memgraph/tests/unit/interpreter.cpp
2023-12-13 13:19:01 +00:00

1576 lines
65 KiB
C++

// Copyright 2023 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 <algorithm>
#include <cstdlib>
#include <filesystem>
#include "communication/bolt/v1/value.hpp"
#include "communication/result_stream_faker.hpp"
#include "csv/parsing.hpp"
#include "disk_test_utils.hpp"
#include "flags/run_time_configurable.hpp"
#include "glue/communication.hpp"
#include "gmock/gmock.h"
#include "gtest/gtest.h"
#include "interpreter_faker.hpp"
#include "query/auth_checker.hpp"
#include "query/config.hpp"
#include "query/exceptions.hpp"
#include "query/interpreter.hpp"
#include "query/interpreter_context.hpp"
#include "query/metadata.hpp"
#include "query/stream.hpp"
#include "query/typed_value.hpp"
#include "query_common.hpp"
#include "replication/state.hpp"
#include "storage/v2/inmemory/storage.hpp"
#include "storage/v2/isolation_level.hpp"
#include "storage/v2/property_value.hpp"
#include "storage/v2/storage_mode.hpp"
#include "utils/logging.hpp"
#include "utils/lru_cache.hpp"
#include "utils/synchronized.hpp"
namespace {
auto ToEdgeList(const memgraph::communication::bolt::Value &v) {
std::vector<memgraph::communication::bolt::Edge> list;
for (auto x : v.ValueList()) {
list.push_back(x.ValueEdge());
}
return list;
};
} // namespace
// TODO: This is not a unit test, but tests/integration dir is chaotic at the
// moment. After tests refactoring is done, move/rename this.
constexpr auto kNoHandler = nullptr;
template <typename StorageType>
class InterpreterTest : public ::testing::Test {
public:
const std::string testSuite = "interpreter";
const std::string testSuiteCsv = "interpreter_csv";
std::filesystem::path data_directory = std::filesystem::temp_directory_path() / "MG_tests_unit_interpreter";
InterpreterTest() = default;
memgraph::storage::Config config{
[&]() {
memgraph::storage::Config config{};
config.durability.storage_directory = data_directory;
config.disk.main_storage_directory = config.durability.storage_directory / "disk";
if constexpr (std::is_same_v<StorageType, memgraph::storage::DiskStorage>) {
config.disk = disk_test_utils::GenerateOnDiskConfig(testSuite).disk;
config.force_on_disk = true;
}
return config;
}() // iile
};
memgraph::replication::ReplicationState repl_state{memgraph::storage::ReplicationStateRootPath(config)};
memgraph::utils::Gatekeeper<memgraph::dbms::Database> db_gk{config, repl_state};
memgraph::dbms::DatabaseAccess db{
[&]() {
auto db_acc_opt = db_gk.access();
MG_ASSERT(db_acc_opt, "Failed to access db");
auto &db_acc = *db_acc_opt;
MG_ASSERT(db_acc->GetStorageMode() == (std::is_same_v<StorageType, memgraph::storage::DiskStorage>
? memgraph::storage::StorageMode::ON_DISK_TRANSACTIONAL
: memgraph::storage::StorageMode::IN_MEMORY_TRANSACTIONAL),
"Wrong storage mode!");
return db_acc;
}() // iile
};
memgraph::query::InterpreterContext interpreter_context{{}, kNoHandler, &repl_state};
void TearDown() override {
if (std::is_same<StorageType, memgraph::storage::DiskStorage>::value) {
disk_test_utils::RemoveRocksDbDirs(testSuite);
disk_test_utils::RemoveRocksDbDirs(testSuiteCsv);
}
std::filesystem::remove_all(data_directory);
}
InterpreterFaker default_interpreter{&interpreter_context, db};
auto Prepare(const std::string &query, const std::map<std::string, memgraph::storage::PropertyValue> &params = {}) {
return default_interpreter.Prepare(query, params);
}
void Pull(ResultStreamFaker *stream, std::optional<int> n = {}, std::optional<int> qid = {}) {
default_interpreter.Pull(stream, n, qid);
}
auto Interpret(const std::string &query, const std::map<std::string, memgraph::storage::PropertyValue> &params = {}) {
return default_interpreter.Interpret(query, params);
}
};
using StorageTypes = ::testing::Types<memgraph::storage::InMemoryStorage, memgraph::storage::DiskStorage>;
TYPED_TEST_CASE(InterpreterTest, StorageTypes);
TYPED_TEST(InterpreterTest, MultiplePulls) {
{
auto [stream, qid] = this->Prepare("UNWIND [1,2,3,4,5] as n RETURN n");
ASSERT_EQ(stream.GetHeader().size(), 1U);
EXPECT_EQ(stream.GetHeader()[0], "n");
this->Pull(&stream, 1);
ASSERT_EQ(stream.GetSummary().count("has_more"), 1);
ASSERT_TRUE(stream.GetSummary().at("has_more").ValueBool());
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
ASSERT_EQ(stream.GetResults()[0][0].ValueInt(), 1);
this->Pull(&stream, 2);
ASSERT_EQ(stream.GetSummary().count("has_more"), 1);
ASSERT_TRUE(stream.GetSummary().at("has_more").ValueBool());
ASSERT_EQ(stream.GetResults().size(), 3U);
ASSERT_EQ(stream.GetResults()[1][0].ValueInt(), 2);
ASSERT_EQ(stream.GetResults()[2][0].ValueInt(), 3);
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("has_more"), 1);
ASSERT_FALSE(stream.GetSummary().at("has_more").ValueBool());
ASSERT_EQ(stream.GetResults().size(), 5U);
ASSERT_EQ(stream.GetResults()[3][0].ValueInt(), 4);
ASSERT_EQ(stream.GetResults()[4][0].ValueInt(), 5);
}
}
// Run query with different ast twice to see if query executes correctly when
// ast is read from cache.
