorServer &;

2
src/coordination/include/coordination/instance_status.hpp
View File

@ -26,7 +26,7 @@ struct InstanceStatus {
std::string raft_socket_address;
std::string coord_socket_address;
std::string cluster_role;
bool is_alive;
std::string health;
};
} // namespace memgraph::coordination

34
src/coordination/include/coordination/raft_state.hpp
View File

@ -14,11 +14,17 @@
#ifdef MG_ENTERPRISE
#include <flags/replication.hpp>
#include "io/network/endpoint.hpp"
#include "nuraft/coordinator_state_machine.hpp"
#include "nuraft/coordinator_state_manager.hpp"
#include <libnuraft/nuraft.hxx>
namespace memgraph::coordination {
class CoordinatorInstance;
struct CoordinatorClientConfig;
using BecomeLeaderCb = std::function<void()>;
using BecomeFollowerCb = std::function<void()>;
@ -47,26 +53,38 @@ class RaftState {
RaftState &operator=(RaftState &&other) noexcept = default;
~RaftState();
static auto MakeRaftState(BecomeLeaderCb become_leader_cb, BecomeFollowerCb become_follower_cb) -> RaftState;
static auto MakeRaftState(BecomeLeaderCb &&become_leader_cb, BecomeFollowerCb &&become_follower_cb) -> RaftState;
auto InstanceName() const -> std::string;
auto RaftSocketAddress() const -> std::string;
auto AddCoordinatorInstance(uint32_t raft_server_id, uint32_t raft_port, std::string raft_address) -> void;
auto AddCoordinatorInstance(uint32_t raft_server_id, uint32_t raft_port, std::string_view raft_address) -> void;
auto GetAllCoordinators() const -> std::vector<ptr<srv_config>>;
auto RequestLeadership() -> bool;
auto IsLeader() const -> bool;
auto AppendRegisterReplicationInstance(std::string const &instance) -> ptr<raft_result>;
auto FindCurrentMainInstanceName() const -> std::optional<std::string>;
auto MainExists() const -> bool;
auto IsMain(std::string_view instance_name) const -> bool;
auto IsReplica(std::string_view instance_name) const -> bool;
// TODO: (andi) I think variables below can be abstracted
auto AppendRegisterReplicationInstanceLog(CoordinatorClientConfig const &config) -> bool;
auto AppendUnregisterReplicationInstanceLog(std::string_view instance_name) -> bool;
auto AppendSetInstanceAsMainLog(std::string_view instance_name) -> bool;
auto AppendSetInstanceAsReplicaLog(std::string_view instance_name) -> bool;
auto AppendUpdateUUIDLog(utils::UUID const &uuid) -> bool;
auto GetInstances() const -> std::vector<InstanceState>;
auto GetUUID() const -> utils::UUID;
private:
// TODO: (andi) I think variables below can be abstracted/clean them.
io::network::Endpoint raft_endpoint_;
uint32_t raft_server_id_;
uint32_t raft_port_;
std::string raft_address_;
ptr<state_machine> state_machine_;
ptr<state_mgr> state_manager_;
ptr<CoordinatorStateMachine> state_machine_;
ptr<CoordinatorStateManager> state_manager_;
ptr<raft_server> raft_server_;
ptr<logger> logger_;
raft_launcher launcher_;

17
src/coordination/include/coordination/register_main_replica_coordinator_status.hpp
View File

@ -19,12 +19,12 @@ namespace memgraph::coordination {
enum class RegisterInstanceCoordinatorStatus : uint8_t {
NAME_EXISTS,
ENDPOINT_EXISTS,
COORD_ENDPOINT_EXISTS,
REPL_ENDPOINT_EXISTS,
NOT_COORDINATOR,
RPC_FAILED,
NOT_LEADER,
RAFT_COULD_NOT_ACCEPT,
RAFT_COULD_NOT_APPEND,
RPC_FAILED,
RAFT_LOG_ERROR,
SUCCESS
};
@ -32,8 +32,9 @@ enum class UnregisterInstanceCoordinatorStatus : uint8_t {
NO_INSTANCE_WITH_NAME,
IS_MAIN,
NOT_COORDINATOR,
NOT_LEADER,
RPC_FAILED,
NOT_LEADER,
RAFT_LOG_ERROR,
SUCCESS,
};
@ -41,9 +42,11 @@ enum class SetInstanceToMainCoordinatorStatus : uint8_t {
NO_INSTANCE_WITH_NAME,
MAIN_ALREADY_EXISTS,
NOT_COORDINATOR,
SUCCESS,
NOT_LEADER,
RAFT_LOG_ERROR,
COULD_NOT_PROMOTE_TO_MAIN,
SWAP_UUID_FAILED
SWAP_UUID_FAILED,
SUCCESS,
};
} // namespace memgraph::coordination

43
src/coordination/include/coordination/replication_instance.hpp
View File

@ -17,18 +17,24 @@
#include "coordination/coordinator_exceptions.hpp"
#include "replication_coordination_glue/role.hpp"
#include <libnuraft/nuraft.hxx>
#include "utils/resource_lock.hpp"
#include "utils/result.hpp"
#include "utils/uuid.hpp"
#include <libnuraft/nuraft.hxx>
namespace memgraph::coordination {
class CoordinatorInstance;
class ReplicationInstance;
using HealthCheckInstanceCallback = void (CoordinatorInstance::*)(std::string_view);
class ReplicationInstance {
public:
ReplicationInstance(CoordinatorInstance *peer, CoordinatorClientConfig config, HealthCheckCallback succ_cb,
HealthCheckCallback fail_cb);
ReplicationInstance(CoordinatorInstance *peer, CoordinatorClientConfig config, HealthCheckClientCallback succ_cb,
HealthCheckClientCallback fail_cb, HealthCheckInstanceCallback succ_instance_cb,
HealthCheckInstanceCallback fail_instance_cb);
ReplicationInstance(ReplicationInstance const &other) = delete;
ReplicationInstance &operator=(ReplicationInstance const &other) = delete;
@ -45,14 +51,16 @@ class ReplicationInstance {
auto IsAlive() const -> bool;
auto InstanceName() const -> std::string;
auto SocketAddress() const -> std::string;
auto CoordinatorSocketAddress() const -> std::string;
auto ReplicationSocketAddress() const -> std::string;
auto IsReplica() const -> bool;
auto IsMain() const -> bool;
auto PromoteToMain(utils::UUID const &uuid, ReplicationClientsInfo repl_clients_info,
HealthCheckInstanceCallback main_succ_cb, HealthCheckInstanceCallback main_fail_cb) -> bool;
auto PromoteToMain(utils::UUID uuid, ReplicationClientsInfo repl_clients_info, HealthCheckCallback main_succ_cb,
HealthCheckCallback main_fail_cb) -> bool;
auto DemoteToReplica(HealthCheckCallback replica_succ_cb, HealthCheckCallback replica_fail_cb) -> bool;
auto SendDemoteToReplicaRpc() -> bool;
auto DemoteToReplica(HealthCheckInstanceCallback replica_succ_cb, HealthCheckInstanceCallback replica_fail_cb)
-> bool;
auto StartFrequentCheck() -> void;
auto StopFrequentCheck() -> void;
@ -63,9 +71,8 @@ class ReplicationInstance {
auto EnsureReplicaHasCorrectMainUUID(utils::UUID const &curr_main_uuid) -> bool;
auto SendSwapAndUpdateUUID(const utils::UUID &new_main_uuid) -> bool;
auto SendUnregisterReplicaRpc(std::string const &instance_name) -> bool;
auto SendSwapAndUpdateUUID(utils::UUID const &new_main_uuid) -> bool;
auto SendUnregisterReplicaRpc(std::string_view instance_name) -> bool;
auto SendGetInstanceUUID() -> utils::BasicResult<coordination::GetInstanceUUIDError, std::optional<utils::UUID>>;
auto GetClient() -> CoordinatorClient &;
@ -74,11 +81,13 @@ class ReplicationInstance {
auto SetNewMainUUID(utils::UUID const &main_uuid) -> void;
auto ResetMainUUID() -> void;
auto GetMainUUID() const -> const std::optional<utils::UUID> &;
auto GetMainUUID() const -> std::optional<utils::UUID> const &;
auto GetSuccessCallback() -> HealthCheckInstanceCallback &;
auto GetFailCallback() -> HealthCheckInstanceCallback &;
private:
CoordinatorClient client_;
replication_coordination_glue::ReplicationRole replication_role_;
std::chrono::system_clock::time_point last_response_time_{};
bool is_alive_{false};
std::chrono::system_clock::time_point last_check_of_uuid_{};
@ -90,8 +99,12 @@ class ReplicationInstance {
// so we need to send swap uuid again
std::optional<utils::UUID> main_uuid_;
HealthCheckInstanceCallback succ_cb_;
HealthCheckInstanceCallback fail_cb_;
friend bool operator==(ReplicationInstance const &first, ReplicationInstance const &second) {
return first.client_ == second.client_ && first.replication_role_ == second.replication_role_;
return first.client_ == second.client_ && first.last_response_time_ == second.last_response_time_ &&
first.is_alive_ == second.is_alive_ && first.main_uuid_ == second.main_uuid_;
}
};

1
src/coordination/include/coordination/rpc_errors.hpp
View File

@ -11,4 +11,5 @@
namespace memgraph::coordination {
enum class GetInstanceUUIDError { NO_RESPONSE, RPC_EXCEPTION };
enum class GetInstanceTimestampsError { NO_RESPONSE, RPC_EXCEPTION };
} // namespace memgraph::coordination

92
src/coordination/include/nuraft/coordinator_cluster_state.hpp Normal file
View File

@ -0,0 +1,92 @@
// Copyright 2024 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
#ifdef MG_ENTERPRISE
#include "coordination/coordinator_config.hpp"
#include "nuraft/raft_log_action.hpp"
#include "replication_coordination_glue/role.hpp"
#include "utils/resource_lock.hpp"
#include "utils/uuid.hpp"
#include <libnuraft/nuraft.hxx>
#include <range/v3/view.hpp>
#include "json/json.hpp"
#include <map>
#include <numeric>
#include <string>
#include <variant>
namespace memgraph::coordination {
using replication_coordination_glue::ReplicationRole;
struct InstanceState {
CoordinatorClientConfig config;
ReplicationRole status;
friend auto operator==(InstanceState const &lhs, InstanceState const &rhs) -> bool {
return lhs.config == rhs.config && lhs.status == rhs.status;
}
};
void to_json(nlohmann::json &j, InstanceState const &instance_state);
void from_json(nlohmann::json const &j, InstanceState &instance_state);
using TRaftLog = std::variant<CoordinatorClientConfig, std::string, utils::UUID>;
using nuraft::buffer;
using nuraft::buffer_serializer;
using nuraft::ptr;
class CoordinatorClusterState {
public:
CoordinatorClusterState() = default;
explicit CoordinatorClusterState(std::map<std::string, InstanceState, std::less<>> instances);
CoordinatorClusterState(CoordinatorClusterState const &);
CoordinatorClusterState &operator=(CoordinatorClusterState const &);
CoordinatorClusterState(CoordinatorClusterState &&other) noexcept;
CoordinatorClusterState &operator=(CoordinatorClusterState &&other) noexcept;
~CoordinatorClusterState() = default;
auto FindCurrentMainInstanceName() const -> std::optional<std::string>;
auto MainExists() const -> bool;
auto IsMain(std::string_view instance_name) const -> bool;
auto IsReplica(std::string_view instance_name) const -> bool;
auto InsertInstance(std::string instance_name, InstanceState instance_state) -> void;
auto DoAction(TRaftLog log_entry, RaftLogAction log_action) -> void;
auto Serialize(ptr<buffer> &data) -> void;
static auto Deserialize(buffer &data) -> CoordinatorClusterState;
auto GetInstances() const -> std::vector<InstanceState>;
auto GetUUID() const -> utils::UUID;
private:
std::map<std::string, InstanceState, std::less<>> instances_{};
utils::UUID uuid_{};
mutable utils::ResourceLock log_lock_{};
};
} // namespace memgraph::coordination
#endif

38
src/coordination/include/nuraft/coordinator_state_machine.hpp
View File

@ -13,9 +13,15 @@
#ifdef MG_ENTERPRISE
#include "coordination/coordinator_config.hpp"
#include "nuraft/coordinator_cluster_state.hpp"
#include "nuraft/raft_log_action.hpp"
#include <spdlog/spdlog.h>
#include <libnuraft/nuraft.hxx>
#include <variant>
namespace memgraph::coordination {
using nuraft::async_result;
@ -36,9 +42,19 @@ class CoordinatorStateMachine : public state_machine {
CoordinatorStateMachine &operator=(CoordinatorStateMachine &&) = delete;
~CoordinatorStateMachine() override {}
static auto EncodeRegisterReplicationInstance(const std::string &name) -> ptr<buffer>;
auto FindCurrentMainInstanceName() const -> std::optional<std::string>;
auto MainExists() const -> bool;
auto IsMain(std::string_view instance_name) const -> bool;
auto IsReplica(std::string_view instance_name) const -> bool;
static auto DecodeRegisterReplicationInstance(buffer &data) -> std::string;
static auto CreateLog(nlohmann::json &&log) -> ptr<buffer>;
static auto SerializeRegisterInstance(CoordinatorClientConfig const &config) -> ptr<buffer>;
static auto SerializeUnregisterInstance(std::string_view instance_name) -> ptr<buffer>;
static auto SerializeSetInstanceAsMain(std::string_view instance_name) -> ptr<buffer>;
static auto SerializeSetInstanceAsReplica(std::string_view instance_name) -> ptr<buffer>;
static auto SerializeUpdateUUID(utils::UUID const &uuid) -> ptr<buffer>;
static auto DecodeLog(buffer &data) -> std::pair<TRaftLog, RaftLogAction>;
auto pre_commit(ulong log_idx, buffer &data) -> ptr<buffer> override;
@ -64,11 +80,27 @@ class CoordinatorStateMachine : public state_machine {
auto create_snapshot(snapshot &s, async_result<bool>::handler_type &when_done) -> void override;
auto GetInstances() const -> std::vector<InstanceState>;
auto GetUUID() const -> utils::UUID;
private:
struct SnapshotCtx {
SnapshotCtx(ptr<snapshot> &snapshot, CoordinatorClusterState const &cluster_state)
: snapshot_(snapshot), cluster_state_(cluster_state) {}
ptr<snapshot> snapshot_;
CoordinatorClusterState cluster_state_;
};
auto create_snapshot_internal(ptr<snapshot> snapshot) -> void;
CoordinatorClusterState cluster_state_;
std::atomic<uint64_t> last_committed_idx_{0};
ptr<snapshot> last_snapshot_;
std::map<uint64_t, ptr<SnapshotCtx>> snapshots_;
std::mutex snapshots_lock_;
ptr<snapshot> last_snapshot_;
std::mutex last_snapshot_lock_;
};

42
src/coordination/include/nuraft/raft_log_action.hpp Normal file
View File

@ -0,0 +1,42 @@
// Copyright 2024 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
#ifdef MG_ENTERPRISE
#include "coordination/coordinator_exceptions.hpp"
#include <cstdint>
#include <string>
#include "json/json.hpp"
namespace memgraph::coordination {
enum class RaftLogAction : uint8_t {
REGISTER_REPLICATION_INSTANCE,
UNREGISTER_REPLICATION_INSTANCE,
SET_INSTANCE_AS_MAIN,
SET_INSTANCE_AS_REPLICA,
UPDATE_UUID
};
NLOHMANN_JSON_SERIALIZE_ENUM(RaftLogAction, {
{RaftLogAction::REGISTER_REPLICATION_INSTANCE, "register"},
{RaftLogAction::UNREGISTER_REPLICATION_INSTANCE, "unregister"},
{RaftLogAction::SET_INSTANCE_AS_MAIN, "promote"},
{RaftLogAction::SET_INSTANCE_AS_REPLICA, "demote"},
{RaftLogAction::UPDATE_UUID, "update_uuid"},
})
} // namespace memgraph::coordination
#endif

209
src/coordination/raft_state.cpp
View File

@ -10,12 +10,11 @@
// licenses/APL.txt.
#ifdef MG_ENTERPRISE
#include <chrono>
#include "coordination/raft_state.hpp"
#include "coordination/coordinator_config.hpp"
#include "coordination/coordinator_exceptions.hpp"
#include "nuraft/coordinator_state_machine.hpp"
#include "nuraft/coordinator_state_manager.hpp"
#include "coordination/raft_state.hpp"
#include "utils/counter.hpp"
namespace memgraph::coordination {
@ -33,31 +32,35 @@ using raft_result = cmd_result<ptr<buffer>>;
RaftState::RaftState(BecomeLeaderCb become_leader_cb, BecomeFollowerCb become_follower_cb, uint32_t raft_server_id,
uint32_t raft_port, std::string raft_address)
: raft_server_id_(raft_server_id),
raft_port_(raft_port),
raft_address_(std::move(raft_address)),
: raft_endpoint_(raft_address, raft_port),
raft_server_id_(raft_server_id),
state_machine_(cs_new<CoordinatorStateMachine>()),
state_manager_(
cs_new<CoordinatorStateManager>(raft_server_id_, raft_address_ + ":" + std::to_string(raft_port_))),
state_manager_(cs_new<CoordinatorStateManager>(raft_server_id_, raft_endpoint_.SocketAddress())),
logger_(nullptr),
become_leader_cb_(std::move(become_leader_cb)),
become_follower_cb_(std::move(become_follower_cb)) {}
auto RaftState::InitRaftServer() -> void {
asio_service::options asio_opts;
asio_opts.thread_pool_size_ = 1; // TODO: (andi) Improve this
asio_opts.thread_pool_size_ = 1;
raft_params params;
params.heart_beat_interval_ = 100;
params.election_timeout_lower_bound_ = 200;
params.election_timeout_upper_bound_ = 400;
// 5 logs are preserved before the last snapshot
params.reserved_log_items_ = 5;
// Create snapshot for every 5 log appends
params.snapshot_distance_ = 5;
params.client_req_timeout_ = 3000;
params.return_method_ = raft_params::blocking;
// If the leader doesn't receive any response from quorum nodes
// in 200ms, it will step down.
// This allows us to achieve strong consistency even if network partition
// happens between the current leader and followers.
// The value must be <= election_timeout_lower_bound_ so that cluster can never
// have multiple leaders.
params.leadership_expiry_ = 200;
raft_server::init_options init_opts;
init_opts.raft_callback_ = [this](cb_func::Type event_type, cb_func::Param *param) -> nuraft::CbReturnCode {
if (event_type == cb_func::BecomeLeader) {
@ -72,11 +75,11 @@ auto RaftState::InitRaftServer() -> void {
raft_launcher launcher;
raft_server_ = launcher.init(state_machine_, state_manager_, logger_, static_cast<int>(raft_port_), asio_opts, params,
init_opts);
raft_server_ =
launcher.init(state_machine_, state_manager_, logger_, raft_endpoint_.port, asio_opts, params, init_opts);
if (!raft_server_) {
throw RaftServerStartException("Failed to launch raft server on {}:{}", raft_address_, raft_port_);
throw RaftServerStartException("Failed to launch raft server on {}", raft_endpoint_.SocketAddress());
}
auto maybe_stop = utils::ResettableCounter<20>();
@ -87,38 +90,61 @@ auto RaftState::InitRaftServer() -> void {
std::this_thread::sleep_for(std::chrono::milliseconds(250));
} while (!maybe_stop());
throw RaftServerStartException("Failed to initialize raft server on {}:{}", raft_address_, raft_port_);
throw RaftServerStartException("Failed to initialize raft server on {}", raft_endpoint_.SocketAddress());
}
auto RaftState::MakeRaftState(BecomeLeaderCb become_leader_cb, BecomeFollowerCb become_follower_cb) -> RaftState {
uint32_t raft_server_id{0};
uint32_t raft_port{0};
try {
raft_server_id = FLAGS_raft_server_id;
raft_port = FLAGS_raft_server_port;
} catch (std::exception const &e) {
throw RaftCouldNotParseFlagsException("Failed to parse flags: {}", e.what());
}
auto RaftState::MakeRaftState(BecomeLeaderCb &&become_leader_cb, BecomeFollowerCb &&become_follower_cb) -> RaftState {
uint32_t raft_server_id = FLAGS_raft_server_id;
uint32_t raft_port = FLAGS_raft_server_port;
auto raft_state =
RaftState(std::move(become_leader_cb), std::move(become_follower_cb), raft_server_id, raft_port, "127.0.0.1");
raft_state.InitRaftServer();
return raft_state;
}
RaftState::~RaftState() { launcher_.shutdown(); }
auto RaftState::InstanceName() const -> std::string { return "coordinator_" + std::to_string(raft_server_id_); }
auto RaftState::InstanceName() const -> std::string {
return fmt::format("coordinator_{}", std::to_string(raft_server_id_));
}
auto RaftState::RaftSocketAddress() const -> std::string { return raft_address_ + ":" + std::to_string(raft_port_); }
auto RaftState::RaftSocketAddress() const -> std::string { return raft_endpoint_.SocketAddress(); }
auto RaftState::AddCoordinatorInstance(uint32_t raft_server_id, uint32_t raft_port, std::string raft_address) -> void {
auto const endpoint = raft_address + ":" + std::to_string(raft_port);
auto RaftState::AddCoordinatorInstance(uint32_t raft_server_id, uint32_t raft_port, std::string_view raft_address)
-> void {
auto const endpoint = fmt::format("{}:{}", raft_address, raft_port);
srv_config const srv_config_to_add(static_cast<int>(raft_server_id), endpoint);
if (!raft_server_->add_srv(srv_config_to_add)->get_accepted()) {
throw RaftAddServerException("Failed to add server {} to the cluster", endpoint);
auto cmd_result = raft_server_->add_srv(srv_config_to_add);
if (cmd_result->get_result_code() == nuraft::cmd_result_code::OK) {
spdlog::info("Request to add server {} to the cluster accepted", endpoint);
} else {
throw RaftAddServerException("Failed to accept request to add server {} to the cluster with error code {}",
endpoint, cmd_result->get_result_code());
}
// Waiting for server to join
constexpr int max_tries{10};
auto maybe_stop = utils::ResettableCounter<max_tries>();
constexpr int waiting_period{200};
bool added{false};
while (!maybe_stop()) {
std::this_thread::sleep_for(std::chrono::milliseconds(waiting_period));
const auto server_config = raft_server_->get_srv_config(static_cast<nuraft::int32>(raft_server_id));
if (server_config) {
spdlog::trace("Server with id {} added to cluster", raft_server_id);
added = true;
break;
}
}
if (!added) {
throw RaftAddServerException("Failed to add server {} to the cluster in {}ms", endpoint,
max_tries * waiting_period);
}
spdlog::info("Request to add server {} to the cluster accepted", endpoint);
}
auto RaftState::GetAllCoordinators() const -> std::vector<ptr<srv_config>> {
@ -131,10 +157,123 @@ auto RaftState::IsLeader() const -> bool { return raft_server_->is_leader(); }
auto RaftState::RequestLeadership() -> bool { return raft_server_->is_leader() || raft_server_->request_leadership(); }
auto RaftState::AppendRegisterReplicationInstance(std::string const &instance) -> ptr<raft_result> {
auto new_log = CoordinatorStateMachine::EncodeRegisterReplicationInstance(instance);
return raft_server_->append_entries({new_log});
auto RaftState::AppendRegisterReplicationInstanceLog(CoordinatorClientConfig const &config) -> bool {
auto new_log = CoordinatorStateMachine::SerializeRegisterInstance(config);
auto const res = raft_server_->append_entries({new_log});
if (!res->get_accepted()) {
spdlog::error(
"Failed to accept request for registering instance {}. Most likely the reason is that the instance is not "
"the "
"leader.",
config.instance_name);
return false;
}
spdlog::info("Request for registering instance {} accepted", config.instance_name);
if (res->get_result_code() != nuraft::cmd_result_code::OK) {
spdlog::error("Failed to register instance {} with error code {}", config.instance_name, res->get_result_code());
return false;
}
return true;
}
auto RaftState::AppendUnregisterReplicationInstanceLog(std::string_view instance_name) -> bool {
auto new_log = CoordinatorStateMachine::SerializeUnregisterInstance(instance_name);
auto const res = raft_server_->append_entries({new_log});
if (!res->get_accepted()) {
spdlog::error(
"Failed to accept request for unregistering instance {}. Most likely the reason is that the instance is not "
"the leader.",
instance_name);
return false;
}
spdlog::info("Request for unregistering instance {} accepted", instance_name);
if (res->get_result_code() != nuraft::cmd_result_code::OK) {
spdlog::error("Failed to unregister instance {} with error code {}", instance_name, res->get_result_code());
return false;
}
return true;
}
auto RaftState::AppendSetInstanceAsMainLog(std::string_view instance_name) -> bool {
auto new_log = CoordinatorStateMachine::SerializeSetInstanceAsMain(instance_name);
auto const res = raft_server_->append_entries({new_log});
if (!res->get_accepted()) {
spdlog::error(
"Failed to accept request for promoting instance {}. Most likely the reason is that the instance is not "
"the leader.",
instance_name);
return false;
}
spdlog::info("Request for promoting instance {} accepted", instance_name);
if (res->get_result_code() != nuraft::cmd_result_code::OK) {
spdlog::error("Failed to promote instance {} with error code {}", instance_name, res->get_result_code());
return false;
}
return true;
}
auto RaftState::AppendSetInstanceAsReplicaLog(std::string_view instance_name) -> bool {
auto new_log = CoordinatorStateMachine::SerializeSetInstanceAsReplica(instance_name);
auto const res = raft_server_->append_entries({new_log});
if (!res->get_accepted()) {
spdlog::error(
"Failed to accept request for demoting instance {}. Most likely the reason is that the instance is not "
"the leader.",
instance_name);
return false;
}
spdlog::info("Request for demoting instance {} accepted", instance_name);
if (res->get_result_code() != nuraft::cmd_result_code::OK) {
spdlog::error("Failed to promote instance {} with error code {}", instance_name, res->get_result_code());
return false;
}
return true;
}
auto RaftState::AppendUpdateUUIDLog(utils::UUID const &uuid) -> bool {
auto new_log = CoordinatorStateMachine::SerializeUpdateUUID(uuid);
auto const res = raft_server_->append_entries({new_log});
if (!res->get_accepted()) {
spdlog::error(
"Failed to accept request for updating UUID. Most likely the reason is that the instance is not "
"the leader.");
return false;
}
spdlog::info("Request for updating UUID accepted");
if (res->get_result_code() != nuraft::cmd_result_code::OK) {
spdlog::error("Failed to update UUID with error code {}", res->get_result_code());
return false;
}
return true;
}
auto RaftState::FindCurrentMainInstanceName() const -> std::optional<std::string> {
return state_machine_->FindCurrentMainInstanceName();
}
auto RaftState::MainExists() const -> bool { return state_machine_->MainExists(); }
auto RaftState::IsMain(std::string_view instance_name) const -> bool { return state_machine_->IsMain(instance_name); }
auto RaftState::IsReplica(std::string_view instance_name) const -> bool {
return state_machine_->IsReplica(instance_name);
}
auto RaftState::GetInstances() const -> std::vector<InstanceState> { return state_machine_->GetInstances(); }
auto RaftState::GetUUID() const -> utils::UUID { return state_machine_->GetUUID(); }
} // namespace memgraph::coordination
#endif

53
src/coordination/replication_instance.cpp
View File

@ -13,21 +13,20 @@
#include "coordination/replication_instance.hpp"
#include <utility>
#include "replication_coordination_glue/handler.hpp"
#include "utils/result.hpp"
namespace memgraph::coordination {
ReplicationInstance::ReplicationInstance(CoordinatorInstance *peer, CoordinatorClientConfig config,
HealthCheckCallback succ_cb, HealthCheckCallback fail_cb)
HealthCheckClientCallback succ_cb, HealthCheckClientCallback fail_cb,
HealthCheckInstanceCallback succ_instance_cb,
HealthCheckInstanceCallback fail_instance_cb)
: client_(peer, std::move(config), std::move(succ_cb), std::move(fail_cb)),
replication_role_(replication_coordination_glue::ReplicationRole::REPLICA) {
if (!client_.DemoteToReplica()) {
throw CoordinatorRegisterInstanceException("Failed to demote instance {} to replica", client_.InstanceName());
}
client_.StartFrequentCheck();
}
succ_cb_(succ_instance_cb),
fail_cb_(fail_instance_cb) {}
auto ReplicationInstance::OnSuccessPing() -> void {
last_response_time_ = std::chrono::system_clock::now();
@ -46,37 +45,34 @@ auto ReplicationInstance::IsReadyForUUIDPing() -> bool {
}
auto ReplicationInstance::InstanceName() const -> std::string { return client_.InstanceName(); }
auto ReplicationInstance::SocketAddress() const -> std::string { return client_.SocketAddress(); }
auto ReplicationInstance::CoordinatorSocketAddress() const -> std::string { return client_.CoordinatorSocketAddress(); }
auto ReplicationInstance::ReplicationSocketAddress() const -> std::string { return client_.ReplicationSocketAddress(); }
auto ReplicationInstance::IsAlive() const -> bool { return is_alive_; }
auto ReplicationInstance::IsReplica() const -> bool {
return replication_role_ == replication_coordination_glue::ReplicationRole::REPLICA;
}
auto ReplicationInstance::IsMain() const -> bool {
return replication_role_ == replication_coordination_glue::ReplicationRole::MAIN;
}
auto ReplicationInstance::PromoteToMain(utils::UUID new_uuid, ReplicationClientsInfo repl_clients_info,
HealthCheckCallback main_succ_cb, HealthCheckCallback main_fail_cb) -> bool {
auto ReplicationInstance::PromoteToMain(utils::UUID const &new_uuid, ReplicationClientsInfo repl_clients_info,
HealthCheckInstanceCallback main_succ_cb,
HealthCheckInstanceCallback main_fail_cb) -> bool {
if (!client_.SendPromoteReplicaToMainRpc(new_uuid, std::move(repl_clients_info))) {
return false;
}
replication_role_ = replication_coordination_glue::ReplicationRole::MAIN;
main_uuid_ = new_uuid;
client_.SetCallbacks(std::move(main_succ_cb), std::move(main_fail_cb));
succ_cb_ = main_succ_cb;
fail_cb_ = main_fail_cb;
return true;
}
auto ReplicationInstance::DemoteToReplica(HealthCheckCallback replica_succ_cb, HealthCheckCallback replica_fail_cb)
-> bool {
auto ReplicationInstance::SendDemoteToReplicaRpc() -> bool { return client_.DemoteToReplica(); }
auto ReplicationInstance::DemoteToReplica(HealthCheckInstanceCallback replica_succ_cb,
HealthCheckInstanceCallback replica_fail_cb) -> bool {
if (!client_.DemoteToReplica()) {
return false;
}
replication_role_ = replication_coordination_glue::ReplicationRole::REPLICA;
client_.SetCallbacks(std::move(replica_succ_cb), std::move(replica_fail_cb));
succ_cb_ = replica_succ_cb;
fail_cb_ = replica_fail_cb;
return true;
}
@ -90,10 +86,12 @@ auto ReplicationInstance::ReplicationClientInfo() const -> CoordinatorClientConf
return client_.ReplicationClientInfo();
}
auto ReplicationInstance::GetSuccessCallback() -> HealthCheckInstanceCallback & { return succ_cb_; }
auto ReplicationInstance::GetFailCallback() -> HealthCheckInstanceCallback & { return fail_cb_; }
auto ReplicationInstance::GetClient() -> CoordinatorClient & { return client_; }
auto ReplicationInstance::SetNewMainUUID(utils::UUID const &main_uuid) -> void { main_uuid_ = main_uuid; }
auto ReplicationInstance::ResetMainUUID() -> void { main_uuid_ = std::nullopt; }
auto ReplicationInstance::GetMainUUID() const -> std::optional<utils::UUID> const & { return main_uuid_; }
auto ReplicationInstance::EnsureReplicaHasCorrectMainUUID(utils::UUID const &curr_main_uuid) -> bool {
@ -106,6 +104,7 @@ auto ReplicationInstance::EnsureReplicaHasCorrectMainUUID(utils::UUID const &cur
}
UpdateReplicaLastResponseUUID();
// NOLINTNEXTLINE
if (res.GetValue().has_value() && res.GetValue().value() == curr_main_uuid) {
return true;
}
@ -113,7 +112,7 @@ auto ReplicationInstance::EnsureReplicaHasCorrectMainUUID(utils::UUID const &cur
return SendSwapAndUpdateUUID(curr_main_uuid);
}
auto ReplicationInstance::SendSwapAndUpdateUUID(const utils::UUID &new_main_uuid) -> bool {
auto ReplicationInstance::SendSwapAndUpdateUUID(utils::UUID const &new_main_uuid) -> bool {
if (!replication_coordination_glue::SendSwapMainUUIDRpc(client_.RpcClient(), new_main_uuid)) {
return false;
}
@ -121,7 +120,7 @@ auto ReplicationInstance::SendSwapAndUpdateUUID(const utils::UUID &new_main_uuid
return true;
}
auto ReplicationInstance::SendUnregisterReplicaRpc(std::string const &instance_name) -> bool {
auto ReplicationInstance::SendUnregisterReplicaRpc(std::string_view instance_name) -> bool {
return client_.SendUnregisterReplicaRpc(instance_name);
}

16
src/dbms/coordinator_handler.cpp
View File

@ -20,28 +20,28 @@ namespace memgraph::dbms {
CoordinatorHandler::CoordinatorHandler(coordination::CoordinatorState &coordinator_state)
: coordinator_state_(coordinator_state) {}
auto CoordinatorHandler::RegisterReplicationInstance(memgraph::coordination::CoordinatorClientConfig config)
auto CoordinatorHandler::RegisterReplicationInstance(coordination::CoordinatorClientConfig const &config)
-> coordination::RegisterInstanceCoordinatorStatus {
return coordinator_state_.RegisterReplicationInstance(config);
}
auto CoordinatorHandler::UnregisterReplicationInstance(std::string instance_name)
auto CoordinatorHandler::UnregisterReplicationInstance(std::string_view instance_name)
-> coordination::UnregisterInstanceCoordinatorStatus {
return coordinator_state_.UnregisterReplicationInstance(std::move(instance_name));
return coordinator_state_.UnregisterReplicationInstance(instance_name);
}
auto CoordinatorHandler::SetReplicationInstanceToMain(std::string instance_name)
auto CoordinatorHandler::SetReplicationInstanceToMain(std::string_view instance_name)
-> coordination::SetInstanceToMainCoordinatorStatus {
return coordinator_state_.SetReplicationInstanceToMain(std::move(instance_name));
return coordinator_state_.SetReplicationInstanceToMain(instance_name);
}
auto CoordinatorHandler::ShowInstances() const -> std::vector<coordination::InstanceStatus> {
return coordinator_state_.ShowInstances();
}
auto CoordinatorHandler::AddCoordinatorInstance(uint32_t raft_server_id, uint32_t raft_port, std::string raft_address)
-> void {
coordinator_state_.AddCoordinatorInstance(raft_server_id, raft_port, std::move(raft_address));
auto CoordinatorHandler::AddCoordinatorInstance(uint32_t raft_server_id, uint32_t raft_port,
std::string_view raft_address) -> void {
coordinator_state_.AddCoordinatorInstance(raft_server_id, raft_port, raft_address);
}
} // namespace memgraph::dbms

9
src/dbms/coordinator_handler.hpp
View File

@ -30,16 +30,17 @@ class CoordinatorHandler {
// TODO: (andi) When moving coordinator state on same instances, rename from RegisterReplicationInstance to
// RegisterInstance
auto RegisterReplicationInstance(coordination::CoordinatorClientConfig config)
auto RegisterReplicationInstance(coordination::CoordinatorClientConfig const &config)
-> coordination::RegisterInstanceCoordinatorStatus;
auto UnregisterReplicationInstance(std::string instance_name) -> coordination::UnregisterInstanceCoordinatorStatus;
auto UnregisterReplicationInstance(std::string_view instance_name)
-> coordination::UnregisterInstanceCoordinatorStatus;
auto SetReplicationInstanceToMain(std::string instance_name) -> coordination::SetInstanceToMainCoordinatorStatus;
auto SetReplicationInstanceToMain(std::string_view instance_name) -> coordination::SetInstanceToMainCoordinatorStatus;
auto ShowInstances() const -> std::vector<coordination::InstanceStatus>;
auto AddCoordinatorInstance(uint32_t raft_server_id, uint32_t raft_port, std::string raft_address) -> void;
auto AddCoordinatorInstance(uint32_t raft_server_id, uint32_t raft_port, std::string_view raft_address) -> void;
private:
coordination::CoordinatorState &coordinator_state_;

4
src/dbms/database.hpp
View File

@ -110,9 +110,9 @@ class Database {
* @param force_directory Use the configured directory, do not try to decipher the multi-db version
* @return DatabaseInfo
*/
DatabaseInfo GetInfo(bool force_directory, replication_coordination_glue::ReplicationRole replication_role) const {
DatabaseInfo GetInfo(replication_coordination_glue::ReplicationRole replication_role) const {
DatabaseInfo info;
info.storage_info = storage_->GetInfo(force_directory, replication_role);
info.storage_info = storage_->GetInfo(replication_role);
info.triggers = trigger_store_.GetTriggerInfo().size();
info.streams = streams_.GetStreamInfo().size();
return info;

2
src/dbms/database_handler.hpp
View File

@ -1,4 +1,4 @@
// Copyright 2023 Memgraph Ltd.
// Copyright 2024 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

13
src/dbms/dbms_handler.cpp
View File

@ -185,6 +185,16 @@ DbmsHandler::DbmsHandler(storage::Config config, replication::ReplicationState &
auto directories = std::set{std::string{kDefaultDB}};
// Recover previous databases
if (flags::AreExperimentsEnabled(flags::Experiments::SYSTEM_REPLICATION) && !recovery_on_startup) {
// This will result in dropping databases on SystemRecoveryHandler
// for MT case, and for single DB case we might not even set replication as commit timestamp is checked
spdlog::warn(
"Data recovery on startup not set, this will result in dropping database in case of multi-tenancy enabled.");
}
// TODO: Problem is if user doesn't set this up "database" name won't be recovered
// but if storage-recover-on-startup is true storage will be recovered which is an issue
spdlog::info("Data recovery on startup set to {}", recovery_on_startup);
if (recovery_on_startup) {
auto it = durability_->begin(std::string(kDBPrefix));
auto end = durability_->end(std::string(kDBPrefix));
@ -410,9 +420,10 @@ void DbmsHandler::UpdateDurability(const storage::Config &config, std::optional<
if (!durability_) return;
// Save database in a list of active databases
const auto &key = Durability::GenKey(config.salient.name);
if (rel_dir == std::nullopt)
if (rel_dir == std::nullopt) {
rel_dir =
std::filesystem::relative(config.durability.storage_directory, default_config_.durability.storage_directory);
}
const auto &val = Durability::GenVal(config.salient.uuid, *rel_dir);
durability_->Put(key, val);
}

18
src/dbms/dbms_handler.hpp
View File

@ -155,6 +155,8 @@ class DbmsHandler {
spdlog::debug("Trying to create db '{}' on replica which already exists.", config.name);
auto db = Get_(config.name);
spdlog::debug("Aligning database with name {} which has UUID {}, where config UUID is {}", config.name,
std::string(db->uuid()), std::string(config.uuid));
if (db->uuid() == config.uuid) { // Same db
return db;
}
@ -163,18 +165,22 @@ class DbmsHandler {
// TODO: Fix this hack
if (config.name == kDefaultDB) {
spdlog::debug("Last commit timestamp for DB {} is {}", kDefaultDB,
db->storage()->repl_storage_state_.last_commit_timestamp_);
// This seems correct, if database made progress
if (db->storage()->repl_storage_state_.last_commit_timestamp_ != storage::kTimestampInitialId) {
spdlog::debug("Default storage is not clean, cannot update UUID...");
return NewError::GENERIC; // Update error
}
spdlog::debug("Update default db's UUID");
spdlog::debug("Updated default db's UUID");
// Default db cannot be deleted and remade, have to just update the UUID
db->storage()->config_.salient.uuid = config.uuid;
UpdateDurability(db->storage()->config_, ".");
return db;
}
spdlog::debug("Drop database and recreate with the correct UUID");
spdlog::debug("Dropping database {} with UUID: {} and recreating with the correct UUID: {}", config.name,
std::string(db->uuid()), std::string(config.uuid));
// Defer drop
(void)Delete_(db->name());
// Second attempt
@ -266,10 +272,6 @@ class DbmsHandler {
bool IsMain() const { return repl_state_.IsMain(); }
bool IsReplica() const { return repl_state_.IsReplica(); }
#ifdef MG_ENTERPRISE
// coordination::CoordinatorState &CoordinatorState() { return coordinator_state_; }
#endif
/**
* @brief Return all active databases.
*
@ -302,7 +304,7 @@ class DbmsHandler {
auto db_acc_opt = db_gk.access();
if (db_acc_opt) {
auto &db_acc = *db_acc_opt;
const auto &info = db_acc->GetInfo(false, replication_role);
const auto &info = db_acc->GetInfo(replication_role);
const auto &storage_info = info.storage_info;
stats.num_vertex += storage_info.vertex_count;
stats.num_edges += storage_info.edge_count;
@ -338,7 +340,7 @@ class DbmsHandler {
auto db_acc_opt = db_gk.access();
if (db_acc_opt) {
auto &db_acc = *db_acc_opt;
res.push_back(db_acc->GetInfo(false, replication_role));
res.push_back(db_acc->GetInfo(replication_role));
}
}
return res;

96
src/dbms/inmemory/replication_handlers.cpp
View File

@ -19,7 +19,6 @@
#include "storage/v2/durability/durability.hpp"
#include "storage/v2/durability/snapshot.hpp"
#include "storage/v2/durability/version.hpp"
#include "storage/v2/fmt.hpp"
#include "storage/v2/indices/label_index_stats.hpp"
#include "storage/v2/inmemory/storage.hpp"
#include "storage/v2/inmemory/unique_constraints.hpp"
@ -119,9 +118,14 @@ void InMemoryReplicationHandlers::Register(dbms::DbmsHandler *dbms_handler, repl
});
server.rpc_server_.Register<replication_coordination_glue::SwapMainUUIDRpc>(
[&data, dbms_handler](auto *req_reader, auto *res_builder) {
spdlog::debug("Received SwapMainUUIDHandler");
spdlog::debug("Received SwapMainUUIDRpc");
InMemoryReplicationHandlers::SwapMainUUIDHandler(dbms_handler, data, req_reader, res_builder);
});
server.rpc_server_.Register<storage::replication::ForceResetStorageRpc>(
[&data, dbms_handler](auto *req_reader, auto *res_builder) {
spdlog::debug("Received ForceResetStorageRpc");
InMemoryReplicationHandlers::ForceResetStorageHandler(dbms_handler, data.uuid_, req_reader, res_builder);
});
}
void InMemoryReplicationHandlers::SwapMainUUIDHandler(dbms::DbmsHandler *dbms_handler,
@ -135,7 +139,7 @@ void InMemoryReplicationHandlers::SwapMainUUIDHandler(dbms::DbmsHandler *dbms_ha
replication_coordination_glue::SwapMainUUIDReq req;
slk::Load(&req, req_reader);
spdlog::info(fmt::format("Set replica data UUID to main uuid {}", std::string(req.uuid)));
spdlog::info("Set replica data UUID to main uuid {}", std::string(req.uuid));
dbms_handler->ReplicationState().TryPersistRoleReplica(role_replica_data.config, req.uuid);
role_replica_data.uuid_ = req.uuid;
@ -330,6 +334,78 @@ void InMemoryReplicationHandlers::SnapshotHandler(dbms::DbmsHandler *dbms_handle
spdlog::debug("Replication recovery from snapshot finished!");
}
void InMemoryReplicationHandlers::ForceResetStorageHandler(dbms::DbmsHandler *dbms_handler,
const std::optional<utils::UUID> &current_main_uuid,
slk::Reader *req_reader, slk::Builder *res_builder) {
storage::replication::ForceResetStorageReq req;
slk::Load(&req, req_reader);
auto db_acc = GetDatabaseAccessor(dbms_handler, req.db_uuid);
if (!db_acc) {
storage::replication::ForceResetStorageRes res{false, 0};
slk::Save(res, res_builder);
return;
}
if (!current_main_uuid.has_value() || req.main_uuid != current_main_uuid) [[unlikely]] {
LogWrongMain(current_main_uuid, req.main_uuid, storage::replication::SnapshotReq::kType.name);
storage::replication::ForceResetStorageRes res{false, 0};
slk::Save(res, res_builder);
return;
}
storage::replication::Decoder decoder(req_reader);
auto *storage = static_cast<storage::InMemoryStorage *>(db_acc->get()->storage());
auto storage_guard = std::unique_lock{storage->main_lock_};
// Clear the database
storage->vertices_.clear();
storage->edges_.clear();
storage->commit_log_.reset();
storage->commit_log_.emplace();
storage->constraints_.existence_constraints_ = std::make_unique<storage::ExistenceConstraints>();
storage->constraints_.unique_constraints_ = std::make_unique<storage::InMemoryUniqueConstraints>();
storage->indices_.label_index_ = std::make_unique<storage::InMemoryLabelIndex>();
storage->indices_.label_property_index_ = std::make_unique<storage::InMemoryLabelPropertyIndex>();
// Fine since we will force push when reading from WAL just random epoch with 0 timestamp, as it should be if it
// acted as MAIN before
storage->repl_storage_state_.epoch_.SetEpoch(std::string(utils::UUID{}));
storage->repl_storage_state_.last_commit_timestamp_ = 0;
storage->repl_storage_state_.history.clear();
storage->vertex_id_ = 0;
storage->edge_id_ = 0;
storage->timestamp_ = storage::kTimestampInitialId;
storage->CollectGarbage<true>(std::move(storage_guard), false);
storage->vertices_.run_gc();
storage->edges_.run_gc();
storage::replication::ForceResetStorageRes res{true, storage->repl_storage_state_.last_commit_timestamp_.load()};
slk::Save(res, res_builder);
spdlog::trace("Deleting old snapshot files.");
// Delete other durability files
auto snapshot_files = storage::durability::GetSnapshotFiles(storage->recovery_.snapshot_directory_, storage->uuid_);
for (const auto &[path, uuid, _] : snapshot_files) {
spdlog::trace("Deleting snapshot file {}", path);
storage->file_retainer_.DeleteFile(path);
}
spdlog::trace("Deleting old WAL files.");
auto wal_files = storage::durability::GetWalFiles(storage->recovery_.wal_directory_, storage->uuid_);
if (wal_files) {
for (const auto &wal_file : *wal_files) {
spdlog::trace("Deleting WAL file {}", wal_file.path);
storage->file_retainer_.DeleteFile(wal_file.path);
}
storage->wal_file_.reset();
}
}
void InMemoryReplicationHandlers::WalFilesHandler(dbms::DbmsHandler *dbms_handler,
const std::optional<utils::UUID> &current_main_uuid,
slk::Reader *req_reader, slk::Builder *res_builder) {
@ -764,6 +840,20 @@ uint64_t InMemoryReplicationHandlers::ReadAndApplyDelta(storage::InMemoryStorage
transaction->DeleteLabelPropertyIndexStats(storage->NameToLabel(info.label));
break;
}
case WalDeltaData::Type::EDGE_INDEX_CREATE: {
spdlog::trace(" Create edge index on :{}", delta.operation_edge_type.edge_type);
auto *transaction = get_transaction(timestamp, kUniqueAccess);
if (transaction->CreateIndex(storage->NameToEdgeType(delta.operation_label.label)).HasError())
throw utils::BasicException("Invalid transaction! Please raise an issue, {}:{}", __FILE__, __LINE__);
break;
}
case WalDeltaData::Type::EDGE_INDEX_DROP: {
spdlog::trace(" Drop edge index on :{}", delta.operation_edge_type.edge_type);
auto *transaction = get_transaction(timestamp, kUniqueAccess);
if (transaction->DropIndex(storage->NameToEdgeType(delta.operation_label.label)).HasError())
throw utils::BasicException("Invalid transaction! Please raise an issue, {}:{}", __FILE__, __LINE__);
break;
}
case WalDeltaData::Type::EXISTENCE_CONSTRAINT_CREATE: {
spdlog::trace(" Create existence constraint on :{} ({})", delta.operation_label_property.label,
delta.operation_label_property.property);

3
src/dbms/inmemory/replication_handlers.hpp
View File

@ -48,6 +48,9 @@ class InMemoryReplicationHandlers {
static void SwapMainUUIDHandler(dbms::DbmsHandler *dbms_handler, replication::RoleReplicaData &role_replica_data,
slk::Reader *req_reader, slk::Builder *res_builder);
static void ForceResetStorageHandler(dbms::DbmsHandler *dbms_handler,
const std::optional<utils::UUID> &current_main_uuid, slk::Reader *req_reader,
slk::Builder *res_builder);
static void LoadWal(storage::InMemoryStorage *storage, storage::replication::Decoder *decoder);

2
src/glue/communication.hpp
View File

@ -1,4 +1,4 @@
// Copyright 2023 Memgraph Ltd.
// Copyright 2024 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