TYPED_TEST(InterpreterTest, AstCache) {
{
auto stream = this->Interpret("RETURN 2 + 3");
ASSERT_EQ(stream.GetHeader().size(), 1U);
EXPECT_EQ(stream.GetHeader()[0], "2 + 3");
ASSERT_EQ(stream.GetResults().size(), 1U);
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
ASSERT_EQ(stream.GetResults()[0][0].ValueInt(), 5);
}
{
// Cached ast, different literals.
auto stream = this->Interpret("RETURN 5 + 4");
ASSERT_EQ(stream.GetResults().size(), 1U);
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
ASSERT_EQ(stream.GetResults()[0][0].ValueInt(), 9);
}
{
// Different ast (because of different types).
auto stream = this->Interpret("RETURN 5.5 + 4");
ASSERT_EQ(stream.GetResults().size(), 1U);
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
ASSERT_EQ(stream.GetResults()[0][0].ValueDouble(), 9.5);
}
{
// Cached ast, same literals.
auto stream = this->Interpret("RETURN 2 + 3");
ASSERT_EQ(stream.GetResults().size(), 1U);
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
ASSERT_EQ(stream.GetResults()[0][0].ValueInt(), 5);
}
{
// Cached ast, different literals.
auto stream = this->Interpret("RETURN 10.5 + 1");
ASSERT_EQ(stream.GetResults().size(), 1U);
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
ASSERT_EQ(stream.GetResults()[0][0].ValueDouble(), 11.5);
}
{
// Cached ast, same literals, different whitespaces.
auto stream = this->Interpret("RETURN 10.5 + 1");
ASSERT_EQ(stream.GetResults().size(), 1U);
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
ASSERT_EQ(stream.GetResults()[0][0].ValueDouble(), 11.5);
}
{
// Cached ast, same literals, different named header.
auto stream = this->Interpret("RETURN 10.5+1");
ASSERT_EQ(stream.GetHeader().size(), 1U);
EXPECT_EQ(stream.GetHeader()[0], "10.5+1");
ASSERT_EQ(stream.GetResults().size(), 1U);
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
ASSERT_EQ(stream.GetResults()[0][0].ValueDouble(), 11.5);
}
}
// Run query with same ast multiple times with different parameters.
TYPED_TEST(InterpreterTest, Parameters) {
{
auto stream = this->Interpret("RETURN $2 + $`a b`", {{"2", memgraph::storage::PropertyValue(10)},
{"a b", memgraph::storage::PropertyValue(15)}});
ASSERT_EQ(stream.GetHeader().size(), 1U);
EXPECT_EQ(stream.GetHeader()[0], "$2 + $`a b`");
ASSERT_EQ(stream.GetResults().size(), 1U);
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
ASSERT_EQ(stream.GetResults()[0][0].ValueInt(), 25);
}
{
// Not needed parameter.
auto stream = this->Interpret("RETURN $2 + $`a b`", {{"2", memgraph::storage::PropertyValue(10)},
{"a b", memgraph::storage::PropertyValue(15)},
{"c", memgraph::storage::PropertyValue(10)}});
ASSERT_EQ(stream.GetHeader().size(), 1U);
EXPECT_EQ(stream.GetHeader()[0], "$2 + $`a b`");
ASSERT_EQ(stream.GetResults().size(), 1U);
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
ASSERT_EQ(stream.GetResults()[0][0].ValueInt(), 25);
}
{
// Cached ast, different parameters.