202
src/io/network/endpoint.cpp
View File

@ -22,113 +22,15 @@
#include "utils/message.hpp"
#include "utils/string.hpp"
namespace {
constexpr std::string_view delimiter = ":";
} // namespace
namespace memgraph::io::network {
Endpoint::IpFamily Endpoint::GetIpFamily(const std::string &address) {
in_addr addr4;
in6_addr addr6;
int ipv4_result = inet_pton(AF_INET, address.c_str(), &addr4);
int ipv6_result = inet_pton(AF_INET6, address.c_str(), &addr6);
if (ipv4_result == 1) {
return IpFamily::IP4;
} else if (ipv6_result == 1) {
return IpFamily::IP6;
} else {
return IpFamily::NONE;
}
}
std::optional<std::pair<std::string, uint16_t>> Endpoint::ParseSocketOrIpAddress(
const std::string &address, const std::optional<uint16_t> default_port) {
/// expected address format:
/// - "ip_address:port_number"
/// - "ip_address"
/// We parse the address first. If it's an IP address, a default port must
// be given, or we return nullopt. If it's a socket address, we try to parse
// it into an ip address and a port number; even if a default port is given,
// it won't be used, as we expect that it is given in the address string.
const std::string delimiter = ":";
std::string ip_address;
std::vector<std::string> parts = utils::Split(address, delimiter);
if (parts.size() == 1) {
if (default_port) {
if (GetIpFamily(address) == IpFamily::NONE) {
return std::nullopt;
}
return std::pair{address, *default_port};
}
} else if (parts.size() == 2) {
ip_address = std::move(parts[0]);
if (GetIpFamily(ip_address) == IpFamily::NONE) {
return std::nullopt;
}
int64_t int_port{0};
try {
int_port = utils::ParseInt(parts[1]);
} catch (utils::BasicException &e) {
spdlog::error(utils::MessageWithLink("Invalid port number {}.", parts[1], "https://memgr.ph/ports"));
return std::nullopt;
}
if (int_port < 0) {
spdlog::error(utils::MessageWithLink("Invalid port number {}. The port number must be a positive integer.",
int_port, "https://memgr.ph/ports"));
return std::nullopt;
}
if (int_port > std::numeric_limits<uint16_t>::max()) {
spdlog::error(utils::MessageWithLink("Invalid port number. The port number exceedes the maximum possible size.",
"https://memgr.ph/ports"));
return std::nullopt;
}
return std::pair{ip_address, static_cast<uint16_t>(int_port)};
}
return std::nullopt;
}
std::optional<std::pair<std::string, uint16_t>> Endpoint::ParseHostname(
const std::string &address, const std::optional<uint16_t> default_port = {}) {
const std::string delimiter = ":";
std::string ip_address;
std::vector<std::string> parts = utils::Split(address, delimiter);
if (parts.size() == 1) {
if (default_port) {
if (!IsResolvableAddress(address, *default_port)) {
return std::nullopt;
}
return std::pair{address, *default_port};
}
} else if (parts.size() == 2) {
int64_t int_port{0};
auto hostname = std::move(parts[0]);
try {
int_port = utils::ParseInt(parts[1]);
} catch (utils::BasicException &e) {
spdlog::error(utils::MessageWithLink("Invalid port number {}.", parts[1], "https://memgr.ph/ports"));
return std::nullopt;
}
if (int_port < 0) {
spdlog::error(utils::MessageWithLink("Invalid port number {}. The port number must be a positive integer.",
int_port, "https://memgr.ph/ports"));
return std::nullopt;
}
if (int_port > std::numeric_limits<uint16_t>::max()) {
spdlog::error(utils::MessageWithLink("Invalid port number. The port number exceedes the maximum possible size.",
"https://memgr.ph/ports"));
return std::nullopt;
}
if (IsResolvableAddress(hostname, static_cast<uint16_t>(int_port))) {
return std::pair{hostname, static_cast<u_int16_t>(int_port)};
}
}
return std::nullopt;
}
std::string Endpoint::SocketAddress() const {
auto ip_address = address.empty() ? "EMPTY" : address;
return ip_address + ":" + std::to_string(port);
}
// NOLINTNEXTLINE
Endpoint::Endpoint(needs_resolving_t, std::string hostname, uint16_t port)
: address(std::move(hostname)), port(port), family{GetIpFamily(address)} {}
Endpoint::Endpoint(std::string ip_address, uint16_t port) : address(std::move(ip_address)), port(port) {
IpFamily ip_family = GetIpFamily(address);
@ -138,9 +40,23 @@ Endpoint::Endpoint(std::string ip_address, uint16_t port) : address(std::move(ip
family = ip_family;
}
// NOLINTNEXTLINE
Endpoint::Endpoint(needs_resolving_t, std::string hostname, uint16_t port)
: address(std::move(hostname)), port(port), family{GetIpFamily(address)} {}
std::string Endpoint::SocketAddress() const { return fmt::format("{}:{}", address, port); }
Endpoint::IpFamily Endpoint::GetIpFamily(std::string_view address) {
// Ensure null-terminated
auto const tmp = std::string(address);
in_addr addr4;
in6_addr addr6;
int ipv4_result = inet_pton(AF_INET, tmp.c_str(), &addr4);
int ipv6_result = inet_pton(AF_INET6, tmp.c_str(), &addr6);
if (ipv4_result == 1) {
return IpFamily::IP4;
}
if (ipv6_result == 1) {
return IpFamily::IP6;
}
return IpFamily::NONE;
}
std::ostream &operator<<(std::ostream &os, const Endpoint &endpoint) {
// no need to cover the IpFamily::NONE case, as you can't even construct an
@ -153,35 +69,73 @@ std::ostream &operator<<(std::ostream &os, const Endpoint &endpoint) {
return os << endpoint.address << ":" << endpoint.port;
}
bool Endpoint::IsResolvableAddress(const std::string &address, uint16_t port) {
// NOTE: Intentional copy to ensure null-terminated string
bool Endpoint::IsResolvableAddress(std::string_view address, uint16_t port) {
addrinfo hints{
.ai_flags = AI_PASSIVE,
.ai_family = AF_UNSPEC, // IPv4 and IPv6
.ai_socktype = SOCK_STREAM // TCP socket
};
addrinfo *info = nullptr;
auto status = getaddrinfo(address.c_str(), std::to_string(port).c_str(), &hints, &info);
auto status = getaddrinfo(std::string(address).c_str(), std::to_string(port).c_str(), &hints, &info);
if (info) freeaddrinfo(info);
return status == 0;
}
std::optional<std::pair<std::string, uint16_t>> Endpoint::ParseSocketOrAddress(
const std::string &address, const std::optional<uint16_t> default_port) {
const std::string delimiter = ":";
std::vector<std::string> parts = utils::Split(address, delimiter);
if (parts.size() == 1) {
if (GetIpFamily(address) == IpFamily::NONE) {
return ParseHostname(address, default_port);
}
return ParseSocketOrIpAddress(address, default_port);
std::optional<ParsedAddress> Endpoint::ParseSocketOrAddress(std::string_view address,
std::optional<uint16_t> default_port) {
auto const parts = utils::SplitView(address, delimiter);
if (parts.size() > 2) {
return std::nullopt;
}
if (parts.size() == 2) {
if (GetIpFamily(parts[0]) == IpFamily::NONE) {
return ParseHostname(address, default_port);
auto const port = [default_port, &parts]() -> std::optional<uint16_t> {
if (parts.size() == 2) {
return static_cast<uint16_t>(utils::ParseInt(parts[1]));
}
return ParseSocketOrIpAddress(address, default_port);
return default_port;
}();
if (!ValidatePort(port)) {
return std::nullopt;
}
return std::nullopt;
auto const addr = [address, &parts]() {
if (parts.size() == 2) {
return parts[0];
}
return address;
}();
if (GetIpFamily(addr) == IpFamily::NONE) {
if (IsResolvableAddress(addr, *port)) { // NOLINT
return std::pair{addr, *port}; // NOLINT
}
return std::nullopt;
}
return std::pair{addr, *port}; // NOLINT
}
auto Endpoint::ValidatePort(std::optional<uint16_t> port) -> bool {
if (!port) {
return false;
}
if (port < 0) {
spdlog::error(utils::MessageWithLink("Invalid port number {}. The port number must be a positive integer.", *port,
"https://memgr.ph/ports"));
return false;
}
if (port > std::numeric_limits<uint16_t>::max()) {
spdlog::error(utils::MessageWithLink("Invalid port number. The port number exceedes the maximum possible size.",
"https://memgr.ph/ports"));
return false;
}
return true;
}
} // namespace memgraph::io::network

55
src/io/network/endpoint.hpp
View File

@ -19,11 +19,8 @@
namespace memgraph::io::network {
/**
* This class represents a network endpoint that is used in Socket.
* It is used when connecting to an address and to get the current
* connection address.
*/
using ParsedAddress = std::pair<std::string_view, uint16_t>;
struct Endpoint {
static const struct needs_resolving_t {
} needs_resolving;
@ -31,59 +28,35 @@ struct Endpoint {
Endpoint() = default;
Endpoint(std::string ip_address, uint16_t port);
Endpoint(needs_resolving_t, std::string hostname, uint16_t port);
Endpoint(Endpoint const &) = default;
Endpoint(Endpoint &&) noexcept = default;
Endpoint &operator=(Endpoint const &) = default;
Endpoint &operator=(Endpoint &&) noexcept = default;
~Endpoint() = default;
enum class IpFamily : std::uint8_t { NONE, IP4, IP6 };
std::string SocketAddress() const;
static std::optional<ParsedAddress> ParseSocketOrAddress(std::string_view address,
std::optional<uint16_t> default_port = {});
bool operator==(const Endpoint &other) const = default;
friend std::ostream &operator<<(std::ostream &os, const Endpoint &endpoint);
std::string SocketAddress() const;
std::string address;
uint16_t port{0};
IpFamily family{IpFamily::NONE};
static std::optional<std::pair<std::string, uint16_t>> ParseSocketOrAddress(const std::string &address,
std::optional<uint16_t> default_port);
bool operator==(const Endpoint &other) const = default;
friend std::ostream &operator<<(std::ostream &os, const Endpoint &endpoint);
/**
* Tries to parse the given string as either a socket address or ip address.
* Expected address format:
* - "ip_address:port_number"
* - "ip_address"
* We parse the address first. If it's an IP address, a default port must
* be given, or we return nullopt. If it's a socket address, we try to parse
* it into an ip address and a port number; even if a default port is given,
* it won't be used, as we expect that it is given in the address string.
*/
static std::optional<std::pair<std::string, uint16_t>> ParseSocketOrIpAddress(
const std::string &address, std::optional<uint16_t> default_port = {});
private:
static IpFamily GetIpFamily(std::string_view address);
/**
* Tries to parse given string as either socket address or hostname.
* Expected address format:
* - "hostname:port_number"
* - "hostname"
* After we parse hostname and port we try to resolve the hostname into an ip_address.
*/
static std::optional<std::pair<std::string, uint16_t>> ParseHostname(const std::string &address,
std::optional<uint16_t> default_port);
static bool IsResolvableAddress(std::string_view address, uint16_t port);
static IpFamily GetIpFamily(const std::string &address);
static bool IsResolvableAddress(const std::string &address, uint16_t port);
/**
* Tries to resolve hostname to its corresponding IP address.
* Given a DNS hostname, this function performs resolution and returns
* the IP address associated with the hostname.
*/
static std::string ResolveHostnameIntoIpAddress(const std::string &address, uint16_t port);
static auto ValidatePort(std::optional<uint16_t> port) -> bool;
};
} // namespace memgraph::io::network

3
src/memgraph.cpp
View File

@ -334,7 +334,8 @@ int main(int argc, char **argv) {
.salient.items = {.properties_on_edges = FLAGS_storage_properties_on_edges,
.enable_schema_metadata = FLAGS_storage_enable_schema_metadata},
.salient.storage_mode = memgraph::flags::ParseStorageMode()};
spdlog::info("config recover on startup {}, flags {} {}", db_config.durability.recover_on_startup,
FLAGS_storage_recover_on_startup, FLAGS_data_recovery_on_startup);
memgraph::utils::Scheduler jemalloc_purge_scheduler;
jemalloc_purge_scheduler.Run("Jemalloc purge", std::chrono::seconds(FLAGS_storage_gc_cycle_sec),
[] { memgraph::memory::PurgeUnusedMemory(); });

4
src/memory/global_memory_control.cpp
View File

@ -122,11 +122,11 @@ static bool my_commit(extent_hooks_t *extent_hooks, void *addr, size_t size, siz
[[maybe_unused]] auto blocker = memgraph::utils::MemoryTracker::OutOfMemoryExceptionBlocker{};
if (GetQueriesMemoryControl().IsThreadTracked()) [[unlikely]] {
bool ok = GetQueriesMemoryControl().TrackAllocOnCurrentThread(length);
[[maybe_unused]] bool ok = GetQueriesMemoryControl().TrackAllocOnCurrentThread(length);
DMG_ASSERT(ok);
}
auto ok = memgraph::utils::total_memory_tracker.Alloc(static_cast<int64_t>(length));
[[maybe_unused]] auto ok = memgraph::utils::total_memory_tracker.Alloc(static_cast<int64_t>(length));
DMG_ASSERT(ok);
return false;

2
src/mg_import_csv.cpp
View File

@ -416,7 +416,7 @@ memgraph::storage::PropertyValue StringToValue(const std::string &str, const std
std::string GetIdSpace(const std::string &type) {
// The format of this field is as follows:
// [START_|END_]ID[(<id_space>)]
std::regex format(R"(^(START_|END_)?ID(\(([^\(\)]+)\))?$)", std::regex::extended);
static std::regex format(R"(^(START_|END_)?ID(\(([^\(\)]+)\))?$)", std::regex::extended);
std::smatch res;
if (!std::regex_match(type, res, format))
throw LoadException(

70
src/query/db_accessor.hpp
View File

@ -371,6 +371,62 @@ class VerticesIterable final {
}
};
class EdgesIterable final {
std::variant<storage::EdgesIterable, std::unordered_set<EdgeAccessor, std::hash<EdgeAccessor>, std::equal_to<void>,
utils::Allocator<EdgeAccessor>> *>
iterable_;
public:
class Iterator final {
std::variant<storage::EdgesIterable::Iterator,
std::unordered_set<EdgeAccessor, std::hash<EdgeAccessor>, std::equal_to<void>,
utils::Allocator<EdgeAccessor>>::iterator>
it_;
public:
explicit Iterator(storage::EdgesIterable::Iterator it) : it_(std::move(it)) {}
explicit Iterator(std::unordered_set<EdgeAccessor, std::hash<EdgeAccessor>, std::equal_to<void>,
utils::Allocator<EdgeAccessor>>::iterator it)
: it_(it) {}
EdgeAccessor operator*() const {
return std::visit([](auto &it_) { return EdgeAccessor(*it_); }, it_);
}
Iterator &operator++() {
std::visit([](auto &it_) { ++it_; }, it_);
return *this;
}
bool operator==(const Iterator &other) const { return it_ == other.it_; }
bool operator!=(const Iterator &other) const { return !(other == *this); }
};
explicit EdgesIterable(storage::EdgesIterable iterable) : iterable_(std::move(iterable)) {}
explicit EdgesIterable(std::unordered_set<EdgeAccessor, std::hash<EdgeAccessor>, std::equal_to<void>,
utils::Allocator<EdgeAccessor>> *edges)
: iterable_(edges) {}
Iterator begin() {
return std::visit(
memgraph::utils::Overloaded{
[](storage::EdgesIterable &iterable_) { return Iterator(iterable_.begin()); },
[](std::unordered_set<EdgeAccessor, std::hash<EdgeAccessor>, std::equal_to<void>,
utils::Allocator<EdgeAccessor>> *iterable_) { return Iterator(iterable_->begin()); }},
iterable_);
}
Iterator end() {
return std::visit(
memgraph::utils::Overloaded{
[](storage::EdgesIterable &iterable_) { return Iterator(iterable_.end()); },
[](std::unordered_set<EdgeAccessor, std::hash<EdgeAccessor>, std::equal_to<void>,
utils::Allocator<EdgeAccessor>> *iterable_) { return Iterator(iterable_->end()); }},
iterable_);
}
};
class DbAccessor final {
storage::Storage::Accessor *accessor_;
@ -416,6 +472,10 @@ class DbAccessor final {
return VerticesIterable(accessor_->Vertices(label, property, lower, upper, view));
}
EdgesIterable Edges(storage::View view, storage::EdgeTypeId edge_type) {
return EdgesIterable(accessor_->Edges(edge_type, view));
}
VertexAccessor InsertVertex() { return VertexAccessor(accessor_->CreateVertex()); }
storage::Result<EdgeAccessor> InsertEdge(VertexAccessor *from, VertexAccessor *to,
@ -572,6 +632,8 @@ class DbAccessor final {
return accessor_->LabelPropertyIndexExists(label, prop);
}
bool EdgeTypeIndexExists(storage::EdgeTypeId edge_type) const { return accessor_->EdgeTypeIndexExists(edge_type); }
std::optional<storage::LabelIndexStats> GetIndexStats(const storage::LabelId &label) const {
return accessor_->GetIndexStats(label);
}
@ -638,6 +700,10 @@ class DbAccessor final {
return accessor_->CreateIndex(label, property);
}
utils::BasicResult<storage::StorageIndexDefinitionError, void> CreateIndex(storage::EdgeTypeId edge_type) {
return accessor_->CreateIndex(edge_type);
}
utils::BasicResult<storage::StorageIndexDefinitionError, void> DropIndex(storage::LabelId label) {
return accessor_->DropIndex(label);
}
@ -647,6 +713,10 @@ class DbAccessor final {
return accessor_->DropIndex(label, property);
}
utils::BasicResult<storage::StorageIndexDefinitionError, void> DropIndex(storage::EdgeTypeId edge_type) {
return accessor_->DropIndex(edge_type);
}
utils::BasicResult<storage::StorageExistenceConstraintDefinitionError, void> CreateExistenceConstraint(
storage::LabelId label, storage::PropertyId property) {
return accessor_->CreateExistenceConstraint(label, property);

35
src/query/dump.cpp
View File

@ -242,6 +242,10 @@ void DumpLabelIndex(std::ostream *os, query::DbAccessor *dba, const storage::Lab
*os << "CREATE INDEX ON :" << EscapeName(dba->LabelToName(label)) << ";";
}
void DumpEdgeTypeIndex(std::ostream *os, query::DbAccessor *dba, const storage::EdgeTypeId edge_type) {
*os << "CREATE EDGE INDEX ON :" << EscapeName(dba->EdgeTypeToName(edge_type)) << ";";
}
void DumpLabelPropertyIndex(std::ostream *os, query::DbAccessor *dba, storage::LabelId label,
storage::PropertyId property) {
*os << "CREATE INDEX ON :" << EscapeName(dba->LabelToName(label)) << "(" << EscapeName(dba->PropertyToName(property))
@ -297,7 +301,9 @@ PullPlanDump::PullPlanDump(DbAccessor *dba, dbms::DatabaseAccess db_acc)
// Internal index cleanup
CreateInternalIndexCleanupPullChunk(),
// Dump all triggers
CreateTriggersPullChunk()} {}
CreateTriggersPullChunk(),
// Dump all edge-type indices
CreateEdgeTypeIndicesPullChunk()} {}
bool PullPlanDump::Pull(AnyStream *stream, std::optional<int> n) {
// Iterate all functions that stream some results.
@ -352,6 +358,33 @@ PullPlanDump::PullChunk PullPlanDump::CreateLabelIndicesPullChunk() {
};
}
PullPlanDump::PullChunk PullPlanDump::CreateEdgeTypeIndicesPullChunk() {
// Dump all label indices
return [this, global_index = 0U](AnyStream *stream, std::optional<int> n) mutable -> std::optional<size_t> {
// Delay the construction of indices vectors
if (!indices_info_) {
indices_info_.emplace(dba_->ListAllIndices());
}
const auto &edge_type = indices_info_->edge_type;
size_t local_counter = 0;
while (global_index < edge_type.size() && (!n || local_counter < *n)) {
std::ostringstream os;
DumpEdgeTypeIndex(&os, dba_, edge_type[global_index]);
stream->Result({TypedValue(os.str())});
++global_index;
++local_counter;
}
if (global_index == edge_type.size()) {
return local_counter;
}
return std::nullopt;
};
}
PullPlanDump::PullChunk PullPlanDump::CreateLabelPropertyIndicesPullChunk() {
return [this, global_index = 0U](AnyStream *stream, std::optional<int> n) mutable -> std::optional<size_t> {
// Delay the construction of indices vectors

1
src/query/dump.hpp
View File

@ -63,5 +63,6 @@ struct PullPlanDump {
PullChunk CreateDropInternalIndexPullChunk();
PullChunk CreateInternalIndexCleanupPullChunk();
PullChunk CreateTriggersPullChunk();
PullChunk CreateEdgeTypeIndicesPullChunk();
};
} // namespace memgraph::query

3
src/query/frontend/ast/ast.cpp
View File

@ -186,6 +186,9 @@ constexpr utils::TypeInfo query::ProfileQuery::kType{utils::TypeId::AST_PROFILE_
constexpr utils::TypeInfo query::IndexQuery::kType{utils::TypeId::AST_INDEX_QUERY, "IndexQuery", &query::Query::kType};
constexpr utils::TypeInfo query::EdgeIndexQuery::kType{utils::TypeId::AST_EDGE_INDEX_QUERY, "EdgeIndexQuery",
&query::Query::kType};
constexpr utils::TypeInfo query::Create::kType{utils::TypeId::AST_CREATE, "Create", &query::Clause::kType};
constexpr utils::TypeInfo query::CallProcedure::kType{utils::TypeId::AST_CALL_PROCEDURE, "CallProcedure",

39
src/query/frontend/ast/ast.hpp
View File

@ -21,6 +21,7 @@
#include "query/interpret/awesome_memgraph_functions.hpp"
#include "query/typed_value.hpp"
#include "storage/v2/property_value.hpp"
#include "utils/exceptions.hpp"
#include "utils/typeinfo.hpp"
namespace memgraph::query {
@ -2223,6 +2224,34 @@ class IndexQuery : public memgraph::query::Query {
friend class AstStorage;
};
class EdgeIndexQuery : public memgraph::query::Query {
public:
static const utils::TypeInfo kType;
const utils::TypeInfo &GetTypeInfo() const override { return kType; }
enum class Action { CREATE, DROP };
EdgeIndexQuery() = default;
DEFVISITABLE(QueryVisitor<void>);
memgraph::query::EdgeIndexQuery::Action action_;
memgraph::query::EdgeTypeIx edge_type_;
EdgeIndexQuery *Clone(AstStorage *storage) const override {
EdgeIndexQuery *object = storage->Create<EdgeIndexQuery>();
object->action_ = action_;
object->edge_type_ = storage->GetEdgeTypeIx(edge_type_.name);
return object;
}
protected:
EdgeIndexQuery(Action action, EdgeTypeIx edge_type) : action_(action), edge_type_(edge_type) {}
private:
friend class AstStorage;
};
class Create : public memgraph::query::Clause {
public:
static const utils::TypeInfo kType;
@ -3035,8 +3064,6 @@ class ReplicationQuery : public memgraph::query::Query {
enum class SyncMode { SYNC, ASYNC };
enum class ReplicaState { READY, REPLICATING, RECOVERY, MAYBE_BEHIND, DIVERGED_FROM_MAIN };
ReplicationQuery() = default;
DEFVISITABLE(QueryVisitor<void>);
@ -3588,7 +3615,7 @@ class PatternComprehension : public memgraph::query::Expression {
bool Accept(HierarchicalTreeVisitor &visitor) override {
if (visitor.PreVisit(*this)) {
if (variable_) {
variable_->Accept(visitor);
throw utils::NotYetImplemented("Variable in pattern comprehension.");
}
pattern_->Accept(visitor);
if (filter_) {
@ -3617,7 +3644,8 @@ class PatternComprehension : public memgraph::query::Expression {
int32_t symbol_pos_{-1};
PatternComprehension *Clone(AstStorage *storage) const override {
PatternComprehension *object = storage->Create<PatternComprehension>();
auto *object = storage->Create<PatternComprehension>();
object->variable_ = variable_ ? variable_->Clone(storage) : nullptr;
object->pattern_ = pattern_ ? pattern_->Clone(storage) : nullptr;
object->filter_ = filter_ ? filter_->Clone(storage) : nullptr;
object->resultExpr_ = resultExpr_ ? resultExpr_->Clone(storage) : nullptr;
@ -3627,7 +3655,8 @@ class PatternComprehension : public memgraph::query::Expression {
}
protected:
PatternComprehension(Identifier *variable, Pattern *pattern) : variable_(variable), pattern_(pattern) {}
PatternComprehension(Identifier *variable, Pattern *pattern, Where *filter, Expression *resultExpr)
: variable_(variable), pattern_(pattern), filter_(filter), resultExpr_(resultExpr) {}
private:
friend class AstStorage;

13
src/query/frontend/ast/ast_visitor.hpp
View File

@ -82,6 +82,7 @@ class AuthQuery;
class ExplainQuery;
class ProfileQuery;
class IndexQuery;
class EdgeIndexQuery;
class DatabaseInfoQuery;
class SystemInfoQuery;
class ConstraintQuery;
@ -143,11 +144,11 @@ class ExpressionVisitor
template <class TResult>
class QueryVisitor
: public utils::Visitor<TResult, CypherQuery, ExplainQuery, ProfileQuery, IndexQuery, AuthQuery, DatabaseInfoQuery,
SystemInfoQuery, ConstraintQuery, DumpQuery, ReplicationQuery, LockPathQuery,
FreeMemoryQuery, TriggerQuery, IsolationLevelQuery, CreateSnapshotQuery, StreamQuery,
SettingQuery, VersionQuery, ShowConfigQuery, TransactionQueueQuery, StorageModeQuery,
AnalyzeGraphQuery, MultiDatabaseQuery, ShowDatabasesQuery, EdgeImportModeQuery,
CoordinatorQuery> {};
: public utils::Visitor<TResult, CypherQuery, ExplainQuery, ProfileQuery, IndexQuery, EdgeIndexQuery, AuthQuery,
DatabaseInfoQuery, SystemInfoQuery, ConstraintQuery, DumpQuery, ReplicationQuery,
LockPathQuery, FreeMemoryQuery, TriggerQuery, IsolationLevelQuery, CreateSnapshotQuery,
StreamQuery, SettingQuery, VersionQuery, ShowConfigQuery, TransactionQueueQuery,
StorageModeQuery, AnalyzeGraphQuery, MultiDatabaseQuery, ShowDatabasesQuery,
EdgeImportModeQuery, CoordinatorQuery> {};
} // namespace memgraph::query

21
src/query/frontend/ast/cypher_main_visitor.cpp
View File

@ -265,6 +265,27 @@ antlrcpp::Any CypherMainVisitor::visitDropIndex(MemgraphCypher::DropIndexContext
return index_query;
}
antlrcpp::Any CypherMainVisitor::visitEdgeIndexQuery(MemgraphCypher::EdgeIndexQueryContext *ctx) {
MG_ASSERT(ctx->children.size() == 1, "EdgeIndexQuery should have exactly one child!");
auto *index_query = std::any_cast<EdgeIndexQuery *>(ctx->children[0]->accept(this));
query_ = index_query;
return index_query;
}
antlrcpp::Any CypherMainVisitor::visitCreateEdgeIndex(MemgraphCypher::CreateEdgeIndexContext *ctx) {
auto *index_query = storage_->Create<EdgeIndexQuery>();
index_query->action_ = EdgeIndexQuery::Action::CREATE;
index_query->edge_type_ = AddEdgeType(std::any_cast<std::string>(ctx->labelName()->accept(this)));
return index_query;
}
antlrcpp::Any CypherMainVisitor::visitDropEdgeIndex(MemgraphCypher::DropEdgeIndexContext *ctx) {
auto *index_query = storage_->Create<EdgeIndexQuery>();
index_query->action_ = EdgeIndexQuery::Action::DROP;
index_query->edge_type_ = AddEdgeType(std::any_cast<std::string>(ctx->labelName()->accept(this)));
return index_query;
}
antlrcpp::Any CypherMainVisitor::visitAuthQuery(MemgraphCypher::AuthQueryContext *ctx) {
MG_ASSERT(ctx->children.size() == 1, "AuthQuery should have exactly one child!");
auto *auth_query = std::any_cast<AuthQuery *>(ctx->children[0]->accept(this));

15
src/query/frontend/ast/cypher_main_visitor.hpp
View File

@ -148,6 +148,11 @@ class CypherMainVisitor : public antlropencypher::MemgraphCypherBaseVisitor {
*/
antlrcpp::Any visitIndexQuery(MemgraphCypher::IndexQueryContext *ctx) override;
/**
* @return IndexQuery*
*/
antlrcpp::Any visitEdgeIndexQuery(MemgraphCypher::EdgeIndexQueryContext *ctx) override;
/**
* @return ExplainQuery*
*/
@ -499,6 +504,16 @@ class CypherMainVisitor : public antlropencypher::MemgraphCypherBaseVisitor {
*/
antlrcpp::Any visitDropIndex(MemgraphCypher::DropIndexContext *ctx) override;
/**
* @return EdgeIndexQuery*
*/
antlrcpp::Any visitCreateEdgeIndex(MemgraphCypher::CreateEdgeIndexContext *ctx) override;
/**
* @return DropEdgeIndex*
*/
antlrcpp::Any visitDropEdgeIndex(MemgraphCypher::DropEdgeIndexContext *ctx) override;
/**
* @return AuthQuery*
*/

7
src/query/frontend/opencypher/grammar/MemgraphCypher.g4
View File

@ -133,6 +133,7 @@ symbolicName : UnescapedSymbolicName
query : cypherQuery
| indexQuery
| edgeIndexQuery
| explainQuery
| profileQuery
| databaseInfoQuery
@ -527,3 +528,9 @@ showDatabase : SHOW DATABASE ;
showDatabases : SHOW DATABASES ;
edgeImportModeQuery : EDGE IMPORT MODE ( ACTIVE | INACTIVE ) ;
createEdgeIndex : CREATE EDGE INDEX ON ':' labelName ;
dropEdgeIndex : DROP EDGE INDEX ON ':' labelName ;
edgeIndexQuery : createEdgeIndex | dropEdgeIndex ;

2
src/query/frontend/semantic/required_privileges.cpp
View File

@ -27,6 +27,8 @@ class PrivilegeExtractor : public QueryVisitor<void>, public HierarchicalTreeVis
void Visit(IndexQuery & /*unused*/) override { AddPrivilege(AuthQuery::Privilege::INDEX); }
void Visit(EdgeIndexQuery & /*unused*/) override { AddPrivilege(AuthQuery::Privilege::INDEX); }
void Visit(AnalyzeGraphQuery & /*unused*/) override { AddPrivilege(AuthQuery::Privilege::INDEX); }
void Visit(AuthQuery & /*unused*/) override { AddPrivilege(AuthQuery::Privilege::AUTH); }

2
src/query/frontend/semantic/symbol.hpp
View File

@ -53,6 +53,8 @@ class Symbol {
bool user_declared() const { return user_declared_; }
int token_position() const { return token_position_; }
bool IsSymbolAnonym() const { return name_.substr(0U, 4U) == "anon"; }
std::string name_;
int64_t position_;
bool user_declared_{true};

26
src/query/frontend/semantic/symbol_generator.cpp
View File

@ -721,6 +721,32 @@ bool SymbolGenerator::PostVisit(EdgeAtom &) {
return true;
}
bool SymbolGenerator::PreVisit(PatternComprehension &pc) {
auto &scope = scopes_.back();
if (scope.in_set_property) {
throw utils::NotYetImplemented("Pattern Comprehension cannot be used within SET clause.!");
}
if (scope.in_with) {
throw utils::NotYetImplemented("Pattern Comprehension cannot be used within WITH!");
}
if (scope.in_reduce) {
throw utils::NotYetImplemented("Pattern Comprehension cannot be used within REDUCE!");
}
if (scope.num_if_operators) {
throw utils::NotYetImplemented("IF operator cannot be used with Pattern Comprehension!");
}
const auto &symbol = CreateAnonymousSymbol();
pc.MapTo(symbol);
return true;
}
bool SymbolGenerator::PostVisit(PatternComprehension & /*pc*/) { return true; }
void SymbolGenerator::VisitWithIdentifiers(Expression *expr, const std::vector<Identifier *> &identifiers) {
auto &scope = scopes_.back();
std::vector<std::pair<std::optional<Symbol>, Identifier *>> prev_symbols;

4
src/query/frontend/semantic/symbol_generator.hpp
View File

@ -1,4 +1,4 @@
// Copyright 2023 Memgraph Ltd.
// Copyright 2024 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
@ -97,6 +97,8 @@ class SymbolGenerator : public HierarchicalTreeVisitor {
bool PostVisit(NodeAtom &) override;
bool PreVisit(EdgeAtom &) override;
bool PostVisit(EdgeAtom &) override;
bool PreVisit(PatternComprehension &) override;
bool PostVisit(PatternComprehension &) override;
private:
// Scope stores the state of where we are when visiting the AST and a map of

334
src/query/interpreter.cpp
View File

@ -297,16 +297,6 @@ inline auto convertToReplicationMode(const ReplicationQuery::SyncMode &sync_mode
class ReplQueryHandler {
public:
struct ReplicaInfo {
std::string name;
std::string socket_address;
ReplicationQuery::SyncMode sync_mode;
std::optional<double> timeout;
uint64_t current_timestamp_of_replica;
uint64_t current_number_of_timestamp_behind_master;
ReplicationQuery::ReplicaState state;
};
explicit ReplQueryHandler(query::ReplicationQueryHandler &replication_query_handler)
: handler_{&replication_query_handler} {}
@ -365,7 +355,7 @@ class ReplQueryHandler {
const auto replication_config =
replication::ReplicationClientConfig{.name = name,
.mode = repl_mode,
.ip_address = ip,
.ip_address = std::string(ip),
.port = port,
.replica_check_frequency = replica_check_frequency,
.ssl = std::nullopt};
@ -397,58 +387,16 @@ class ReplQueryHandler {
}
}
std::vector<ReplicaInfo> ShowReplicas(const dbms::Database &db) const {
if (handler_->IsReplica()) {
// replica can't show registered replicas (it shouldn't have any)
throw QueryRuntimeException("Replica can't show registered replicas (it shouldn't have any)!");
std::vector<ReplicasInfo> ShowReplicas() const {
auto info = handler_->ShowReplicas();
if (info.HasError()) {
switch (info.GetError()) {
case ShowReplicaError::NOT_MAIN:
throw QueryRuntimeException("Replica can't show registered replicas (it shouldn't have any)!");
}
}
// TODO: Combine results? Have a single place with clients???
// Also authentication checks (replica + database visibility)
const auto repl_infos = db.storage()->ReplicasInfo();
std::vector<ReplicaInfo> replicas;
replicas.reserve(repl_infos.size());
const auto from_info = [](const auto &repl_info) -> ReplicaInfo {
ReplicaInfo replica;
replica.name = repl_info.name;
replica.socket_address = repl_info.endpoint.SocketAddress();
switch (repl_info.mode) {
case replication_coordination_glue::ReplicationMode::SYNC:
replica.sync_mode = ReplicationQuery::SyncMode::SYNC;
break;
case replication_coordination_glue::ReplicationMode::ASYNC:
replica.sync_mode = ReplicationQuery::SyncMode::ASYNC;
break;
}
replica.current_timestamp_of_replica = repl_info.timestamp_info.current_timestamp_of_replica;
replica.current_number_of_timestamp_behind_master =
repl_info.timestamp_info.current_number_of_timestamp_behind_master;
switch (repl_info.state) {
case storage::replication::ReplicaState::READY:
replica.state = ReplicationQuery::ReplicaState::READY;
break;
case storage::replication::ReplicaState::REPLICATING:
replica.state = ReplicationQuery::ReplicaState::REPLICATING;
break;
case storage::replication::ReplicaState::RECOVERY:
replica.state = ReplicationQuery::ReplicaState::RECOVERY;
break;
case storage::replication::ReplicaState::MAYBE_BEHIND:
replica.state = ReplicationQuery::ReplicaState::MAYBE_BEHIND;
break;
case storage::replication::ReplicaState::DIVERGED_FROM_MAIN:
replica.state = ReplicationQuery::ReplicaState::DIVERGED_FROM_MAIN;
break;
}
return replica;
};
std::transform(repl_infos.begin(), repl_infos.end(), std::back_inserter(replicas), from_info);
return replicas;
return info.GetValue().entries_;
}
private:
@ -462,7 +410,7 @@ class CoordQueryHandler final : public query::CoordinatorQueryHandler {
: coordinator_handler_(coordinator_state) {}
void UnregisterInstance(std::string const &instance_name) override {
void UnregisterInstance(std::string_view instance_name) override {
auto status = coordinator_handler_.UnregisterReplicationInstance(instance_name);
switch (status) {
using enum memgraph::coordination::UnregisterInstanceCoordinatorStatus;
@ -475,6 +423,8 @@ class CoordQueryHandler final : public query::CoordinatorQueryHandler {
throw QueryRuntimeException("UNREGISTER INSTANCE query can only be run on a coordinator!");
case NOT_LEADER:
throw QueryRuntimeException("Couldn't unregister replica instance since coordinator is not a leader!");
case RAFT_LOG_ERROR:
throw QueryRuntimeException("Couldn't unregister replica instance since raft server couldn't append the log!");
case RPC_FAILED:
throw QueryRuntimeException(
"Couldn't unregister replica instance because current main instance couldn't unregister replica!");
@ -483,20 +433,18 @@ class CoordQueryHandler final : public query::CoordinatorQueryHandler {
}
}
void RegisterReplicationInstance(std::string const &coordinator_socket_address,
std::string const &replication_socket_address,
void RegisterReplicationInstance(std::string_view coordinator_socket_address,
std::string_view replication_socket_address,
std::chrono::seconds const &instance_check_frequency,
std::chrono::seconds const &instance_down_timeout,
std::chrono::seconds const &instance_get_uuid_frequency,
std::string const &instance_name, CoordinatorQuery::SyncMode sync_mode) override {
const auto maybe_replication_ip_port =
io::network::Endpoint::ParseSocketOrAddress(replication_socket_address, std::nullopt);
std::string_view instance_name, CoordinatorQuery::SyncMode sync_mode) override {
const auto maybe_replication_ip_port = io::network::Endpoint::ParseSocketOrAddress(replication_socket_address);
if (!maybe_replication_ip_port) {
throw QueryRuntimeException("Invalid replication socket address!");
}
const auto maybe_coordinator_ip_port =
io::network::Endpoint::ParseSocketOrAddress(coordinator_socket_address, std::nullopt);
const auto maybe_coordinator_ip_port = io::network::Endpoint::ParseSocketOrAddress(coordinator_socket_address);
if (!maybe_replication_ip_port) {
throw QueryRuntimeException("Invalid replication socket address!");
}
@ -504,14 +452,14 @@ class CoordQueryHandler final : public query::CoordinatorQueryHandler {
const auto [replication_ip, replication_port] = *maybe_replication_ip_port;
const auto [coordinator_server_ip, coordinator_server_port] = *maybe_coordinator_ip_port;
const auto repl_config = coordination::CoordinatorClientConfig::ReplicationClientInfo{
.instance_name = instance_name,
.instance_name = std::string(instance_name),
.replication_mode = convertFromCoordinatorToReplicationMode(sync_mode),
.replication_ip_address = replication_ip,
.replication_ip_address = std::string(replication_ip),
.replication_port = replication_port};
auto coordinator_client_config =
coordination::CoordinatorClientConfig{.instance_name = instance_name,
.ip_address = coordinator_server_ip,
coordination::CoordinatorClientConfig{.instance_name = std::string(instance_name),
.ip_address = std::string(coordinator_server_ip),
.port = coordinator_server_port,
.instance_health_check_frequency_sec = instance_check_frequency,
.instance_down_timeout_sec = instance_down_timeout,
@ -524,18 +472,17 @@ class CoordQueryHandler final : public query::CoordinatorQueryHandler {
using enum memgraph::coordination::RegisterInstanceCoordinatorStatus;
case NAME_EXISTS:
throw QueryRuntimeException("Couldn't register replica instance since instance with such name already exists!");
case ENDPOINT_EXISTS:
case COORD_ENDPOINT_EXISTS:
throw QueryRuntimeException(
"Couldn't register replica instance since instance with such endpoint already exists!");
"Couldn't register replica instance since instance with such coordinator endpoint already exists!");
case REPL_ENDPOINT_EXISTS:
throw QueryRuntimeException(
"Couldn't register replica instance since instance with such replication endpoint already exists!");
case NOT_COORDINATOR:
throw QueryRuntimeException("REGISTER INSTANCE query can only be run on a coordinator!");
case NOT_LEADER:
throw QueryRuntimeException("Couldn't register replica instance since coordinator is not a leader!");
case RAFT_COULD_NOT_ACCEPT:
throw QueryRuntimeException(
"Couldn't register replica instance since raft server couldn't accept the log! Most likely the raft "
"instance is not a leader!");
case RAFT_COULD_NOT_APPEND:
case RAFT_LOG_ERROR:
throw QueryRuntimeException("Couldn't register replica instance since raft server couldn't append the log!");
case RPC_FAILED:
throw QueryRuntimeException(
@ -546,19 +493,19 @@ class CoordQueryHandler final : public query::CoordinatorQueryHandler {
}
}
auto AddCoordinatorInstance(uint32_t raft_server_id, std::string const &raft_socket_address) -> void override {
auto const maybe_ip_and_port = io::network::Endpoint::ParseSocketOrIpAddress(raft_socket_address);
auto AddCoordinatorInstance(uint32_t raft_server_id, std::string_view raft_socket_address) -> void override {
auto const maybe_ip_and_port = io::network::Endpoint::ParseSocketOrAddress(raft_socket_address);
if (maybe_ip_and_port) {
auto const [ip, port] = *maybe_ip_and_port;
spdlog::info("Adding instance {} with raft socket address {}:{}.", raft_server_id, port, ip);
spdlog::info("Adding instance {} with raft socket address {}:{}.", raft_server_id, ip, port);
coordinator_handler_.AddCoordinatorInstance(raft_server_id, port, ip);
} else {
spdlog::error("Invalid raft socket address {}.", raft_socket_address);
}
}
void SetReplicationInstanceToMain(const std::string &instance_name) override {
auto status = coordinator_handler_.SetReplicationInstanceToMain(instance_name);
void SetReplicationInstanceToMain(std::string_view instance_name) override {
auto const status = coordinator_handler_.SetReplicationInstanceToMain(instance_name);
switch (status) {
using enum memgraph::coordination::SetInstanceToMainCoordinatorStatus;
case NO_INSTANCE_WITH_NAME:
@ -567,6 +514,10 @@ class CoordQueryHandler final : public query::CoordinatorQueryHandler {
throw QueryRuntimeException("Couldn't set instance to main since there is already a main instance in cluster!");
case NOT_COORDINATOR:
throw QueryRuntimeException("SET INSTANCE TO MAIN query can only be run on a coordinator!");
case NOT_LEADER:
throw QueryRuntimeException("Couldn't set instance to main since coordinator is not a leader!");
case RAFT_LOG_ERROR:
throw QueryRuntimeException("Couldn't promote instance since raft server couldn't append the log!");
case COULD_NOT_PROMOTE_TO_MAIN:
throw QueryRuntimeException(
"Couldn't set replica instance to main! Check coordinator and replica for more logs");
@ -1092,50 +1043,98 @@ Callback HandleReplicationQuery(ReplicationQuery *repl_query, const Parameters &
}
#endif
callback.header = {
"name", "socket_address", "sync_mode", "current_timestamp_of_replica", "number_of_timestamp_behind_master",
"state"};
bool full_info = false;
#ifdef MG_ENTERPRISE
full_info = license::global_license_checker.IsEnterpriseValidFast();
#endif
callback.header = {"name", "socket_address", "sync_mode", "system_info", "data_info"};
callback.fn = [handler = ReplQueryHandler{replication_query_handler}, replica_nfields = callback.header.size(),
db_acc = current_db.db_acc_] {
const auto &replicas = handler.ShowReplicas(*db_acc->get());
full_info] {
auto const sync_mode_to_tv = [](memgraph::replication_coordination_glue::ReplicationMode sync_mode) {
using namespace std::string_view_literals;
switch (sync_mode) {
using enum memgraph::replication_coordination_glue::ReplicationMode;
case SYNC:
return TypedValue{"sync"sv};
case ASYNC:
return TypedValue{"async"sv};
}
};
auto const replica_sys_state_to_tv = [](memgraph::replication::ReplicationClient::State state) {
using namespace std::string_view_literals;
switch (state) {
using enum memgraph::replication::ReplicationClient::State;
case BEHIND:
return TypedValue{"invalid"sv};
case READY:
return TypedValue{"ready"sv};
case RECOVERY:
return TypedValue{"recovery"sv};
}
};
auto const sys_info_to_tv = [&](ReplicaSystemInfoState orig) {
auto info = std::map<std::string, TypedValue>{};
info.emplace("ts", TypedValue{static_cast<int64_t>(orig.ts_)});
// TODO: behind not implemented
info.emplace("behind", TypedValue{/* static_cast<int64_t>(orig.behind_) */});
info.emplace("status", replica_sys_state_to_tv(orig.state_));
return TypedValue{std::move(info)};
};
auto const replica_state_to_tv = [](memgraph::storage::replication::ReplicaState state) {
using namespace std::string_view_literals;
switch (state) {
using enum memgraph::storage::replication::ReplicaState;
case READY:
return TypedValue{"ready"sv};
case REPLICATING:
return TypedValue{"replicating"sv};
case RECOVERY:
return TypedValue{"recovery"sv};
case MAYBE_BEHIND:
return TypedValue{"invalid"sv};
case DIVERGED_FROM_MAIN:
return TypedValue{"diverged"sv};
}
};
auto const info_to_tv = [&](ReplicaInfoState orig) {
auto info = std::map<std::string, TypedValue>{};
info.emplace("ts", TypedValue{static_cast<int64_t>(orig.ts_)});
info.emplace("behind", TypedValue{static_cast<int64_t>(orig.behind_)});
info.emplace("status", replica_state_to_tv(orig.state_));
return TypedValue{std::move(info)};
};
auto const data_info_to_tv = [&](std::map<std::string, ReplicaInfoState> orig) {
auto data_info = std::map<std::string, TypedValue>{};
for (auto &[name, info] : orig) {
data_info.emplace(name, info_to_tv(info));
}
return TypedValue{std::move(data_info)};
};
auto replicas = handler.ShowReplicas();
auto typed_replicas = std::vector<std::vector<TypedValue>>{};
typed_replicas.reserve(replicas.size());
for (const auto &replica : replicas) {
for (auto &replica : replicas) {
std::vector<TypedValue> typed_replica;
typed_replica.reserve(replica_nfields);
typed_replica.emplace_back(replica.name);
typed_replica.emplace_back(replica.socket_address);
switch (replica.sync_mode) {
case ReplicationQuery::SyncMode::SYNC:
typed_replica.emplace_back("sync");
break;
case ReplicationQuery::SyncMode::ASYNC:
typed_replica.emplace_back("async");
break;
}
typed_replica.emplace_back(static_cast<int64_t>(replica.current_timestamp_of_replica));
typed_replica.emplace_back(static_cast<int64_t>(replica.current_number_of_timestamp_behind_master));
switch (replica.state) {
case ReplicationQuery::ReplicaState::READY:
typed_replica.emplace_back("ready");
break;
case ReplicationQuery::ReplicaState::REPLICATING:
typed_replica.emplace_back("replicating");
break;
case ReplicationQuery::ReplicaState::RECOVERY:
typed_replica.emplace_back("recovery");
break;
case ReplicationQuery::ReplicaState::MAYBE_BEHIND:
typed_replica.emplace_back("invalid");
break;
case ReplicationQuery::ReplicaState::DIVERGED_FROM_MAIN:
typed_replica.emplace_back("diverged");
break;
typed_replica.emplace_back(replica.name_);
typed_replica.emplace_back(replica.socket_address_);
typed_replica.emplace_back(sync_mode_to_tv(replica.sync_mode_));
if (full_info) {
typed_replica.emplace_back(sys_info_to_tv(replica.system_info_));
} else {
// Set to NULL
typed_replica.emplace_back(TypedValue{});
}
typed_replica.emplace_back(data_info_to_tv(replica.data_info_));
typed_replicas.emplace_back(std::move(typed_replica));
}
@ -1213,7 +1212,7 @@ Callback HandleCoordinatorQuery(CoordinatorQuery *coordinator_query, const Param
};
notifications->emplace_back(
SeverityLevel::INFO, NotificationCode::REGISTER_COORDINATOR_SERVER,
SeverityLevel::INFO, NotificationCode::REGISTER_REPLICATION_INSTANCE,
fmt::format("Coordinator has registered coordinator server on {} for instance {}.",
coordinator_socket_address_tv.ValueString(), coordinator_query->instance_name_));
return callback;
@ -1255,17 +1254,16 @@ Callback HandleCoordinatorQuery(CoordinatorQuery *coordinator_query, const Param
throw QueryRuntimeException("Only coordinator can run SHOW INSTANCES.");
}
callback.header = {"name", "raft_socket_address", "coordinator_socket_address", "alive", "role"};
callback.header = {"name", "raft_socket_address", "coordinator_socket_address", "health", "role"};
callback.fn = [handler = CoordQueryHandler{*coordinator_state},
replica_nfields = callback.header.size()]() mutable {
auto const instances = handler.ShowInstances();
auto const converter = [](const auto &status) -> std::vector<TypedValue> {
return {TypedValue{status.instance_name}, TypedValue{status.raft_socket_address},
TypedValue{status.coord_socket_address}, TypedValue{status.is_alive},
TypedValue{status.cluster_role}};
TypedValue{status.coord_socket_address}, TypedValue{status.health}, TypedValue{status.cluster_role}};
};
return utils::fmap(converter, instances);
return utils::fmap(instances, converter);
};
return callback;
}
@ -2681,6 +2679,75 @@ PreparedQuery PrepareIndexQuery(ParsedQuery parsed_query, bool in_explicit_trans
RWType::W};
}
PreparedQuery PrepareEdgeIndexQuery(ParsedQuery parsed_query, bool in_explicit_transaction,
std::vector<Notification> *notifications, CurrentDB &current_db) {
if (in_explicit_transaction) {
throw IndexInMulticommandTxException();
}
auto *index_query = utils::Downcast<EdgeIndexQuery>(parsed_query.query);
std::function<void(Notification &)> handler;
MG_ASSERT(current_db.db_acc_, "Index query expects a current DB");
auto &db_acc = *current_db.db_acc_;
MG_ASSERT(current_db.db_transactional_accessor_, "Index query expects a current DB transaction");
auto *dba = &*current_db.execution_db_accessor_;
auto invalidate_plan_cache = [plan_cache = db_acc->plan_cache()] {
plan_cache->WithLock([&](auto &cache) { cache.reset(); });
};
auto *storage = db_acc->storage();
auto edge_type = storage->NameToEdgeType(index_query->edge_type_.name);
Notification index_notification(SeverityLevel::INFO);
switch (index_query->action_) {
case EdgeIndexQuery::Action::CREATE: {
index_notification.code = NotificationCode::CREATE_INDEX;
index_notification.title = fmt::format("Created index on edge-type {}.", index_query->edge_type_.name);
handler = [dba, edge_type, label_name = index_query->edge_type_.name,
invalidate_plan_cache = std::move(invalidate_plan_cache)](Notification &index_notification) {
auto maybe_index_error = dba->CreateIndex(edge_type);
utils::OnScopeExit invalidator(invalidate_plan_cache);
if (maybe_index_error.HasError()) {
index_notification.code = NotificationCode::EXISTENT_INDEX;
index_notification.title = fmt::format("Index on edge-type {} already exists.", label_name);
}
};
break;
}
case EdgeIndexQuery::Action::DROP: {
index_notification.code = NotificationCode::DROP_INDEX;
index_notification.title = fmt::format("Dropped index on edge-type {}.", index_query->edge_type_.name);
handler = [dba, edge_type, label_name = index_query->edge_type_.name,
invalidate_plan_cache = std::move(invalidate_plan_cache)](Notification &index_notification) {
auto maybe_index_error = dba->DropIndex(edge_type);
utils::OnScopeExit invalidator(invalidate_plan_cache);
if (maybe_index_error.HasError()) {
index_notification.code = NotificationCode::NONEXISTENT_INDEX;
index_notification.title = fmt::format("Index on edge-type {} doesn't exist.", label_name);
}
};
break;
}
}
return PreparedQuery{
{},
std::move(parsed_query.required_privileges),
[handler = std::move(handler), notifications, index_notification = std::move(index_notification)](
AnyStream * /*stream*/, std::optional<int> /*unused*/) mutable {
handler(index_notification);
notifications->push_back(index_notification);
return QueryHandlerResult::COMMIT;
},
RWType::W};
}
PreparedQuery PrepareAuthQuery(ParsedQuery parsed_query, bool in_explicit_transaction,
InterpreterContext *interpreter_context, Interpreter &interpreter) {
if (in_explicit_transaction) {
@ -3485,6 +3552,7 @@ PreparedQuery PrepareDatabaseInfoQuery(ParsedQuery parsed_query, bool in_explici
auto *storage = database->storage();
const std::string_view label_index_mark{"label"};
const std::string_view label_property_index_mark{"label+property"};
const std::string_view edge_type_index_mark{"edge-type"};
auto info = dba->ListAllIndices();
auto storage_acc = database->Access();
std::vector<std::vector<TypedValue>> results;
@ -3499,6 +3567,10 @@ PreparedQuery PrepareDatabaseInfoQuery(ParsedQuery parsed_query, bool in_explici
TypedValue(storage->PropertyToName(item.second)),
TypedValue(static_cast<int>(storage_acc->ApproximateVertexCount(item.first, item.second)))});
}
for (const auto &item : info.edge_type) {
results.push_back({TypedValue(edge_type_index_mark), TypedValue(storage->EdgeTypeToName(item)), TypedValue(),
TypedValue(static_cast<int>(storage_acc->ApproximateEdgeCount(item)))});
}
std::sort(results.begin(), results.end(), [&label_index_mark](const auto &record_1, const auto &record_2) {
const auto type_1 = record_1[0].ValueString();
const auto type_2 = record_2[0].ValueString();
@ -4285,13 +4357,14 @@ 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) || utils::Downcast<AnalyzeGraphQuery>(parsed_query.query) ||
utils::Downcast<IndexQuery>(parsed_query.query) || utils::Downcast<DatabaseInfoQuery>(parsed_query.query) ||
utils::Downcast<ConstraintQuery>(parsed_query.query);
utils::Downcast<IndexQuery>(parsed_query.query) || utils::Downcast<EdgeIndexQuery>(parsed_query.query) ||
utils::Downcast<DatabaseInfoQuery>(parsed_query.query) || utils::Downcast<ConstraintQuery>(parsed_query.query);
if (!in_explicit_transaction_ && requires_db_transaction) {
// TODO: ATM only a single database, will change when we have multiple database transactions
bool could_commit = utils::Downcast<CypherQuery>(parsed_query.query) != nullptr;
bool unique = utils::Downcast<IndexQuery>(parsed_query.query) != nullptr ||
utils::Downcast<EdgeIndexQuery>(parsed_query.query) != nullptr ||
utils::Downcast<ConstraintQuery>(parsed_query.query) != nullptr ||
upper_case_query.find(kSchemaAssert) != std::string::npos;
SetupDatabaseTransaction(could_commit, unique);
@ -4328,6 +4401,9 @@ Interpreter::PrepareResult Interpreter::Prepare(const std::string &query_string,
} else if (utils::Downcast<IndexQuery>(parsed_query.query)) {
prepared_query = PrepareIndexQuery(std::move(parsed_query), in_explicit_transaction_,
&query_execution->notifications, current_db_);
} else if (utils::Downcast<EdgeIndexQuery>(parsed_query.query)) {
prepared_query = PrepareEdgeIndexQuery(std::move(parsed_query), in_explicit_transaction_,
&query_execution->notifications, current_db_);
} else if (utils::Downcast<AnalyzeGraphQuery>(parsed_query.query)) {
prepared_query = PrepareAnalyzeGraphQuery(std::move(parsed_query), in_explicit_transaction_, current_db_);
} else if (utils::Downcast<AuthQuery>(parsed_query.query)) {