auto stream = this->Interpret("RETURN $2 + $`a b`", {{"2", memgraph::storage::PropertyValue("da")},
{"a b", memgraph::storage::PropertyValue("ne")}});
ASSERT_EQ(stream.GetResults().size(), 1U);
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
ASSERT_EQ(stream.GetResults()[0][0].ValueString(), "dane");
}
{
// Non-primitive literal.
auto stream = this->Interpret("RETURN $2",
{{"2", memgraph::storage::PropertyValue(std::vector<memgraph::storage::PropertyValue>{
memgraph::storage::PropertyValue(5), memgraph::storage::PropertyValue(2),
memgraph::storage::PropertyValue(3)})}});
ASSERT_EQ(stream.GetResults().size(), 1U);
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
auto result = memgraph::query::test_common::ToIntList(memgraph::glue::ToTypedValue(stream.GetResults()[0][0]));
ASSERT_THAT(result, testing::ElementsAre(5, 2, 3));
}
{
// Cached ast, unprovided parameter.
ASSERT_THROW(this->Interpret("RETURN $2 + $`a b`", {{"2", memgraph::storage::PropertyValue("da")},
{"ab", memgraph::storage::PropertyValue("ne")}}),
memgraph::query::UnprovidedParameterError);
}
}
// Run CREATE/MATCH/MERGE queries with property map
TYPED_TEST(InterpreterTest, ParametersAsPropertyMap) {
{
std::map<std::string, memgraph::storage::PropertyValue> property_map{};
property_map["name"] = memgraph::storage::PropertyValue("name1");
property_map["age"] = memgraph::storage::PropertyValue(25);
auto stream =
this->Interpret("CREATE (n $prop) RETURN n", {
{"prop", memgraph::storage::PropertyValue(property_map)},
});
ASSERT_EQ(stream.GetHeader().size(), 1U);
ASSERT_EQ(stream.GetHeader()[0], "n");
ASSERT_EQ(stream.GetResults().size(), 1U);
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
auto result = stream.GetResults()[0][0].ValueVertex();
EXPECT_EQ(result.properties["name"].ValueString(), "name1");
EXPECT_EQ(result.properties["age"].ValueInt(), 25);
}
{
std::map<std::string, memgraph::storage::PropertyValue> property_map{};
property_map["name"] = memgraph::storage::PropertyValue("name1");
property_map["age"] = memgraph::storage::PropertyValue(25);
this->Interpret("CREATE (:Person)");
auto stream = this->Interpret("MATCH (m: Person) CREATE (n $prop) RETURN n",
{
{"prop", memgraph::storage::PropertyValue(property_map)},
});
ASSERT_EQ(stream.GetHeader().size(), 1U);
ASSERT_EQ(stream.GetHeader()[0], "n");
ASSERT_EQ(stream.GetResults().size(), 1U);
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
auto result = stream.GetResults()[0][0].ValueVertex();
EXPECT_EQ(result.properties["name"].ValueString(), "name1");
EXPECT_EQ(result.properties["age"].ValueInt(), 25);
}
{
std::map<std::string, memgraph::storage::PropertyValue> property_map{};
property_map["name"] = memgraph::storage::PropertyValue("name1");
property_map["weight"] = memgraph::storage::PropertyValue(121);
auto stream = this->Interpret("CREATE ()-[r:TO $prop]->() RETURN r",
{
{"prop", memgraph::storage::PropertyValue(property_map)},
});
ASSERT_EQ(stream.GetHeader().size(), 1U);
ASSERT_EQ(stream.GetHeader()[0], "r");
ASSERT_EQ(stream.GetResults().size(), 1U);
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
auto result = stream.GetResults()[0][0].ValueEdge();
EXPECT_EQ(result.properties["name"].ValueString(), "name1");
EXPECT_EQ(result.properties["weight"].ValueInt(), 121);
}
{
std::map<std::string, memgraph::storage::PropertyValue> property_map{};
property_map["name"] = memgraph::storage::PropertyValue("name1");
property_map["age"] = memgraph::storage::PropertyValue(15);
ASSERT_THROW(this->Interpret("MATCH (n $prop) RETURN n",
{
{"prop", memgraph::storage::PropertyValue(property_map)},
}),
memgraph::query::SemanticException);
}
{
std::map<std::string, memgraph::storage::PropertyValue> property_map{};
property_map["name"] = memgraph::storage::PropertyValue("name1");
property_map["age"] = memgraph::storage::PropertyValue(15);
ASSERT_THROW(this->Interpret("MERGE (n $prop) RETURN n",
{
{"prop", memgraph::storage::PropertyValue(property_map)},
}),
memgraph::query::SemanticException);
}
}
// Test bfs end to end.