23
src/query/interpreter.hpp
View File

@ -84,16 +84,6 @@ class CoordinatorQueryHandler {
CoordinatorQueryHandler(CoordinatorQueryHandler &&) = default;
CoordinatorQueryHandler &operator=(CoordinatorQueryHandler &&) = default;
struct Replica {
std::string name;
std::string socket_address;
ReplicationQuery::SyncMode sync_mode;
std::optional<double> timeout;
uint64_t current_timestamp_of_replica;
uint64_t current_number_of_timestamp_behind_master;
ReplicationQuery::ReplicaState state;
};
struct MainReplicaStatus {
std::string_view name;
std::string_view socket_address;
@ -105,25 +95,24 @@ class CoordinatorQueryHandler {
};
/// @throw QueryRuntimeException if an error ocurred.
virtual void RegisterReplicationInstance(std::string const &coordinator_socket_address,
std::string const &replication_socket_address,
virtual void RegisterReplicationInstance(std::string_view coordinator_socket_address,
std::string_view replication_socket_address,
std::chrono::seconds const &instance_health_check_frequency,
std::chrono::seconds const &instance_down_timeout,
std::chrono::seconds const &instance_get_uuid_frequency,
std::string const &instance_name, CoordinatorQuery::SyncMode sync_mode) = 0;
std::string_view instance_name, CoordinatorQuery::SyncMode sync_mode) = 0;
/// @throw QueryRuntimeException if an error ocurred.
virtual void UnregisterInstance(std::string const &instance_name) = 0;
virtual void UnregisterInstance(std::string_view instance_name) = 0;
/// @throw QueryRuntimeException if an error ocurred.
virtual void SetReplicationInstanceToMain(const std::string &instance_name) = 0;
virtual void SetReplicationInstanceToMain(std::string_view instance_name) = 0;
/// @throw QueryRuntimeException if an error ocurred.
virtual std::vector<coordination::InstanceStatus> ShowInstances() const = 0;
/// @throw QueryRuntimeException if an error ocurred.
virtual auto AddCoordinatorInstance(uint32_t raft_server_id, std::string const &coordinator_socket_address)
-> void = 0;
virtual auto AddCoordinatorInstance(uint32_t raft_server_id, std::string_view coordinator_socket_address) -> void = 0;
};
#endif

4
src/query/metadata.cpp
View File

@ -67,8 +67,8 @@ constexpr std::string_view GetCodeString(const NotificationCode code) {
case NotificationCode::REGISTER_REPLICA:
return "RegisterReplica"sv;
#ifdef MG_ENTERPRISE
case NotificationCode::REGISTER_COORDINATOR_SERVER:
return "RegisterCoordinatorServer"sv;
case NotificationCode::REGISTER_REPLICATION_INSTANCE:
return "RegisterReplicationInstance"sv;
case NotificationCode::ADD_COORDINATOR_INSTANCE:
return "AddCoordinatorInstance"sv;
case NotificationCode::UNREGISTER_INSTANCE:

2
src/query/metadata.hpp
View File

@ -43,7 +43,7 @@ enum class NotificationCode : uint8_t {
REPLICA_PORT_WARNING,
REGISTER_REPLICA,
#ifdef MG_ENTERPRISE
REGISTER_COORDINATOR_SERVER, // TODO: (andi) What is this?
REGISTER_REPLICATION_INSTANCE,
ADD_COORDINATOR_INSTANCE,
UNREGISTER_INSTANCE,
#endif

13
src/query/plan/hint_provider.hpp
View File

@ -1,4 +1,4 @@
// Copyright 2023 Memgraph Ltd.
// Copyright 2024 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
@ -114,6 +114,9 @@ class PlanHintsProvider final : public HierarchicalLogicalOperatorVisitor {
bool PreVisit(ScanAllById & /*unused*/) override { return true; }
bool PostVisit(ScanAllById & /*unused*/) override { return true; }
bool PreVisit(ScanAllByEdgeType & /*unused*/) override { return true; }
bool PostVisit(ScanAllByEdgeType & /*unused*/) override { return true; }
bool PreVisit(ConstructNamedPath & /*unused*/) override { return true; }
bool PostVisit(ConstructNamedPath & /*unused*/) override { return true; }
@ -206,6 +209,14 @@ class PlanHintsProvider final : public HierarchicalLogicalOperatorVisitor {
bool PostVisit(IndexedJoin & /*unused*/) override { return true; }
bool PreVisit(RollUpApply &op) override {
op.input()->Accept(*this);
op.list_collection_branch_->Accept(*this);
return false;
}
bool PostVisit(RollUpApply & /*unused*/) override { return true; }
private:
const SymbolTable &symbol_table_;
std::vector<std::string> hints_;

103
src/query/plan/operator.cpp
View File

@ -105,6 +105,7 @@ extern const Event ScanAllByLabelPropertyRangeOperator;
extern const Event ScanAllByLabelPropertyValueOperator;
extern const Event ScanAllByLabelPropertyOperator;
extern const Event ScanAllByIdOperator;
extern const Event ScanAllByEdgeTypeOperator;
extern const Event ExpandOperator;
extern const Event ExpandVariableOperator;
extern const Event ConstructNamedPathOperator;
@ -517,6 +518,60 @@ class ScanAllCursor : public Cursor {
const char *op_name_;
};
template <typename TEdgesFun>
class ScanAllByEdgeTypeCursor : public Cursor {
public:
explicit ScanAllByEdgeTypeCursor(const ScanAllByEdgeType &self, Symbol output_symbol, UniqueCursorPtr input_cursor,
storage::View view, TEdgesFun get_edges, const char *op_name)
: self_(self),
output_symbol_(std::move(output_symbol)),
input_cursor_(std::move(input_cursor)),
view_(view),
get_edges_(std::move(get_edges)),
op_name_(op_name) {}
bool Pull(Frame &frame, ExecutionContext &context) override {
OOMExceptionEnabler oom_exception;
SCOPED_PROFILE_OP_BY_REF(self_);
AbortCheck(context);
while (!vertices_ || vertices_it_.value() == vertices_end_it_.value()) {
if (!input_cursor_->Pull(frame, context)) return false;
auto next_vertices = get_edges_(frame, context);
if (!next_vertices) continue;
vertices_.emplace(std::move(next_vertices.value()));
vertices_it_.emplace(vertices_.value().begin());
vertices_end_it_.emplace(vertices_.value().end());
}
frame[output_symbol_] = *vertices_it_.value();
++vertices_it_.value();
return true;
}
void Shutdown() override { input_cursor_->Shutdown(); }
void Reset() override {
input_cursor_->Reset();
vertices_ = std::nullopt;
vertices_it_ = std::nullopt;
vertices_end_it_ = std::nullopt;
}
private:
const ScanAllByEdgeType &self_;
const Symbol output_symbol_;
const UniqueCursorPtr input_cursor_;
storage::View view_;
TEdgesFun get_edges_;
std::optional<typename std::result_of<TEdgesFun(Frame &, ExecutionContext &)>::type::value_type> vertices_;
std::optional<decltype(vertices_.value().begin())> vertices_it_;
std::optional<decltype(vertices_.value().end())> vertices_end_it_;
const char *op_name_;
};
ScanAll::ScanAll(const std::shared_ptr<LogicalOperator> &input, Symbol output_symbol, storage::View view)
: input_(input ? input : std::make_shared<Once>()), output_symbol_(std::move(output_symbol)), view_(view) {}
@ -556,6 +611,33 @@ UniqueCursorPtr ScanAllByLabel::MakeCursor(utils::MemoryResource *mem) const {
view_, std::move(vertices), "ScanAllByLabel");
}
ScanAllByEdgeType::ScanAllByEdgeType(const std::shared_ptr<LogicalOperator> &input, Symbol output_symbol,
storage::EdgeTypeId edge_type, storage::View view)
: input_(input ? input : std::make_shared<Once>()),
output_symbol_(std::move(output_symbol)),
view_(view),
edge_type_(edge_type) {}
ACCEPT_WITH_INPUT(ScanAllByEdgeType)
UniqueCursorPtr ScanAllByEdgeType::MakeCursor(utils::MemoryResource *mem) const {
memgraph::metrics::IncrementCounter(memgraph::metrics::ScanAllByEdgeTypeOperator);
auto edges = [this](Frame &, ExecutionContext &context) {
auto *db = context.db_accessor;
return std::make_optional(db->Edges(view_, edge_type_));
};
return MakeUniqueCursorPtr<ScanAllByEdgeTypeCursor<decltype(edges)>>(
mem, *this, output_symbol_, input_->MakeCursor(mem), view_, std::move(edges), "ScanAllByEdgeType");
}
std::vector<Symbol> ScanAllByEdgeType::ModifiedSymbols(const SymbolTable &table) const {
auto symbols = input_->ModifiedSymbols(table);
symbols.emplace_back(output_symbol_);
return symbols;
}
// TODO(buda): Implement ScanAllByLabelProperty operator to iterate over
// vertices that have the label and some value for the given property.
@ -5624,4 +5706,25 @@ UniqueCursorPtr HashJoin::MakeCursor(utils::MemoryResource *mem) const {
return MakeUniqueCursorPtr<HashJoinCursor>(mem, *this, mem);
}
RollUpApply::RollUpApply(const std::shared_ptr<LogicalOperator> &input,
std::shared_ptr<LogicalOperator> &&second_branch)
: input_(input), list_collection_branch_(second_branch) {}
std::vector<Symbol> RollUpApply::OutputSymbols(const SymbolTable & /*symbol_table*/) const {
std::vector<Symbol> symbols;
return symbols;
}
std::vector<Symbol> RollUpApply::ModifiedSymbols(const SymbolTable &table) const { return OutputSymbols(table); }
bool RollUpApply::Accept(HierarchicalLogicalOperatorVisitor &visitor) {
if (visitor.PreVisit(*this)) {
if (!input_ || !list_collection_branch_) {
throw utils::NotYetImplemented("One of the branches in pattern comprehension is null! Please contact support.");
}
input_->Accept(visitor) && list_collection_branch_->Accept(visitor);
}
return visitor.PostVisit(*this);
}
} // namespace memgraph::query::plan

83
src/query/plan/operator.hpp
View File

@ -1,4 +1,4 @@
// Copyright 2023 Memgraph Ltd.
// Copyright 2024 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
@ -99,6 +99,7 @@ class ScanAllByLabelPropertyRange;
class ScanAllByLabelPropertyValue;
class ScanAllByLabelProperty;
class ScanAllById;
class ScanAllByEdgeType;
class Expand;
class ExpandVariable;
class ConstructNamedPath;
@ -130,14 +131,15 @@ class EvaluatePatternFilter;
class Apply;
class IndexedJoin;
class HashJoin;
class RollUpApply;
using LogicalOperatorCompositeVisitor =
utils::CompositeVisitor<Once, CreateNode, CreateExpand, ScanAll, ScanAllByLabel, ScanAllByLabelPropertyRange,
ScanAllByLabelPropertyValue, ScanAllByLabelProperty, ScanAllById, Expand, ExpandVariable,
ConstructNamedPath, Filter, Produce, Delete, SetProperty, SetProperties, SetLabels,
RemoveProperty, RemoveLabels, EdgeUniquenessFilter, Accumulate, Aggregate, Skip, Limit,
OrderBy, Merge, Optional, Unwind, Distinct, Union, Cartesian, CallProcedure, LoadCsv,
Foreach, EmptyResult, EvaluatePatternFilter, Apply, IndexedJoin, HashJoin>;
ScanAllByLabelPropertyValue, ScanAllByLabelProperty, ScanAllById, ScanAllByEdgeType, Expand,
ExpandVariable, ConstructNamedPath, Filter, Produce, Delete, SetProperty, SetProperties,
SetLabels, RemoveProperty, RemoveLabels, EdgeUniquenessFilter, Accumulate, Aggregate, Skip,
Limit, OrderBy, Merge, Optional, Unwind, Distinct, Union, Cartesian, CallProcedure, LoadCsv,
Foreach, EmptyResult, EvaluatePatternFilter, Apply, IndexedJoin, HashJoin, RollUpApply>;
using LogicalOperatorLeafVisitor = utils::LeafVisitor<Once>;
@ -591,6 +593,42 @@ class ScanAllByLabel : public memgraph::query::plan::ScanAll {
}
};
class ScanAllByEdgeType : public memgraph::query::plan::LogicalOperator {
public:
static const utils::TypeInfo kType;
const utils::TypeInfo &GetTypeInfo() const override { return kType; }
ScanAllByEdgeType() = default;
ScanAllByEdgeType(const std::shared_ptr<LogicalOperator> &input, Symbol output_symbol, storage::EdgeTypeId edge_type,
storage::View view = storage::View::OLD);
bool Accept(HierarchicalLogicalOperatorVisitor &visitor) override;
UniqueCursorPtr MakeCursor(utils::MemoryResource *) const override;
std::vector<Symbol> ModifiedSymbols(const SymbolTable &) const override;
bool HasSingleInput() const override { return true; }
std::shared_ptr<LogicalOperator> input() const override { return input_; }
void set_input(std::shared_ptr<LogicalOperator> input) override { input_ = input; }
std::string ToString() const override {
return fmt::format("ScanAllByEdgeType ({} :{})", output_symbol_.name(), dba_->EdgeTypeToName(edge_type_));
}
std::shared_ptr<memgraph::query::plan::LogicalOperator> input_;
Symbol output_symbol_;
storage::View view_;
storage::EdgeTypeId edge_type_;
std::unique_ptr<LogicalOperator> Clone(AstStorage *storage) const override {
auto object = std::make_unique<ScanAllByEdgeType>();
object->input_ = input_ ? input_->Clone(storage) : nullptr;
object->output_symbol_ = output_symbol_;
object->view_ = view_;
object->edge_type_ = edge_type_;
return object;
}
};
/// Behaves like @c ScanAll, but produces only vertices with given label and
/// property value which is inside a range (inclusive or exlusive).
///
@ -2634,5 +2672,38 @@ class HashJoin : public memgraph::query::plan::LogicalOperator {
}
};
/// RollUpApply operator is used to execute an expression which takes as input a pattern,
/// and returns a list with content from the matched pattern
/// It's used for a pattern expression or pattern comprehension in a query.
class RollUpApply : public memgraph::query::plan::LogicalOperator {
public:
static const utils::TypeInfo kType;
const utils::TypeInfo &GetTypeInfo() const override { return kType; }
RollUpApply() = default;
RollUpApply(const std::shared_ptr<LogicalOperator> &input, std::shared_ptr<LogicalOperator> &&second_branch);
bool HasSingleInput() const override { return false; }
std::shared_ptr<LogicalOperator> input() const override { return input_; }
void set_input(std::shared_ptr<LogicalOperator> input) override { input_ = input; }
bool Accept(HierarchicalLogicalOperatorVisitor &visitor) override;
UniqueCursorPtr MakeCursor(utils::MemoryResource *) const override {
throw utils::NotYetImplemented("Execution of Pattern comprehension is currently unsupported.");
}
std::vector<Symbol> OutputSymbols(const SymbolTable &) const override;
std::vector<Symbol> ModifiedSymbols(const SymbolTable &) const override;
std::unique_ptr<LogicalOperator> Clone(AstStorage *storage) const override {
auto object = std::make_unique<RollUpApply>();
object->input_ = input_ ? input_->Clone(storage) : nullptr;
object->list_collection_branch_ = list_collection_branch_ ? list_collection_branch_->Clone(storage) : nullptr;
return object;
}
std::shared_ptr<memgraph::query::plan::LogicalOperator> input_;
std::shared_ptr<memgraph::query::plan::LogicalOperator> list_collection_branch_;
};
} // namespace plan
} // namespace memgraph::query

7
src/query/plan/operator_type_info.cpp
View File

@ -1,4 +1,4 @@
// Copyright 2023 Memgraph Ltd.
// Copyright 2024 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
@ -49,6 +49,8 @@ constexpr utils::TypeInfo query::plan::ScanAllByLabelProperty::kType{
constexpr utils::TypeInfo query::plan::ScanAllById::kType{utils::TypeId::SCAN_ALL_BY_ID, "ScanAllById",
&query::plan::ScanAll::kType};
constexpr utils::TypeInfo query::plan::ScanAllByEdgeType::kType{utils::TypeId::SCAN_ALL_BY_EDGE_TYPE,
"ScanAllByEdgeType", &query::plan::ScanAll::kType};
constexpr utils::TypeInfo query::plan::ExpandCommon::kType{utils::TypeId::EXPAND_COMMON, "ExpandCommon", nullptr};
@ -154,4 +156,7 @@ constexpr utils::TypeInfo query::plan::IndexedJoin::kType{utils::TypeId::INDEXED
constexpr utils::TypeInfo query::plan::HashJoin::kType{utils::TypeId::HASH_JOIN, "HashJoin",
&query::plan::LogicalOperator::kType};
constexpr utils::TypeInfo query::plan::RollUpApply::kType{utils::TypeId::ROLLUP_APPLY, "RollUpApply",
&query::plan::LogicalOperator::kType};
} // namespace memgraph

8
src/query/plan/planner.hpp
View File

@ -23,6 +23,7 @@
#include "query/plan/operator.hpp"
#include "query/plan/preprocess.hpp"
#include "query/plan/pretty_print.hpp"
#include "query/plan/rewrite/edge_type_index_lookup.hpp"
#include "query/plan/rewrite/index_lookup.hpp"
#include "query/plan/rewrite/join.hpp"
#include "query/plan/rule_based_planner.hpp"
@ -54,8 +55,11 @@ class PostProcessor final {
std::unique_ptr<LogicalOperator> Rewrite(std::unique_ptr<LogicalOperator> plan, TPlanningContext *context) {
auto index_lookup_plan =
RewriteWithIndexLookup(std::move(plan), context->symbol_table, context->ast_storage, context->db, index_hints_);
return RewriteWithJoinRewriter(std::move(index_lookup_plan), context->symbol_table, context->ast_storage,
context->db);
auto join_plan =
RewriteWithJoinRewriter(std::move(index_lookup_plan), context->symbol_table, context->ast_storage, context->db);
auto edge_index_plan = RewriteWithEdgeTypeIndexRewriter(std::move(join_plan), context->symbol_table,
context->ast_storage, context->db);
return edge_index_plan;
}
template <class TVertexCounts>

49
src/query/plan/preprocess.cpp
View File

@ -632,20 +632,20 @@ void AddMatching(const Match &match, SymbolTable &symbol_table, AstStorage &stor
// If there are any pattern filters, we add those as well
for (auto &filter : matching.filters) {
PatternFilterVisitor visitor(symbol_table, storage);
PatternVisitor visitor(symbol_table, storage);
filter.expression->Accept(visitor);
filter.matchings = visitor.getMatchings();
filter.matchings = visitor.getFilterMatchings();
}
}
PatternFilterVisitor::PatternFilterVisitor(SymbolTable &symbol_table, AstStorage &storage)
PatternVisitor::PatternVisitor(SymbolTable &symbol_table, AstStorage &storage)
: symbol_table_(symbol_table), storage_(storage) {}
PatternFilterVisitor::PatternFilterVisitor(const PatternFilterVisitor &) = default;
PatternFilterVisitor::PatternFilterVisitor(PatternFilterVisitor &&) noexcept = default;
PatternFilterVisitor::~PatternFilterVisitor() = default;
PatternVisitor::PatternVisitor(const PatternVisitor &) = default;
PatternVisitor::PatternVisitor(PatternVisitor &&) noexcept = default;
PatternVisitor::~PatternVisitor() = default;
void PatternFilterVisitor::Visit(Exists &op) {
void PatternVisitor::Visit(Exists &op) {
std::vector<Pattern *> patterns;
patterns.push_back(op.pattern_);
@ -655,10 +655,14 @@ void PatternFilterVisitor::Visit(Exists &op) {
filter_matching.type = PatternFilterType::EXISTS;
filter_matching.symbol = std::make_optional<Symbol>(symbol_table_.at(op));
matchings_.push_back(std::move(filter_matching));
filter_matchings_.push_back(std::move(filter_matching));
}
std::vector<FilterMatching> PatternFilterVisitor::getMatchings() { return matchings_; }
std::vector<FilterMatching> PatternVisitor::getFilterMatchings() { return filter_matchings_; }
std::vector<PatternComprehensionMatching> PatternVisitor::getPatternComprehensionMatchings() {
return pattern_comprehension_matchings_;
}
static void ParseForeach(query::Foreach &foreach, SingleQueryPart &query_part, AstStorage &storage,
SymbolTable &symbol_table) {
@ -672,6 +676,30 @@ static void ParseForeach(query::Foreach &foreach, SingleQueryPart &query_part, A
}
}
static void ParseReturn(query::Return &ret, AstStorage &storage, SymbolTable &symbol_table,
std::unordered_map<std::string, PatternComprehensionMatching> &matchings) {
PatternVisitor visitor(symbol_table, storage);
for (auto *expr : ret.body_.named_expressions) {
expr->Accept(visitor);
auto pattern_comprehension_matchings = visitor.getPatternComprehensionMatchings();
for (auto &matching : pattern_comprehension_matchings) {
matchings.emplace(expr->name_, matching);
}
}
}
void PatternVisitor::Visit(NamedExpression &op) { op.expression_->Accept(*this); }
void PatternVisitor::Visit(PatternComprehension &op) {
PatternComprehensionMatching matching;
AddMatching({op.pattern_}, op.filter_, symbol_table_, storage_, matching);
matching.result_expr = storage_.Create<NamedExpression>(symbol_table_.at(op).name(), op.resultExpr_);
matching.result_expr->MapTo(symbol_table_.at(op));
pattern_comprehension_matchings_.push_back(std::move(matching));
}
// Converts a Query to multiple QueryParts. In the process new Ast nodes may be
// created, e.g. filter expressions.
std::vector<SingleQueryPart> CollectSingleQueryParts(SymbolTable &symbol_table, AstStorage &storage,
@ -703,7 +731,8 @@ std::vector<SingleQueryPart> CollectSingleQueryParts(SymbolTable &symbol_table,
// This query part is done, continue with a new one.
query_parts.emplace_back(SingleQueryPart{});
query_part = &query_parts.back();
} else if (utils::IsSubtype(*clause, Return::kType)) {
} else if (auto *ret = utils::Downcast<Return>(clause)) {
ParseReturn(*ret, storage, symbol_table, query_part->pattern_comprehension_matchings);
return query_parts;
}
}

42
src/query/plan/preprocess.hpp
View File

@ -153,19 +153,20 @@ struct Expansion {
ExpansionGroupId expansion_group_id = ExpansionGroupId();
};
struct PatternComprehensionMatching;
struct FilterMatching;
enum class PatternFilterType { EXISTS };
/// Collects matchings from filters that include patterns
class PatternFilterVisitor : public ExpressionVisitor<void> {
/// Collects matchings that include patterns
class PatternVisitor : public ExpressionVisitor<void> {
public:
explicit PatternFilterVisitor(SymbolTable &symbol_table, AstStorage &storage);
PatternFilterVisitor(const PatternFilterVisitor &);
PatternFilterVisitor &operator=(const PatternFilterVisitor &) = delete;
PatternFilterVisitor(PatternFilterVisitor &&) noexcept;
PatternFilterVisitor &operator=(PatternFilterVisitor &&) noexcept = delete;
~PatternFilterVisitor() override;
explicit PatternVisitor(SymbolTable &symbol_table, AstStorage &storage);
PatternVisitor(const PatternVisitor &);
PatternVisitor &operator=(const PatternVisitor &) = delete;
PatternVisitor(PatternVisitor &&) noexcept;
PatternVisitor &operator=(PatternVisitor &&) noexcept = delete;
~PatternVisitor() override;
using ExpressionVisitor<void>::Visit;
@ -233,18 +234,22 @@ class PatternFilterVisitor : public ExpressionVisitor<void> {
void Visit(PropertyLookup &op) override{};
void Visit(AllPropertiesLookup &op) override{};
void Visit(ParameterLookup &op) override{};
void Visit(NamedExpression &op) override{};
void Visit(RegexMatch &op) override{};
void Visit(PatternComprehension &op) override{};
void Visit(NamedExpression &op) override;
void Visit(PatternComprehension &op) override;
std::vector<FilterMatching> getMatchings();
std::vector<FilterMatching> getFilterMatchings();
std::vector<PatternComprehensionMatching> getPatternComprehensionMatchings();
SymbolTable &symbol_table_;
AstStorage &storage_;
private:
/// Collection of matchings in the filter expression being analyzed.
std::vector<FilterMatching> matchings_;
std::vector<FilterMatching> filter_matchings_;
/// Collection of matchings in the pattern comprehension being analyzed.
std::vector<PatternComprehensionMatching> pattern_comprehension_matchings_;
};
/// Stores the symbols and expression used to filter a property.
@ -495,6 +500,11 @@ inline auto Filters::IdFilters(const Symbol &symbol) const -> std::vector<Filter
return filters;
}
struct PatternComprehensionMatching : Matching {
/// Pattern comprehension result named expression
NamedExpression *result_expr = nullptr;
};
/// @brief Represents a read (+ write) part of a query. Parts are split on
/// `WITH` clauses.
///
@ -537,6 +547,14 @@ struct SingleQueryPart {
/// in the `remaining_clauses` but rather in the `Foreach` itself and are guranteed
/// to be processed in the same order by the semantics of the `RuleBasedPlanner`.
std::vector<Matching> merge_matching{};
/// @brief @c NamedExpression name to @c PatternComprehensionMatching for each pattern comprehension.
///
/// Storing the normalized pattern of a @c PatternComprehension does not preclude storing the
/// @c PatternComprehension clause itself inside `remaining_clauses`. The reason is that we
/// need to have access to other parts of the clause, such as pattern, filter clauses.
std::unordered_map<std::string, PatternComprehensionMatching> pattern_comprehension_matchings{};
/// @brief All the remaining clauses (without @c Match).
std::vector<Clause *> remaining_clauses{};
/// The subqueries vector are all the subqueries in this query part ordered in a list by

29
src/query/plan/pretty_print.cpp
View File

@ -1,4 +1,4 @@
// Copyright 2023 Memgraph Ltd.
// Copyright 2024 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
@ -76,6 +76,13 @@ bool PlanPrinter::PreVisit(ScanAllById &op) {
return true;
}
bool PlanPrinter::PreVisit(query::plan::ScanAllByEdgeType &op) {
op.dba_ = dba_;
WithPrintLn([&op](auto &out) { out << "* " << op.ToString(); });
op.dba_ = nullptr;
return true;
}
bool PlanPrinter::PreVisit(query::plan::Expand &op) {
op.dba_ = dba_;
WithPrintLn([&op](auto &out) { out << "* " << op.ToString(); });
@ -143,6 +150,13 @@ bool PlanPrinter::PreVisit(query::plan::Union &op) {
return false;
}
bool PlanPrinter::PreVisit(query::plan::RollUpApply &op) {
WithPrintLn([&op](auto &out) { out << "* " << op.ToString(); });
Branch(*op.list_collection_branch_);
op.input_->Accept(*this);
return false;
}
bool PlanPrinter::PreVisit(query::plan::CallProcedure &op) {
WithPrintLn([&op](auto &out) { out << "* " << op.ToString(); });
return true;
@ -457,6 +471,19 @@ bool PlanToJsonVisitor::PreVisit(ScanAllById &op) {
return false;
}
bool PlanToJsonVisitor::PreVisit(ScanAllByEdgeType &op) {
json self;
self["name"] = "ScanAllByEdgeType";
self["edge_type"] = ToJson(op.edge_type_, *dba_);
self["output_symbol"] = ToJson(op.output_symbol_);
op.input_->Accept(*this);
self["input"] = PopOutput();
output_ = std::move(self);
return false;
}
bool PlanToJsonVisitor::PreVisit(CreateNode &op) {
json self;
self["name"] = "CreateNode";

5
src/query/plan/pretty_print.hpp
View File

@ -1,4 +1,4 @@
// Copyright 2023 Memgraph Ltd.
// Copyright 2024 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
@ -67,6 +67,7 @@ class PlanPrinter : public virtual HierarchicalLogicalOperatorVisitor {
bool PreVisit(ScanAllByLabelPropertyRange &) override;
bool PreVisit(ScanAllByLabelProperty &) override;
bool PreVisit(ScanAllById &) override;
bool PreVisit(ScanAllByEdgeType &) override;
bool PreVisit(Expand &) override;
bool PreVisit(ExpandVariable &) override;
@ -91,6 +92,7 @@ class PlanPrinter : public virtual HierarchicalLogicalOperatorVisitor {
bool PreVisit(OrderBy &) override;
bool PreVisit(Distinct &) override;
bool PreVisit(Union &) override;
bool PreVisit(RollUpApply &) override;
bool PreVisit(Unwind &) override;
bool PreVisit(CallProcedure &) override;
@ -203,6 +205,7 @@ class PlanToJsonVisitor : public virtual HierarchicalLogicalOperatorVisitor {
bool PreVisit(ScanAllByLabelPropertyValue &) override;
bool PreVisit(ScanAllByLabelProperty &) override;
bool PreVisit(ScanAllById &) override;
bool PreVisit(ScanAllByEdgeType &) override;
bool PreVisit(EmptyResult &) override;
bool PreVisit(Produce &) override;

534
src/query/plan/rewrite/edge_type_index_lookup.hpp Normal file
View File

@ -0,0 +1,534 @@
// Copyright 2024 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
/// This file provides a plan rewriter which replaces `ScanAll` and `Expand`
/// operations with `ScanAllByEdgeType` if possible. The public entrypoint is
/// `RewriteWithEdgeTypeIndexRewriter`.
#pragma once
#include <algorithm>
#include <memory>
#include <optional>
#include <unordered_map>
#include <unordered_set>
#include <vector>
#include <gflags/gflags.h>
#include "query/plan/operator.hpp"
#include "query/plan/preprocess.hpp"
#include "query/plan/rewrite/index_lookup.hpp"
#include "utils/algorithm.hpp"
namespace memgraph::query::plan {
namespace impl {
template <class TDbAccessor>
class EdgeTypeIndexRewriter final : public HierarchicalLogicalOperatorVisitor {
public:
EdgeTypeIndexRewriter(SymbolTable *symbol_table, AstStorage *ast_storage, TDbAccessor *db)
: symbol_table_(symbol_table), ast_storage_(ast_storage), db_(db) {}
using HierarchicalLogicalOperatorVisitor::PostVisit;
using HierarchicalLogicalOperatorVisitor::PreVisit;
using HierarchicalLogicalOperatorVisitor::Visit;
bool Visit(Once &) override { return true; }
bool PreVisit(Filter &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(Filter & /*op*/) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(ScanAll &op) override {
prev_ops_.push_back(&op);
if (op.input()->GetTypeInfo() == Once::kType) {
const bool is_node_anon = op.output_symbol_.IsSymbolAnonym();
once_under_scanall_ = is_node_anon;
}
return true;
}
bool PostVisit(ScanAll &op) override {
prev_ops_.pop_back();
if (EdgeTypeIndexingPossible()) {
SetOnParent(op.input());
}
return true;
}
bool PreVisit(Expand &op) override {
prev_ops_.push_back(&op);
if (op.input()->GetTypeInfo() == ScanAll::kType) {
const bool only_one_edge_type = (op.common_.edge_types.size() == 1U);
const bool expansion_is_named = !(op.common_.edge_symbol.IsSymbolAnonym());
const bool expdanded_node_not_named = op.common_.node_symbol.IsSymbolAnonym();
edge_type_index_exist = only_one_edge_type ? db_->EdgeTypeIndexExists(op.common_.edge_types.front()) : false;
scanall_under_expand_ = only_one_edge_type && expansion_is_named && expdanded_node_not_named;
}
return true;
}
bool PostVisit(Expand &op) override {
prev_ops_.pop_back();
if (EdgeTypeIndexingPossible()) {
auto indexed_scan = GenEdgeTypeScan(op);
SetOnParent(std::move(indexed_scan));
}
return true;
}
bool PreVisit(ExpandVariable &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(ExpandVariable &expand) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(Merge &op) override {
prev_ops_.push_back(&op);
op.input()->Accept(*this);
RewriteBranch(&op.merge_match_);
return false;
}
bool PostVisit(Merge &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(Optional &op) override {
prev_ops_.push_back(&op);
op.input()->Accept(*this);
RewriteBranch(&op.optional_);
return false;
}
bool PostVisit(Optional &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(Cartesian &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(Cartesian &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(IndexedJoin &op) override {
prev_ops_.push_back(&op);
RewriteBranch(&op.main_branch_);
RewriteBranch(&op.sub_branch_);
return false;
}
bool PostVisit(IndexedJoin &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(HashJoin &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(HashJoin &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(Union &op) override {
prev_ops_.push_back(&op);
RewriteBranch(&op.left_op_);
RewriteBranch(&op.right_op_);
return false;
}
bool PostVisit(Union &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(CreateNode &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(CreateNode &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(CreateExpand &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(CreateExpand &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(ScanAllByLabel &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(ScanAllByLabel &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(ScanAllByLabelPropertyRange &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(ScanAllByLabelPropertyRange &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(ScanAllByLabelPropertyValue &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(ScanAllByLabelPropertyValue &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(ScanAllByLabelProperty &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(ScanAllByLabelProperty &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(ScanAllById &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(ScanAllById &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(ScanAllByEdgeType &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(ScanAllByEdgeType &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(ConstructNamedPath &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(ConstructNamedPath &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(Produce &op) override {
prev_ops_.push_back(&op);
if (op.input()->GetTypeInfo() == Expand::kType) {
expand_under_produce_ = true;
}
return true;
}
bool PostVisit(Produce &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(EmptyResult &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(EmptyResult &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(Delete &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(Delete &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(SetProperty &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(SetProperty &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(SetProperties &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(SetProperties &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(SetLabels &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(SetLabels &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(RemoveProperty &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(RemoveProperty &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(RemoveLabels &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(RemoveLabels &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(EdgeUniquenessFilter &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(EdgeUniquenessFilter &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(Accumulate &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(Accumulate &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(Aggregate &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(Aggregate &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(Skip &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(Skip &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(Limit &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(Limit &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(OrderBy &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(OrderBy &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(Unwind &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(Unwind &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(Distinct &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(Distinct &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(CallProcedure &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(CallProcedure &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(Foreach &op) override {
prev_ops_.push_back(&op);
op.input()->Accept(*this);
RewriteBranch(&op.update_clauses_);
return false;
}
bool PostVisit(Foreach &) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(EvaluatePatternFilter &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(EvaluatePatternFilter & /*op*/) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(Apply &op) override {
prev_ops_.push_back(&op);
op.input()->Accept(*this);
RewriteBranch(&op.subquery_);
return false;
}
bool PostVisit(Apply & /*op*/) override {
prev_ops_.pop_back();
return true;
}
bool PreVisit(LoadCsv &op) override {
prev_ops_.push_back(&op);
return true;
}
bool PostVisit(LoadCsv & /*op*/) override {
prev_ops_.pop_back();
return true;
}
std::shared_ptr<LogicalOperator> new_root_;
private:
SymbolTable *symbol_table_;
AstStorage *ast_storage_;
TDbAccessor *db_;
// Collected filters, pending for examination if they can be used for advanced
// lookup operations (by index, node ID, ...).
Filters filters_;
// Expressions which no longer need a plain Filter operator.
std::unordered_set<Expression *> filter_exprs_for_removal_;
std::vector<LogicalOperator *> prev_ops_;
std::unordered_set<Symbol> cartesian_symbols_;
bool EdgeTypeIndexingPossible() const {
return expand_under_produce_ && scanall_under_expand_ && once_under_scanall_ && edge_type_index_exist;
}
bool expand_under_produce_ = false;
bool scanall_under_expand_ = false;
bool once_under_scanall_ = false;
bool edge_type_index_exist = false;
bool DefaultPreVisit() override {
throw utils::NotYetImplemented("Operator not yet covered by EdgeTypeIndexRewriter");
}
std::unique_ptr<ScanAllByEdgeType> GenEdgeTypeScan(const Expand &expand) {
const auto &input = expand.input();
const auto &output_symbol = expand.common_.edge_symbol;
const auto &view = expand.view_;
// Extract edge_type from symbol
auto edge_type = expand.common_.edge_types.front();
return std::make_unique<ScanAllByEdgeType>(input, output_symbol, edge_type, view);
}
void SetOnParent(const std::shared_ptr<LogicalOperator> &input) {
MG_ASSERT(input);
if (prev_ops_.empty()) {
MG_ASSERT(!new_root_);
new_root_ = input;
return;
}
prev_ops_.back()->set_input(input);
}
void RewriteBranch(std::shared_ptr<LogicalOperator> *branch) {
EdgeTypeIndexRewriter<TDbAccessor> rewriter(symbol_table_, ast_storage_, db_);
(*branch)->Accept(rewriter);
if (rewriter.new_root_) {
*branch = rewriter.new_root_;
}
}
};
} // namespace impl
template <class TDbAccessor>
std::unique_ptr<LogicalOperator> RewriteWithEdgeTypeIndexRewriter(std::unique_ptr<LogicalOperator> root_op,
SymbolTable *symbol_table, AstStorage *ast_storage,
TDbAccessor *db) {
impl::EdgeTypeIndexRewriter<TDbAccessor> rewriter(symbol_table, ast_storage, db);
root_op->Accept(rewriter);
return root_op;
}
} // namespace memgraph::query::plan

12
src/query/plan/rewrite/index_lookup.hpp
View File

@ -595,6 +595,18 @@ class IndexLookupRewriter final : public HierarchicalLogicalOperatorVisitor {
return true;
}
bool PreVisit(RollUpApply &op) override {
prev_ops_.push_back(&op);
op.input()->Accept(*this);
RewriteBranch(&op.list_collection_branch_);
return false;
}
bool PostVisit(RollUpApply &) override {
prev_ops_.pop_back();
return true;
}
std::shared_ptr<LogicalOperator> new_root_;
private:

14
src/query/plan/rewrite/join.hpp
View File

@ -1,4 +1,4 @@
// Copyright 2023 Memgraph Ltd.
// Copyright 2024 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
@ -455,6 +455,18 @@ class JoinRewriter final : public HierarchicalLogicalOperatorVisitor {
return true;
}
bool PreVisit(RollUpApply &op) override {
prev_ops_.push_back(&op);
op.input()->Accept(*this);
RewriteBranch(&op.list_collection_branch_);
return false;
}
bool PostVisit(RollUpApply &) override {
prev_ops_.pop_back();
return true;
}
std::shared_ptr<LogicalOperator> new_root_;
private:

62
src/query/plan/rule_based_planner.cpp
View File

@ -14,9 +14,12 @@
#include <algorithm>
#include <functional>
#include <limits>
#include <memory>
#include <stack>
#include <unordered_set>
#include "query/frontend/ast/ast.hpp"
#include "query/plan/operator.hpp"
#include "query/plan/preprocess.hpp"
#include "utils/algorithm.hpp"
#include "utils/exceptions.hpp"
@ -40,7 +43,8 @@ namespace {
class ReturnBodyContext : public HierarchicalTreeVisitor {
public:
ReturnBodyContext(const ReturnBody &body, SymbolTable &symbol_table, const std::unordered_set<Symbol> &bound_symbols,
AstStorage &storage, Where *where = nullptr)
AstStorage &storage, std::unordered_map<std::string, std::shared_ptr<LogicalOperator>> pc_ops,
Where *where = nullptr)
: body_(body), symbol_table_(symbol_table), bound_symbols_(bound_symbols), storage_(storage), where_(where) {
// Collect symbols from named expressions.
output_symbols_.reserve(body_.named_expressions.size());
@ -53,6 +57,14 @@ class ReturnBodyContext : public HierarchicalTreeVisitor {
output_symbols_.emplace_back(symbol_table_.at(*named_expr));
named_expr->Accept(*this);
named_expressions_.emplace_back(named_expr);
if (pattern_comprehension_) {
if (auto it = pc_ops.find(named_expr->name_); it != pc_ops.end()) {
pattern_comprehension_op_ = std::move(it->second);
pc_ops.erase(it);
} else {
throw utils::NotYetImplemented("Operation on top of pattern comprehension");
}
}
}
// Collect symbols used in group by expressions.
if (!aggregations_.empty()) {
@ -386,8 +398,20 @@ class ReturnBodyContext : public HierarchicalTreeVisitor {
return true;
}
bool PostVisit(PatternComprehension & /*unused*/) override {
throw utils::NotYetImplemented("Planner can not handle pattern comprehension.");
bool PreVisit(PatternComprehension & /*unused*/) override {
pattern_compression_aggregations_start_index_ = has_aggregation_.size();
return true;
}
bool PostVisit(PatternComprehension &pattern_comprehension) override {
bool has_aggr = false;
for (auto i = has_aggregation_.size(); i > pattern_compression_aggregations_start_index_; --i) {
has_aggr |= has_aggregation_.back();
has_aggregation_.pop_back();
}
has_aggregation_.emplace_back(has_aggr);
pattern_comprehension_ = &pattern_comprehension;
return true;
}
// Creates NamedExpression with an Identifier for each user declared symbol.
@ -444,6 +468,10 @@ class ReturnBodyContext : public HierarchicalTreeVisitor {
// named_expressions.
const auto &output_symbols() const { return output_symbols_; }
const auto *pattern_comprehension() const { return pattern_comprehension_; }
std::shared_ptr<LogicalOperator> pattern_comprehension_op() const { return pattern_comprehension_op_; }
private:
const ReturnBody &body_;
SymbolTable &symbol_table_;
@ -465,10 +493,13 @@ class ReturnBodyContext : public HierarchicalTreeVisitor {
// group by it.
std::list<bool> has_aggregation_;
std::vector<NamedExpression *> named_expressions_;
PatternComprehension *pattern_comprehension_ = nullptr;
std::shared_ptr<LogicalOperator> pattern_comprehension_op_;
size_t pattern_compression_aggregations_start_index_ = 0;
};
std::unique_ptr<LogicalOperator> GenReturnBody(std::unique_ptr<LogicalOperator> input_op, bool advance_command,
const ReturnBodyContext &body, bool accumulate = false) {
const ReturnBodyContext &body, bool accumulate) {
std::vector<Symbol> used_symbols(body.used_symbols().begin(), body.used_symbols().end());
auto last_op = std::move(input_op);
if (accumulate) {
@ -482,6 +513,11 @@ std::unique_ptr<LogicalOperator> GenReturnBody(std::unique_ptr<LogicalOperator>
std::vector<Symbol> remember(body.group_by_used_symbols().begin(), body.group_by_used_symbols().end());
last_op = std::make_unique<Aggregate>(std::move(last_op), body.aggregations(), body.group_by(), remember);
}
if (body.pattern_comprehension()) {
last_op = std::make_unique<RollUpApply>(std::move(last_op), body.pattern_comprehension_op());
}
last_op = std::make_unique<Produce>(std::move(last_op), body.named_expressions());
// Distinct in ReturnBody only makes Produce values unique, so plan after it.
if (body.distinct()) {
@ -506,6 +542,7 @@ std::unique_ptr<LogicalOperator> GenReturnBody(std::unique_ptr<LogicalOperator>
last_op = std::make_unique<Filter>(std::move(last_op), std::vector<std::shared_ptr<LogicalOperator>>{},
body.where()->expression_);
}
return last_op;
}
@ -543,8 +580,9 @@ Expression *ExtractFilters(const std::unordered_set<Symbol> &bound_symbols, Filt
return filter_expr;
}
std::unordered_set<Symbol> GetSubqueryBoundSymbols(const std::vector<SingleQueryPart> &single_query_parts,
SymbolTable &symbol_table, AstStorage &storage) {
std::unordered_set<Symbol> GetSubqueryBoundSymbols(
const std::vector<SingleQueryPart> &single_query_parts, SymbolTable &symbol_table, AstStorage &storage,
std::unordered_map<std::string, std::shared_ptr<LogicalOperator>> pc_ops) {
const auto &query = single_query_parts[0];
if (!query.matching.expansions.empty() || query.remaining_clauses.empty()) {
@ -552,7 +590,7 @@ std::unordered_set<Symbol> GetSubqueryBoundSymbols(const std::vector<SingleQuery
}
if (std::unordered_set<Symbol> bound_symbols; auto *with = utils::Downcast<query::With>(query.remaining_clauses[0])) {
auto input_op = impl::GenWith(*with, nullptr, symbol_table, false, bound_symbols, storage);
auto input_op = impl::GenWith(*with, nullptr, symbol_table, false, bound_symbols, storage, pc_ops);
return bound_symbols;
}
@ -583,7 +621,8 @@ std::unique_ptr<LogicalOperator> GenNamedPaths(std::unique_ptr<LogicalOperator>
std::unique_ptr<LogicalOperator> GenReturn(Return &ret, std::unique_ptr<LogicalOperator> input_op,
SymbolTable &symbol_table, bool is_write,
const std::unordered_set<Symbol> &bound_symbols, AstStorage &storage) {
const std::unordered_set<Symbol> &bound_symbols, AstStorage &storage,
std::unordered_map<std::string, std::shared_ptr<LogicalOperator>> pc_ops) {
// Similar to WITH clause, but we want to accumulate when the query writes to
// the database. This way we handle the case when we want to return
// expressions with the latest updated results. For example, `MATCH (n) -- ()
@ -592,13 +631,14 @@ std::unique_ptr<LogicalOperator> GenReturn(Return &ret, std::unique_ptr<LogicalO
// final result of 'k' increments.
bool accumulate = is_write;
bool advance_command = false;
ReturnBodyContext body(ret.body_, symbol_table, bound_symbols, storage);
ReturnBodyContext body(ret.body_, symbol_table, bound_symbols, storage, pc_ops);
return GenReturnBody(std::move(input_op), advance_command, body, accumulate);
}
std::unique_ptr<LogicalOperator> GenWith(With &with, std::unique_ptr<LogicalOperator> input_op,
SymbolTable &symbol_table, bool is_write,
std::unordered_set<Symbol> &bound_symbols, AstStorage &storage) {
std::unordered_set<Symbol> &bound_symbols, AstStorage &storage,
std::unordered_map<std::string, std::shared_ptr<LogicalOperator>> pc_ops) {
// WITH clause is Accumulate/Aggregate (advance_command) + Produce and
// optional Filter. In case of update and aggregation, we want to accumulate
// first, so that when aggregating, we get the latest results. Similar to
@ -606,7 +646,7 @@ std::unique_ptr<LogicalOperator> GenWith(With &with, std::unique_ptr<LogicalOper
bool accumulate = is_write;
// No need to advance the command if we only performed reads.
bool advance_command = is_write;
ReturnBodyContext body(with.body_, symbol_table, bound_symbols, storage, with.where_);
ReturnBodyContext body(with.body_, symbol_table, bound_symbols, storage, pc_ops, with.where_);
auto last_op = GenReturnBody(std::move(input_op), advance_command, body, accumulate);
// Reset bound symbols, so that only those in WITH are exposed.
bound_symbols.clear();

39
src/query/plan/rule_based_planner.hpp
View File

@ -21,6 +21,7 @@
#include "query/frontend/ast/ast_visitor.hpp"
#include "query/plan/operator.hpp"
#include "query/plan/preprocess.hpp"
#include "utils/exceptions.hpp"
#include "utils/logging.hpp"
#include "utils/typeinfo.hpp"
@ -87,8 +88,9 @@ bool HasBoundFilterSymbols(const std::unordered_set<Symbol> &bound_symbols, cons
// Returns the set of symbols for the subquery that are actually referenced from the outer scope and
// used in the subquery.
std::unordered_set<Symbol> GetSubqueryBoundSymbols(const std::vector<SingleQueryPart> &single_query_parts,
SymbolTable &symbol_table, AstStorage &storage);
std::unordered_set<Symbol> GetSubqueryBoundSymbols(
const std::vector<SingleQueryPart> &single_query_parts, SymbolTable &symbol_table, AstStorage &storage,
std::unordered_map<std::string, std::shared_ptr<LogicalOperator>> pc_ops);
Symbol GetSymbol(NodeAtom *atom, const SymbolTable &symbol_table);
Symbol GetSymbol(EdgeAtom *atom, const SymbolTable &symbol_table);
@ -142,11 +144,13 @@ std::unique_ptr<LogicalOperator> GenNamedPaths(std::unique_ptr<LogicalOperator>
std::unique_ptr<LogicalOperator> GenReturn(Return &ret, std::unique_ptr<LogicalOperator> input_op,
SymbolTable &symbol_table, bool is_write,
const std::unordered_set<Symbol> &bound_symbols, AstStorage &storage);
const std::unordered_set<Symbol> &bound_symbols, AstStorage &storage,
std::unordered_map<std::string, std::shared_ptr<LogicalOperator>> pc_ops);
std::unique_ptr<LogicalOperator> GenWith(With &with, std::unique_ptr<LogicalOperator> input_op,
SymbolTable &symbol_table, bool is_write,
std::unordered_set<Symbol> &bound_symbols, AstStorage &storage);
std::unordered_set<Symbol> &bound_symbols, AstStorage &storage,
std::unordered_map<std::string, std::shared_ptr<LogicalOperator>> pc_ops);
std::unique_ptr<LogicalOperator> GenUnion(const CypherUnion &cypher_union, std::shared_ptr<LogicalOperator> left_op,
std::shared_ptr<LogicalOperator> right_op, SymbolTable &symbol_table);
@ -190,11 +194,24 @@ class RuleBasedPlanner {
uint64_t merge_id = 0;
uint64_t subquery_id = 0;
std::unordered_map<std::string, std::shared_ptr<LogicalOperator>> pattern_comprehension_ops;
if (single_query_part.pattern_comprehension_matchings.size() > 1) {
throw utils::NotYetImplemented("Multiple pattern comprehensions.");
}
for (const auto &matching : single_query_part.pattern_comprehension_matchings) {
std::unique_ptr<LogicalOperator> new_input;
MatchContext match_ctx{matching.second, *context.symbol_table, context.bound_symbols};
new_input = PlanMatching(match_ctx, std::move(new_input));
new_input = std::make_unique<Produce>(std::move(new_input), std::vector{matching.second.result_expr});
pattern_comprehension_ops.emplace(matching.first, std::move(new_input));
}
for (const auto &clause : single_query_part.remaining_clauses) {
MG_ASSERT(!utils::IsSubtype(*clause, Match::kType), "Unexpected Match in remaining clauses");
if (auto *ret = utils::Downcast<Return>(clause)) {
input_op = impl::GenReturn(*ret, std::move(input_op), *context.symbol_table, context.is_write_query,
context.bound_symbols, *context.ast_storage);
context.bound_symbols, *context.ast_storage, pattern_comprehension_ops);
} else if (auto *merge = utils::Downcast<query::Merge>(clause)) {
input_op = GenMerge(*merge, std::move(input_op), single_query_part.merge_matching[merge_id++]);
// Treat MERGE clause as write, because we do not know if it will
@ -202,7 +219,7 @@ class RuleBasedPlanner {
context.is_write_query = true;
} else if (auto *with = utils::Downcast<query::With>(clause)) {
input_op = impl::GenWith(*with, std::move(input_op), *context.symbol_table, context.is_write_query,
context.bound_symbols, *context.ast_storage);
context.bound_symbols, *context.ast_storage, pattern_comprehension_ops);
// WITH clause advances the command, so reset the flag.
context.is_write_query = false;
} else if (auto op = HandleWriteClause(clause, input_op, *context.symbol_table, context.bound_symbols)) {
@ -241,7 +258,7 @@ class RuleBasedPlanner {
single_query_part, merge_id);
} else if (auto *call_sub = utils::Downcast<query::CallSubquery>(clause)) {
input_op = HandleSubquery(std::move(input_op), single_query_part.subqueries[subquery_id++],
*context.symbol_table, *context_->ast_storage);
*context.symbol_table, *context_->ast_storage, pattern_comprehension_ops);
} else {
throw utils::NotYetImplemented("clause '{}' conversion to operator(s)", clause->GetTypeInfo().name);
}
@ -860,15 +877,15 @@ class RuleBasedPlanner {
symbol);
}
std::unique_ptr<LogicalOperator> HandleSubquery(std::unique_ptr<LogicalOperator> last_op,
std::shared_ptr<QueryParts> subquery, SymbolTable &symbol_table,
AstStorage &storage) {
std::unique_ptr<LogicalOperator> HandleSubquery(
std::unique_ptr<LogicalOperator> last_op, std::shared_ptr<QueryParts> subquery, SymbolTable &symbol_table,
AstStorage &storage, std::unordered_map<std::string, std::shared_ptr<LogicalOperator>> pc_ops) {
std::unordered_set<Symbol> outer_scope_bound_symbols;
outer_scope_bound_symbols.insert(std::make_move_iterator(context_->bound_symbols.begin()),
std::make_move_iterator(context_->bound_symbols.end()));
context_->bound_symbols =
impl::GetSubqueryBoundSymbols(subquery->query_parts[0].single_query_parts, symbol_table, storage);
impl::GetSubqueryBoundSymbols(subquery->query_parts[0].single_query_parts, symbol_table, storage, pc_ops);
auto subquery_op = Plan(*subquery);

2
src/query/plan/vertex_count_cache.hpp
View File

@ -78,6 +78,8 @@ class VertexCountCache {
return db_->LabelPropertyIndexExists(label, property);
}
bool EdgeTypeIndexExists(storage::EdgeTypeId edge_type) { return db_->EdgeTypeIndexExists(edge_type); }
std::optional<storage::LabelIndexStats> GetIndexStats(const storage::LabelId &label) const {
return db_->GetIndexStats(label);
}

2
src/query/procedure/mg_procedure_impl.cpp
View File

@ -3798,7 +3798,7 @@ void PrintFuncSignature(const mgp_func &func, std::ostream &stream) {
bool IsValidIdentifierName(const char *name) {
if (!name) return false;
std::regex regex("[_[:alpha:]][_[:alnum:]]*");
static std::regex regex("[_[:alpha:]][_[:alnum:]]*");
return std::regex_match(name, regex);
}

2
src/query/procedure/module.hpp
View File

@ -1,4 +1,4 @@
// Copyright 2023 Memgraph Ltd.
// Copyright 2024 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

2
src/query/procedure/py_module.hpp
View File

@ -1,4 +1,4 @@
// Copyright 2023 Memgraph Ltd.
// Copyright 2024 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

46
src/query/replication_query_handler.hpp
View File

@ -11,6 +11,8 @@
#pragma once
#include "replication/replication_client.hpp"
#include "replication_coordination_glue/mode.hpp"
#include "replication_coordination_glue/role.hpp"
#include "utils/result.hpp"
#include "utils/uuid.hpp"
@ -31,6 +33,7 @@ enum class RegisterReplicaError : uint8_t {
COULD_NOT_BE_PERSISTED,
ERROR_ACCEPTING_MAIN
};
enum class UnregisterReplicaResult : uint8_t {
NOT_MAIN,
COULD_NOT_BE_PERSISTED,
@ -38,6 +41,47 @@ enum class UnregisterReplicaResult : uint8_t {
SUCCESS,
};
enum class ShowReplicaError : uint8_t {
NOT_MAIN,
};
struct ReplicaSystemInfoState {
uint64_t ts_;
uint64_t behind_;
replication::ReplicationClient::State state_;
};
struct ReplicaInfoState {
ReplicaInfoState(uint64_t ts, uint64_t behind, storage::replication::ReplicaState state)
: ts_(ts), behind_(behind), state_(state) {}
uint64_t ts_;
uint64_t behind_;
storage::replication::ReplicaState state_;
};
struct ReplicasInfo {
ReplicasInfo(std::string name, std::string socket_address, replication_coordination_glue::ReplicationMode sync_mode,
ReplicaSystemInfoState system_info, std::map<std::string, ReplicaInfoState> data_info)
: name_(std::move(name)),
socket_address_(std::move(socket_address)),
sync_mode_(sync_mode),
system_info_(std::move(system_info)),
data_info_(std::move(data_info)) {}
std::string name_;
std::string socket_address_;
memgraph::replication_coordination_glue::ReplicationMode sync_mode_;
ReplicaSystemInfoState system_info_;
std::map<std::string, ReplicaInfoState> data_info_;
};
struct ReplicasInfos {
explicit ReplicasInfos(std::vector<ReplicasInfo> entries) : entries_(std::move(entries)) {}
std::vector<ReplicasInfo> entries_;
};
/// A handler type that keep in sync current ReplicationState and the MAIN/REPLICA-ness of Storage
struct ReplicationQueryHandler {
virtual ~ReplicationQueryHandler() = default;
@ -66,6 +110,8 @@ struct ReplicationQueryHandler {
virtual auto GetRole() const -> memgraph::replication_coordination_glue::ReplicationRole = 0;
virtual bool IsMain() const = 0;
virtual bool IsReplica() const = 0;
virtual auto ShowReplicas() const -> utils::BasicResult<ShowReplicaError, ReplicasInfos> = 0;
};
} // namespace memgraph::query

5
src/replication/include/replication/replication_client.hpp
View File

@ -14,7 +14,9 @@
#include "replication/config.hpp"
#include "replication_coordination_glue/messages.hpp"
#include "rpc/client.hpp"
#include "utils/rw_lock.hpp"
#include "utils/scheduler.hpp"
#include "utils/spin_lock.hpp"
#include "utils/synchronized.hpp"
#include "utils/thread_pool.hpp"
@ -114,8 +116,9 @@ struct ReplicationClient {
enum class State {
BEHIND,
READY,
RECOVERY,
};
utils::Synchronized<State> state_{State::BEHIND};
utils::Synchronized<State, utils::WritePrioritizedRWLock> state_{State::BEHIND};
replication_coordination_glue::ReplicationMode mode_{replication_coordination_glue::ReplicationMode::SYNC};
// This thread pool is used for background tasks so we don't

1
src/replication_coordination_glue/CMakeLists.txt
View File

@ -7,6 +7,7 @@ target_sources(mg-repl_coord_glue
mode.hpp
role.hpp
handler.hpp
common.hpp
PRIVATE
messages.cpp

32
src/replication_coordination_glue/common.hpp Normal file
View File

@ -0,0 +1,32 @@
// Copyright 2024 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 "rpc/client.hpp"
#include "utils/uuid.hpp"
#include <deque>
#include "messages.hpp"
#include "rpc/messages.hpp"
#include "utils/uuid.hpp"
namespace memgraph::replication_coordination_glue {
struct DatabaseHistory {
memgraph::utils::UUID db_uuid;
std::vector<std::pair<std::string, uint64_t>> history;
std::string name;
};
using DatabaseHistories = std::vector<DatabaseHistory>;
} // namespace memgraph::replication_coordination_glue

12
src/replication_coordination_glue/mode.hpp
View File

@ -12,7 +12,19 @@
#pragma once
#include <cstdint>
#include <map>
#include <stdexcept>
#include <string>
#include "json/json.hpp"
namespace memgraph::replication_coordination_glue {
enum class ReplicationMode : std::uint8_t { SYNC, ASYNC };
NLOHMANN_JSON_SERIALIZE_ENUM(ReplicationMode, {
{ReplicationMode::SYNC, "sync"},
{ReplicationMode::ASYNC, "async"},
})
} // namespace memgraph::replication_coordination_glue

6
src/replication_coordination_glue/role.hpp
View File

@ -12,8 +12,14 @@
#pragma once
#include <cstdint>
#include "json/json.hpp"
namespace memgraph::replication_coordination_glue {
// TODO: figure out a way of ensuring that usage of this type is never uninitialed/defaulted incorrectly to MAIN
enum class ReplicationRole : uint8_t { MAIN, REPLICA };
NLOHMANN_JSON_SERIALIZE_ENUM(ReplicationRole, {{ReplicationRole::MAIN, "main"}, {ReplicationRole::REPLICA, "replica"}})
} // namespace memgraph::replication_coordination_glue

15
src/replication_handler/include/replication_handler/replication_handler.hpp
View File

@ -14,6 +14,7 @@
#include "dbms/dbms_handler.hpp"
#include "flags/experimental.hpp"
#include "replication/include/replication/state.hpp"
#include "replication_coordination_glue/common.hpp"
#include "replication_handler/system_replication.hpp"
#include "replication_handler/system_rpc.hpp"
#include "utils/result.hpp"
@ -39,10 +40,12 @@ void SystemRestore(replication::ReplicationClient &client, system::System &syste
const utils::UUID &main_uuid, auth::SynchedAuth &auth) {
// Check if system is up to date
if (client.state_.WithLock(
[](auto &state) { return state == memgraph::replication::ReplicationClient::State::READY; }))
[](auto &state) { return state != memgraph::replication::ReplicationClient::State::BEHIND; }))
return;
// Try to recover...
client.state_.WithLock(
[](auto &state) { return state != memgraph::replication::ReplicationClient::State::RECOVERY; });
{
using enum memgraph::flags::Experiments;
bool full_system_replication =
@ -139,11 +142,16 @@ struct ReplicationHandler : public memgraph::query::ReplicationQueryHandler {
bool IsMain() const override;
bool IsReplica() const override;
auto ShowReplicas() const
-> utils::BasicResult<memgraph::query::ShowReplicaError, memgraph::query::ReplicasInfos> override;
auto GetReplState() const -> const memgraph::replication::ReplicationState &;
auto GetReplState() -> memgraph::replication::ReplicationState &;
auto GetReplicaUUID() -> std::optional<utils::UUID>;
auto GetDatabasesHistories() -> replication_coordination_glue::DatabaseHistories;
private:
template <bool SendSwapUUID>
auto RegisterReplica_(const memgraph::replication::ReplicationClientConfig &config)
@ -202,8 +210,13 @@ struct ReplicationHandler : public memgraph::query::ReplicationQueryHandler {
auto client = std::make_unique<storage::ReplicationStorageClient>(*instance_client_ptr, main_uuid);
client->Start(storage, std::move(db_acc));
bool const success = std::invoke([state = client->State()]() {
// We force sync replicas in other situation
if (state == storage::replication::ReplicaState::DIVERGED_FROM_MAIN) {
#ifdef MG_ENTERPRISE
return FLAGS_coordinator_server_port != 0;
#else
return false;
#endif
}
return true;
});

149
src/replication_handler/replication_handler.cpp
View File

@ -10,26 +10,29 @@
// licenses/APL.txt.
#include "replication_handler/replication_handler.hpp"
#include "dbms/constants.hpp"
#include "dbms/dbms_handler.hpp"
#include "replication/replication_client.hpp"
#include "replication_handler/system_replication.hpp"
#include "utils/functional.hpp"
namespace memgraph::replication {
namespace {
#ifdef MG_ENTERPRISE
void RecoverReplication(memgraph::replication::ReplicationState &repl_state, memgraph::system::System &system,
memgraph::dbms::DbmsHandler &dbms_handler, memgraph::auth::SynchedAuth &auth) {
void RecoverReplication(replication::ReplicationState &repl_state, system::System &system,
dbms::DbmsHandler &dbms_handler, auth::SynchedAuth &auth) {
/*
* REPLICATION RECOVERY AND STARTUP
*/
// Startup replication state (if recovered at startup)
auto replica = [&dbms_handler, &auth, &system](memgraph::replication::RoleReplicaData &data) {
return memgraph::replication::StartRpcServer(dbms_handler, data, auth, system);
auto replica = [&dbms_handler, &auth, &system](replication::RoleReplicaData &data) {
return replication::StartRpcServer(dbms_handler, data, auth, system);
};
// Replication recovery and frequent check start
auto main = [&system, &dbms_handler, &auth](memgraph::replication::RoleMainData &mainData) {
auto main = [&system, &dbms_handler, &auth](replication::RoleMainData &mainData) {
for (auto &client : mainData.registered_replicas_) {
if (client.try_set_uuid &&
replication_coordination_glue::SendSwapMainUUIDRpc(client.rpc_client_, mainData.uuid_)) {
@ -38,7 +41,7 @@ void RecoverReplication(memgraph::replication::ReplicationState &repl_state, mem
SystemRestore(client, system, dbms_handler, mainData.uuid_, auth);
}
// DBMS here
dbms_handler.ForEach([&mainData](memgraph::dbms::DatabaseAccess db_acc) {
dbms_handler.ForEach([&mainData](dbms::DatabaseAccess db_acc) {
dbms::DbmsHandler::RecoverStorageReplication(std::move(db_acc), mainData);
});
@ -48,7 +51,7 @@ void RecoverReplication(memgraph::replication::ReplicationState &repl_state, mem
// Warning
if (dbms_handler.default_config().durability.snapshot_wal_mode ==
memgraph::storage::Config::Durability::SnapshotWalMode::DISABLED) {
storage::Config::Durability::SnapshotWalMode::DISABLED) {
spdlog::warn(
"The instance has the MAIN replication role, but durability logs and snapshots are disabled. Please "
"consider "
@ -59,19 +62,18 @@ void RecoverReplication(memgraph::replication::ReplicationState &repl_state, mem
return true;
};
auto result = std::visit(memgraph::utils::Overloaded{replica, main}, repl_state.ReplicationData());
auto result = std::visit(utils::Overloaded{replica, main}, repl_state.ReplicationData());
MG_ASSERT(result, "Replica recovery failure!");
}
#else
void RecoverReplication(memgraph::replication::ReplicationState &repl_state,
memgraph::dbms::DbmsHandler &dbms_handler) {
void RecoverReplication(replication::ReplicationState &repl_state, dbms::DbmsHandler &dbms_handler) {
// Startup replication state (if recovered at startup)
auto replica = [&dbms_handler](memgraph::replication::RoleReplicaData &data) {
return memgraph::replication::StartRpcServer(dbms_handler, data);
auto replica = [&dbms_handler](replication::RoleReplicaData &data) {
return replication::StartRpcServer(dbms_handler, data);
};
// Replication recovery and frequent check start
auto main = [&dbms_handler](memgraph::replication::RoleMainData &mainData) {
auto main = [&dbms_handler](replication::RoleMainData &mainData) {
dbms::DbmsHandler::RecoverStorageReplication(dbms_handler.Get(), mainData);
for (auto &client : mainData.registered_replicas_) {
@ -79,12 +81,12 @@ void RecoverReplication(memgraph::replication::ReplicationState &repl_state,
replication_coordination_glue::SendSwapMainUUIDRpc(client.rpc_client_, mainData.uuid_)) {
client.try_set_uuid = false;
}
memgraph::replication::StartReplicaClient(client, dbms_handler, mainData.uuid_);
replication::StartReplicaClient(client, dbms_handler, mainData.uuid_);
}
// Warning
if (dbms_handler.default_config().durability.snapshot_wal_mode ==
memgraph::storage::Config::Durability::SnapshotWalMode::DISABLED) {
storage::Config::Durability::SnapshotWalMode::DISABLED) {
spdlog::warn(
"The instance has the MAIN replication role, but durability logs and snapshots are disabled. Please "
"consider "
@ -95,7 +97,7 @@ void RecoverReplication(memgraph::replication::ReplicationState &repl_state,
return true;
};
auto result = std::visit(memgraph::utils::Overloaded{replica, main}, repl_state.ReplicationData());
auto result = std::visit(utils::Overloaded{replica, main}, repl_state.ReplicationData());
MG_ASSERT(result, "Replica recovery failure!");
}
#endif
@ -133,20 +135,19 @@ void StartReplicaClient(replication::ReplicationClient &client, dbms::DbmsHandle
spdlog::trace("Replication client started at: {}:{}", endpoint.address, endpoint.port);
client.StartFrequentCheck([&, license = license::global_license_checker.IsEnterpriseValidFast(), main_uuid](
bool reconnect, replication::ReplicationClient &client) mutable {
if (client.try_set_uuid &&
memgraph::replication_coordination_glue::SendSwapMainUUIDRpc(client.rpc_client_, main_uuid)) {
if (client.try_set_uuid && replication_coordination_glue::SendSwapMainUUIDRpc(client.rpc_client_, main_uuid)) {
client.try_set_uuid = false;
}
// Working connection
// Check if system needs restoration
if (reconnect) {
client.state_.WithLock([](auto &state) { state = memgraph::replication::ReplicationClient::State::BEHIND; });
client.state_.WithLock([](auto &state) { state = replication::ReplicationClient::State::BEHIND; });
}
// Check if license has changed
const auto new_license = license::global_license_checker.IsEnterpriseValidFast();
if (new_license != license) {
license = new_license;
client.state_.WithLock([](auto &state) { state = memgraph::replication::ReplicationClient::State::BEHIND; });
client.state_.WithLock([](auto &state) { state = replication::ReplicationClient::State::BEHIND; });
}
#ifdef MG_ENTERPRISE
SystemRestore<true>(client, system, dbms_handler, main_uuid, auth);
@ -154,10 +155,10 @@ void StartReplicaClient(replication::ReplicationClient &client, dbms::DbmsHandle
// Check if any database has been left behind
dbms_handler.ForEach([&name = client.name_, reconnect](dbms::DatabaseAccess db_acc) {
// Specific database <-> replica client
db_acc->storage()->repl_storage_state_.WithClient(name, [&](storage::ReplicationStorageClient *client) {
if (reconnect || client->State() == storage::replication::ReplicaState::MAYBE_BEHIND) {
db_acc->storage()->repl_storage_state_.WithClient(name, [&](storage::ReplicationStorageClient &client) {
if (reconnect || client.State() == storage::replication::ReplicaState::MAYBE_BEHIND) {
// Database <-> replica might be behind, check and recover
client->TryCheckReplicaStateAsync(db_acc->storage(), db_acc);
client.TryCheckReplicaStateAsync(db_acc->storage(), db_acc);
}
});
});
@ -165,9 +166,8 @@ void StartReplicaClient(replication::ReplicationClient &client, dbms::DbmsHandle
}
#ifdef MG_ENTERPRISE
ReplicationHandler::ReplicationHandler(memgraph::replication::ReplicationState &repl_state,
memgraph::dbms::DbmsHandler &dbms_handler, memgraph::system::System &system,
memgraph::auth::SynchedAuth &auth)
ReplicationHandler::ReplicationHandler(replication::ReplicationState &repl_state, dbms::DbmsHandler &dbms_handler,
system::System &system, auth::SynchedAuth &auth)
: repl_state_{repl_state}, dbms_handler_{dbms_handler}, system_{system}, auth_{auth} {
RecoverReplication(repl_state_, system_, dbms_handler_, auth_);
}
@ -179,20 +179,20 @@ ReplicationHandler::ReplicationHandler(replication::ReplicationState &repl_state
#endif
bool ReplicationHandler::SetReplicationRoleMain() {
auto const main_handler = [](memgraph::replication::RoleMainData &) {
auto const main_handler = [](replication::RoleMainData &) {
// If we are already MAIN, we don't want to change anything
return false;
};
auto const replica_handler = [this](memgraph::replication::RoleReplicaData const &) {
auto const replica_handler = [this](replication::RoleReplicaData const &) {
return DoReplicaToMainPromotion(utils::UUID{});
};
// TODO: under lock
return std::visit(memgraph::utils::Overloaded{main_handler, replica_handler}, repl_state_.ReplicationData());
return std::visit(utils::Overloaded{main_handler, replica_handler}, repl_state_.ReplicationData());
}
bool ReplicationHandler::SetReplicationRoleReplica(const memgraph::replication::ReplicationServerConfig &config,
bool ReplicationHandler::SetReplicationRoleReplica(const replication::ReplicationServerConfig &config,
const std::optional<utils::UUID> &main_uuid) {
return SetReplicationRoleReplica_<true>(config, main_uuid);
}
@ -238,18 +238,16 @@ auto ReplicationHandler::RegisterReplica(const memgraph::replication::Replicatio
return RegisterReplica_<false>(config);
}
auto ReplicationHandler::UnregisterReplica(std::string_view name) -> memgraph::query::UnregisterReplicaResult {
auto const replica_handler =
[](memgraph::replication::RoleReplicaData const &) -> memgraph::query::UnregisterReplicaResult {
return memgraph::query::UnregisterReplicaResult::NOT_MAIN;
auto ReplicationHandler::UnregisterReplica(std::string_view name) -> query::UnregisterReplicaResult {
auto const replica_handler = [](replication::RoleReplicaData const &) -> query::UnregisterReplicaResult {
return query::UnregisterReplicaResult::NOT_MAIN;
};
auto const main_handler =
[this, name](memgraph::replication::RoleMainData &mainData) -> memgraph::query::UnregisterReplicaResult {
auto const main_handler = [this, name](replication::RoleMainData &mainData) -> query::UnregisterReplicaResult {
if (!repl_state_.TryPersistUnregisterReplica(name)) {
return memgraph::query::UnregisterReplicaResult::COULD_NOT_BE_PERSISTED;
return query::UnregisterReplicaResult::COULD_NOT_BE_PERSISTED;
}
// Remove database specific clients
dbms_handler_.ForEach([name](memgraph::dbms::DatabaseAccess db_acc) {
dbms_handler_.ForEach([name](dbms::DatabaseAccess db_acc) {
db_acc->storage()->repl_storage_state_.replication_clients_.WithLock([&name](auto &clients) {
std::erase_if(clients, [name](const auto &client) { return client->Name() == name; });
});
@ -257,28 +255,91 @@ auto ReplicationHandler::UnregisterReplica(std::string_view name) -> memgraph::q
// Remove instance level clients
auto const n_unregistered =
std::erase_if(mainData.registered_replicas_, [name](auto const &client) { return client.name_ == name; });
return n_unregistered != 0 ? memgraph::query::UnregisterReplicaResult::SUCCESS
: memgraph::query::UnregisterReplicaResult::CAN_NOT_UNREGISTER;
return n_unregistered != 0 ? query::UnregisterReplicaResult::SUCCESS
: query::UnregisterReplicaResult::CAN_NOT_UNREGISTER;
};
return std::visit(memgraph::utils::Overloaded{main_handler, replica_handler}, repl_state_.ReplicationData());
return std::visit(utils::Overloaded{main_handler, replica_handler}, repl_state_.ReplicationData());
}
auto ReplicationHandler::GetRole() const -> memgraph::replication_coordination_glue::ReplicationRole {
auto ReplicationHandler::GetRole() const -> replication_coordination_glue::ReplicationRole {
return repl_state_.GetRole();
}
auto ReplicationHandler::GetDatabasesHistories() -> replication_coordination_glue::DatabaseHistories {
replication_coordination_glue::DatabaseHistories results;
dbms_handler_.ForEach([&results](memgraph::dbms::DatabaseAccess db_acc) {
auto &repl_storage_state = db_acc->storage()->repl_storage_state_;
std::vector<std::pair<std::string, uint64_t>> history = utils::fmap(repl_storage_state.history);
history.emplace_back(std::string(repl_storage_state.epoch_.id()), repl_storage_state.last_commit_timestamp_.load());
replication_coordination_glue::DatabaseHistory repl{
.db_uuid = utils::UUID{db_acc->storage()->uuid()}, .history = history, .name = std::string(db_acc->name())};
results.emplace_back(repl);
});
return results;
}
auto ReplicationHandler::GetReplicaUUID() -> std::optional<utils::UUID> {
MG_ASSERT(repl_state_.IsReplica());
MG_ASSERT(repl_state_.IsReplica(), "Instance is not replica");
return std::get<RoleReplicaData>(repl_state_.ReplicationData()).uuid_;
}
auto ReplicationHandler::GetReplState() const -> const memgraph::replication::ReplicationState & { return repl_state_; }
auto ReplicationHandler::GetReplState() -> memgraph::replication::ReplicationState & { return repl_state_; }
auto ReplicationHandler::GetReplState() -> replication::ReplicationState & { return repl_state_; }
bool ReplicationHandler::IsMain() const { return repl_state_.IsMain(); }
bool ReplicationHandler::IsReplica() const { return repl_state_.IsReplica(); }
auto ReplicationHandler::ShowReplicas() const -> utils::BasicResult<query::ShowReplicaError, query::ReplicasInfos> {
using res_t = utils::BasicResult<query::ShowReplicaError, query::ReplicasInfos>;
auto main = [this](RoleMainData const &main) -> res_t {
auto entries = std::vector<query::ReplicasInfo>{};
entries.reserve(main.registered_replicas_.size());
const bool full_info = license::global_license_checker.IsEnterpriseValidFast();
for (auto const &replica : main.registered_replicas_) {
// STEP 1: data_info
auto data_info = std::map<std::string, query::ReplicaInfoState>{};
this->dbms_handler_.ForEach([&](dbms::DatabaseAccess db_acc) {
auto *storage = db_acc->storage();
// ATM we only support IN_MEMORY_TRANSACTIONAL
if (storage->storage_mode_ != storage::StorageMode::IN_MEMORY_TRANSACTIONAL) return;
if (!full_info && storage->name() == dbms::kDefaultDB) return;
auto ok =
storage->repl_storage_state_.WithClient(replica.name_, [&](storage::ReplicationStorageClient &client) {
auto ts_info = client.GetTimestampInfo(storage);
auto state = client.State();
data_info.emplace(storage->name(),
query::ReplicaInfoState{ts_info.current_timestamp_of_replica,
ts_info.current_number_of_timestamp_behind_main, state});
});
DMG_ASSERT(ok);
});
// STEP 2: system_info
#ifdef MG_ENTERPRISE
// Already locked on system transaction via the interpreter
const auto ts = system_.LastCommittedSystemTimestamp();
// NOTE: no system behind at the moment
query::ReplicaSystemInfoState system_info{ts, 0 /* behind ts not implemented */, *replica.state_.ReadLock()};
#else
query::ReplicaSystemInfoState system_info{};
#endif
// STEP 3: add entry
entries.emplace_back(replica.name_, replica.rpc_client_.Endpoint().SocketAddress(), replica.mode_, system_info,
std::move(data_info));
}
return query::ReplicasInfos{std::move(entries)};
};
auto replica = [](RoleReplicaData const &) -> res_t { return query::ShowReplicaError::NOT_MAIN; };
return std::visit(utils::Overloaded{main, replica}, repl_state_.ReplicationData());
}
} // namespace memgraph::replication

3
src/storage/v2/CMakeLists.txt
View File

@ -21,8 +21,10 @@ add_library(mg-storage-v2 STATIC
storage.cpp
indices/indices.cpp
all_vertices_iterable.cpp
edges_iterable.cpp
vertices_iterable.cpp
inmemory/storage.cpp
inmemory/edge_type_index.cpp
inmemory/label_index.cpp
inmemory/label_property_index.cpp
inmemory/unique_constraints.cpp
@ -30,6 +32,7 @@ add_library(mg-storage-v2 STATIC
disk/edge_import_mode_cache.cpp
disk/storage.cpp
disk/rocksdb_storage.cpp
disk/edge_type_index.cpp
disk/label_index.cpp
disk/label_property_index.cpp
disk/unique_constraints.cpp

26
src/storage/v2/delta.hpp
View File

@ -123,6 +123,26 @@ inline bool operator==(const PreviousPtr::Pointer &a, const PreviousPtr::Pointer
inline bool operator!=(const PreviousPtr::Pointer &a, const PreviousPtr::Pointer &b) { return !(a == b); }
struct opt_str {
opt_str(std::optional<std::string> const &other) : str_{other ? new_cstr(*other) : nullptr} {}
~opt_str() { delete[] str_; }
auto as_opt_str() const -> std::optional<std::string> {
if (!str_) return std::nullopt;
return std::optional<std::string>{std::in_place, str_};
}
private:
static auto new_cstr(std::string const &str) -> char const * {
auto *mem = new char[str.length() + 1];
strcpy(mem, str.c_str());
return mem;
}
char const *str_ = nullptr;
};
struct Delta {
enum class Action : std::uint8_t {
/// Use for Vertex and Edge
@ -160,7 +180,7 @@ struct Delta {
// Because of this object was created in past txs, we create timestamp by ourselves inside instead of having it from
// current tx. This timestamp we got from RocksDB timestamp stored in key.
Delta(DeleteDeserializedObjectTag /*tag*/, uint64_t ts, std::optional<std::string> old_disk_key)
: timestamp(new std::atomic<uint64_t>(ts)), command_id(0), old_disk_key{.value = std::move(old_disk_key)} {}
: timestamp(new std::atomic<uint64_t>(ts)), command_id(0), old_disk_key{.value = old_disk_key} {}
Delta(DeleteObjectTag /*tag*/, std::atomic<uint64_t> *timestamp, uint64_t command_id)
: timestamp(timestamp), command_id(command_id), action(Action::DELETE_OBJECT) {}
@ -222,7 +242,7 @@ struct Delta {
case Action::REMOVE_OUT_EDGE:
break;
case Action::DELETE_DESERIALIZED_OBJECT:
old_disk_key.value.reset();
std::destroy_at(&old_disk_key.value);
delete timestamp;
timestamp = nullptr;
break;
@ -242,7 +262,7 @@ struct Delta {
Action action;
struct {
Action action = Action::DELETE_DESERIALIZED_OBJECT;
std::optional<std::string> value;
opt_str value;
} old_disk_key;
struct {
Action action;

49
src/storage/v2/disk/edge_type_index.cpp Normal file
View File

@ -0,0 +1,49 @@
// Copyright 2024 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 "edge_type_index.hpp"
#include "utils/exceptions.hpp"
namespace memgraph::storage {
bool DiskEdgeTypeIndex::DropIndex(EdgeTypeId /*edge_type*/) {
spdlog::warn("Edge-type index related operations are not yet supported using on-disk storage mode.");
return true;
}
bool DiskEdgeTypeIndex::IndexExists(EdgeTypeId /*edge_type*/) const {
spdlog::warn("Edge-type index related operations are not yet supported using on-disk storage mode.");
return false;
}
std::vector<EdgeTypeId> DiskEdgeTypeIndex::ListIndices() const {
spdlog::warn("Edge-type index related operations are not yet supported using on-disk storage mode.");
return {};
}
uint64_t DiskEdgeTypeIndex::ApproximateEdgeCount(EdgeTypeId /*edge_type*/) const {
spdlog::warn("Edge-type index related operations are not yet supported using on-disk storage mode.");
return 0U;
}
void DiskEdgeTypeIndex::UpdateOnEdgeCreation(Vertex * /*from*/, Vertex * /*to*/, EdgeRef /*edge_ref*/,
EdgeTypeId /*edge_type*/, const Transaction & /*tx*/) {
spdlog::warn("Edge-type index related operations are not yet supported using on-disk storage mode.");
}
void DiskEdgeTypeIndex::UpdateOnEdgeModification(Vertex * /*old_from*/, Vertex * /*old_to*/, Vertex * /*new_from*/,
Vertex * /*new_to*/, EdgeRef /*edge_ref*/, EdgeTypeId /*edge_type*/,
const Transaction & /*tx*/) {
spdlog::warn("Edge-type index related operations are not yet supported using on-disk storage mode.");
}
} // namespace memgraph::storage

35
src/storage/v2/disk/edge_type_index.hpp Normal file
View File

@ -0,0 +1,35 @@
// Copyright 2024 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/v2/indices/edge_type_index.hpp"
namespace memgraph::storage {
class DiskEdgeTypeIndex : public storage::EdgeTypeIndex {
public:
bool DropIndex(EdgeTypeId edge_type) override;
bool IndexExists(EdgeTypeId edge_type) const override;
std::vector<EdgeTypeId> ListIndices() const override;
uint64_t ApproximateEdgeCount(EdgeTypeId edge_type) const override;
void UpdateOnEdgeCreation(Vertex *from, Vertex *to, EdgeRef edge_ref, EdgeTypeId edge_type,
const Transaction &tx) override;
void UpdateOnEdgeModification(Vertex *old_from, Vertex *old_to, Vertex *new_from, Vertex *new_to, EdgeRef edge_ref,
EdgeTypeId edge_type, const Transaction &tx) override;
};
} // namespace memgraph::storage

40
src/storage/v2/disk/storage.cpp
View File

@ -41,6 +41,7 @@
#include "storage/v2/edge_accessor.hpp"
#include "storage/v2/edge_import_mode.hpp"
#include "storage/v2/edge_ref.hpp"
#include "storage/v2/edges_iterable.hpp"
#include "storage/v2/id_types.hpp"
#include "storage/v2/modified_edge.hpp"
#include "storage/v2/mvcc.hpp"
@ -807,11 +808,21 @@ void DiskStorage::LoadVerticesFromDiskLabelPropertyIndexForIntervalSearch(
}
}
EdgesIterable DiskStorage::DiskAccessor::Edges(EdgeTypeId /*edge_type*/, View /*view*/) {
throw utils::NotYetImplemented(
"Edge-type index related operations are not yet supported using on-disk storage mode.");
}
uint64_t DiskStorage::DiskAccessor::ApproximateVertexCount() const {
auto *disk_storage = static_cast<DiskStorage *>(storage_);
return disk_storage->vertex_count_.load(std::memory_order_acquire);
}
uint64_t DiskStorage::DiskAccessor::ApproximateEdgeCount(EdgeTypeId /*edge_type*/) const {
spdlog::info("Edge-type index related operations are not yet supported using on-disk storage mode.");
return 0U;
}
uint64_t DiskStorage::GetDiskSpaceUsage() const {
uint64_t main_disk_storage_size = utils::GetDirDiskUsage(config_.disk.main_storage_directory);
uint64_t index_disk_storage_size = utils::GetDirDiskUsage(config_.disk.label_index_directory) +
@ -825,7 +836,7 @@ uint64_t DiskStorage::GetDiskSpaceUsage() const {
durability_disk_storage_size;
}
StorageInfo DiskStorage::GetBaseInfo(bool /* unused */) {
StorageInfo DiskStorage::GetBaseInfo() {
StorageInfo info{};
info.vertex_count = vertex_count_;
info.edge_count = edge_count_.load(std::memory_order_acquire);
@ -838,9 +849,8 @@ StorageInfo DiskStorage::GetBaseInfo(bool /* unused */) {
return info;
}
StorageInfo DiskStorage::GetInfo(bool force_dir,
memgraph::replication_coordination_glue::ReplicationRole replication_role) {
StorageInfo info = GetBaseInfo(force_dir);
StorageInfo DiskStorage::GetInfo(memgraph::replication_coordination_glue::ReplicationRole replication_role) {
StorageInfo info = GetBaseInfo();
{
auto access = Access(replication_role);
const auto &lbl = access->ListAllIndices();
@ -1630,6 +1640,9 @@ utils::BasicResult<StorageManipulationError, void> DiskStorage::DiskAccessor::Co
return StorageManipulationError{PersistenceError{}};
}
} break;
case MetadataDelta::Action::EDGE_INDEX_CREATE: {
throw utils::NotYetImplemented("Edge-type indexing is not yet implemented on on-disk storage mode.");
}
case MetadataDelta::Action::LABEL_INDEX_DROP: {
if (!disk_storage->durable_metadata_.PersistLabelIndexDeletion(md_delta.label)) {
return StorageManipulationError{PersistenceError{}};
@ -1642,6 +1655,9 @@ utils::BasicResult<StorageManipulationError, void> DiskStorage::DiskAccessor::Co
return StorageManipulationError{PersistenceError{}};
}
} break;
case MetadataDelta::Action::EDGE_INDEX_DROP: {
throw utils::NotYetImplemented("Edge-type indexing is not yet implemented on on-disk storage mode.");
}
case MetadataDelta::Action::LABEL_INDEX_STATS_SET: {
throw utils::NotYetImplemented("SetIndexStats(stats) is not implemented for DiskStorage.");
} break;
@ -1918,6 +1934,11 @@ utils::BasicResult<StorageIndexDefinitionError, void> DiskStorage::DiskAccessor:
return {};
}
utils::BasicResult<StorageIndexDefinitionError, void> DiskStorage::DiskAccessor::CreateIndex(EdgeTypeId /*edge_type*/) {
throw utils::NotYetImplemented(
"Edge-type index related operations are not yet supported using on-disk storage mode.");
}
utils::BasicResult<StorageIndexDefinitionError, void> DiskStorage::DiskAccessor::DropIndex(LabelId label) {
MG_ASSERT(unique_guard_.owns_lock(), "Create index requires a unique access to the storage!");
auto *on_disk = static_cast<DiskStorage *>(storage_);
@ -1946,6 +1967,11 @@ utils::BasicResult<StorageIndexDefinitionError, void> DiskStorage::DiskAccessor:
return {};
}
utils::BasicResult<StorageIndexDefinitionError, void> DiskStorage::DiskAccessor::DropIndex(EdgeTypeId /*edge_type*/) {
throw utils::NotYetImplemented(
"Edge-type index related operations are not yet supported using on-disk storage mode.");
}
utils::BasicResult<StorageExistenceConstraintDefinitionError, void>
DiskStorage::DiskAccessor::CreateExistenceConstraint(LabelId label, PropertyId property) {
MG_ASSERT(unique_guard_.owns_lock(), "Create existence constraint requires a unique access to the storage!");
@ -2054,6 +2080,12 @@ std::unique_ptr<Storage::Accessor> DiskStorage::UniqueAccess(
return std::unique_ptr<DiskAccessor>(
new DiskAccessor{Storage::Accessor::unique_access, this, isolation_level, storage_mode_});
}
bool DiskStorage::DiskAccessor::EdgeTypeIndexExists(EdgeTypeId /*edge_type*/) const {
spdlog::info("Edge-type index related operations are not yet supported using on-disk storage mode.");
return false;
}
IndicesInfo DiskStorage::DiskAccessor::ListAllIndices() const {
auto *on_disk = static_cast<DiskStorage *>(storage_);
auto *disk_label_index = static_cast<DiskLabelIndex *>(on_disk->indices_.label_index_.get());

17
src/storage/v2/disk/storage.hpp
View File

@ -72,6 +72,8 @@ class DiskStorage final : public Storage {
const std::optional<utils::Bound<PropertyValue>> &lower_bound,
const std::optional<utils::Bound<PropertyValue>> &upper_bound, View view) override;
EdgesIterable Edges(EdgeTypeId edge_type, View view) override;
uint64_t ApproximateVertexCount() const override;
uint64_t ApproximateVertexCount(LabelId /*label*/) const override { return 10; }
@ -89,6 +91,8 @@ class DiskStorage final : public Storage {
return 10;
}
uint64_t ApproximateEdgeCount(EdgeTypeId edge_type) const override;
std::optional<storage::LabelIndexStats> GetIndexStats(const storage::LabelId & /*label*/) const override {
return {};
}
@ -140,6 +144,8 @@ class DiskStorage final : public Storage {
return disk_storage->indices_.label_property_index_->IndexExists(label, property);
}
bool EdgeTypeIndexExists(EdgeTypeId edge_type) const override;
IndicesInfo ListAllIndices() const override;
ConstraintsInfo ListAllConstraints() const override;
@ -158,10 +164,14 @@ class DiskStorage final : public Storage {
utils::BasicResult<StorageIndexDefinitionError, void> CreateIndex(LabelId label, PropertyId property) override;
utils::BasicResult<StorageIndexDefinitionError, void> CreateIndex(EdgeTypeId edge_type) override;
utils::BasicResult<StorageIndexDefinitionError, void> DropIndex(LabelId label) override;
utils::BasicResult<StorageIndexDefinitionError, void> DropIndex(LabelId label, PropertyId property) override;
utils::BasicResult<StorageIndexDefinitionError, void> DropIndex(EdgeTypeId edge_type) override;
utils::BasicResult<StorageExistenceConstraintDefinitionError, void> CreateExistenceConstraint(
LabelId label, PropertyId property) override;
@ -307,11 +317,10 @@ class DiskStorage final : public Storage {
std::vector<std::pair<std::string, std::string>> SerializeVerticesForLabelPropertyIndex(LabelId label,
PropertyId property);
StorageInfo GetBaseInfo(bool force_directory) override;
StorageInfo GetInfo(bool force_directory,
memgraph::replication_coordination_glue::ReplicationRole replication_role) override;
StorageInfo GetBaseInfo() override;
StorageInfo GetInfo(memgraph::replication_coordination_glue::ReplicationRole replication_role) override;
void FreeMemory(std::unique_lock<utils::ResourceLock> /*lock*/) override {}
void FreeMemory(std::unique_lock<utils::ResourceLock> /*lock*/, bool /*periodic*/) override {}
void PrepareForNewEpoch() override { throw utils::BasicException("Disk storage mode does not support replication."); }

57
src/storage/v2/durability/durability.cpp
View File

@ -1,4 +1,4 @@
// Copyright 2023 Memgraph Ltd.
// Copyright 2024 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
@ -31,6 +31,7 @@
#include "storage/v2/durability/paths.hpp"
#include "storage/v2/durability/snapshot.hpp"
#include "storage/v2/durability/wal.hpp"
#include "storage/v2/inmemory/edge_type_index.hpp"
#include "storage/v2/inmemory/label_index.hpp"
#include "storage/v2/inmemory/label_property_index.hpp"
#include "storage/v2/inmemory/unique_constraints.hpp"
@ -118,6 +119,8 @@ std::optional<std::vector<WalDurabilityInfo>> GetWalFiles(const std::filesystem:
if (!item.is_regular_file()) continue;
try {
auto info = ReadWalInfo(item.path());
spdlog::trace("Getting wal file with following info: uuid: {}, epoch id: {}, from timestamp {}, to_timestamp {} ",
info.uuid, info.epoch_id, info.from_timestamp, info.to_timestamp);
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());
@ -197,9 +200,18 @@ void RecoverIndicesAndStats(const RecoveredIndicesAndConstraints::IndicesMetadat
}
spdlog::info("Label+property indices statistics are recreated.");
spdlog::info("Indices are recreated.");
// Recover edge-type indices.
spdlog::info("Recreating {} edge-type indices from metadata.", indices_metadata.edge.size());
auto *mem_edge_type_index = static_cast<InMemoryEdgeTypeIndex *>(indices->edge_type_index_.get());
for (const auto &item : indices_metadata.edge) {
if (!mem_edge_type_index->CreateIndex(item, vertices->access())) {
throw RecoveryFailure("The edge-type index must be created here!");
}
spdlog::info("Index on :{} is recreated from metadata", name_id_mapper->IdToName(item.AsUint()));
}
spdlog::info("Edge-type indices are recreated.");
spdlog::info("Recreating constraints from metadata.");
spdlog::info("Indices are recreated.");
}
void RecoverExistenceConstraints(const RecoveredIndicesAndConstraints::ConstraintsMetadata &constraints_metadata,
@ -410,22 +422,17 @@ std::optional<RecoveryInfo> Recovery::RecoverData(std::string *uuid, Replication
std::optional<uint64_t> previous_seq_num;
auto last_loaded_timestamp = snapshot_timestamp;
spdlog::info("Trying to load WAL files.");
if (last_loaded_timestamp) {
epoch_history->emplace_back(repl_storage_state.epoch_.id(), *last_loaded_timestamp);
}
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 != repl_storage_state.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);
}
repl_storage_state.epoch_.SetEpoch(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, config.salient.items);
@ -434,13 +441,24 @@ std::optional<RecoveryInfo> Recovery::RecoverData(std::string *uuid, Replication
recovery_info.next_timestamp = std::max(recovery_info.next_timestamp, info.next_timestamp);
recovery_info.last_commit_timestamp = info.last_commit_timestamp;
if (recovery_info.next_timestamp != 0) {
last_loaded_timestamp.emplace(recovery_info.next_timestamp - 1);
}
auto last_loaded_timestamp_value = last_loaded_timestamp.value_or(0);
if (epoch_history->empty() || epoch_history->back().first != wal_file.epoch_id) {
// no history or new epoch, add it
epoch_history->emplace_back(wal_file.epoch_id, last_loaded_timestamp_value);
repl_storage_state.epoch_.SetEpoch(wal_file.epoch_id);
} else if (epoch_history->back().second < last_loaded_timestamp_value) {
// existing epoch, update with newer timestamp
epoch_history->back().second = last_loaded_timestamp_value;
}
} 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.
@ -456,7 +474,12 @@ std::optional<RecoveryInfo> Recovery::RecoverData(std::string *uuid, Replication
memgraph::metrics::Measure(memgraph::metrics::SnapshotRecoveryLatency_us,
std::chrono::duration_cast<std::chrono::microseconds>(timer.Elapsed()).count());
spdlog::trace("Set epoch id: {} with commit timestamp {}", std::string(repl_storage_state.epoch_.id()),
repl_storage_state.last_commit_timestamp_);
std::for_each(repl_storage_state.history.begin(), repl_storage_state.history.end(), [](auto &history) {
spdlog::trace("epoch id: {} with commit timestamp {}", std::string(history.first), history.second);
});
return recovery_info;
}

Some files were not shown because too many files have changed in this diff Show More