TYPED_TEST(InterpreterTest, Bfs) {
srand(0);
auto kNumLevels = 10;
auto kNumNodesPerLevel = 100;
auto kNumEdgesPerNode = 100;
auto kNumUnreachableNodes = 1000;
auto kNumUnreachableEdges = 100000;
auto kResCoeff = 5;
const auto *const kReachable = "reachable";
const auto kId = "id";
if (std::is_same<TypeParam, memgraph::storage::DiskStorage>::value) {
kNumLevels = 5;
kNumNodesPerLevel = 20;
kNumEdgesPerNode = 20;
kNumUnreachableNodes = 200;
kNumUnreachableEdges = 20000;
kResCoeff = 4;
}
std::vector<std::vector<memgraph::query::VertexAccessor>> levels(kNumLevels);
int id = 0;
// Set up.
{
auto storage_dba = this->db->Access();
memgraph::query::DbAccessor dba(storage_dba.get());
auto add_node = [&](int level, bool reachable) {
auto node = dba.InsertVertex();
MG_ASSERT(node.SetProperty(dba.NameToProperty(kId), memgraph::storage::PropertyValue(id++)).HasValue());
MG_ASSERT(
node.SetProperty(dba.NameToProperty(kReachable), memgraph::storage::PropertyValue(reachable)).HasValue());
levels[level].push_back(node);
return node;
};
auto add_edge = [&](auto &v1, auto &v2, bool reachable) {
auto edge = dba.InsertEdge(&v1, &v2, dba.NameToEdgeType("edge"));
MG_ASSERT(
edge->SetProperty(dba.NameToProperty(kReachable), memgraph::storage::PropertyValue(reachable)).HasValue());
};
// Add source node.
add_node(0, true);
// Add reachable nodes.
for (int i = 1; i < kNumLevels; ++i) {
for (int j = 0; j < kNumNodesPerLevel; ++j) {
auto node = add_node(i, true);
for (int k = 0; k < kNumEdgesPerNode; ++k) {
auto &node2 = levels[i - 1][rand() % levels[i - 1].size()];
add_edge(node2, node, true);
}
}
}
// Add unreachable nodes.
for (int i = 0; i < kNumUnreachableNodes; ++i) {
auto node = add_node(rand() % kNumLevels, // Not really important.
false);
for (int j = 0; j < kNumEdgesPerNode; ++j) {
auto &level = levels[rand() % kNumLevels];
auto &node2 = level[rand() % level.size()];
add_edge(node2, node, true);
add_edge(node, node2, true);
}
}
// Add unreachable edges.
for (int i = 0; i < kNumUnreachableEdges; ++i) {
auto &level1 = levels[rand() % kNumLevels];
auto &node1 = level1[rand() % level1.size()];
auto &level2 = levels[rand() % kNumLevels];
auto &node2 = level2[rand() % level2.size()];
add_edge(node1, node2, false);
}
ASSERT_FALSE(dba.Commit().HasError());
}
auto stream = this->Interpret(
"MATCH (n {id: 0})-[r *bfs..5 (e, n | n.reachable and "
"e.reachable)]->(m) RETURN n, r, m");
ASSERT_EQ(stream.GetHeader().size(), 3U);
EXPECT_EQ(stream.GetHeader()[0], "n");
EXPECT_EQ(stream.GetHeader()[1], "r");
EXPECT_EQ(stream.GetHeader()[2], "m");
ASSERT_EQ(stream.GetResults().size(), kResCoeff * kNumNodesPerLevel);
int expected_level = 1;
int remaining_nodes_in_level = kNumNodesPerLevel;
std::unordered_set<int64_t> matched_ids;
for (const auto &result : stream.GetResults()) {
const auto &begin = result[0].ValueVertex();
const auto &edges = ToEdgeList(result[1]);
const auto &end = result[2].ValueVertex();
// Check that path is of expected length. Returned paths should be from
// shorter to longer ones.
EXPECT_EQ(edges.size(), expected_level);
// Check that starting node is correct.
EXPECT_EQ(edges.front().from, begin.id);
EXPECT_EQ(begin.properties.at(kId).ValueInt(), 0);
for (int i = 1; i < static_cast<int>(edges.size()); ++i) {
// Check that edges form a connected path.
EXPECT_EQ(edges[i - 1].to.AsInt(), edges[i].from.AsInt());
}
auto matched_id = end.properties.at(kId).ValueInt();
EXPECT_EQ(edges.back().to, end.id);
// Check that we didn't match that node already.
EXPECT_TRUE(matched_ids.insert(matched_id).second);
// Check that shortest path was found.
EXPECT_TRUE(matched_id > kNumNodesPerLevel * (expected_level - 1) &&
matched_id <= kNumNodesPerLevel * expected_level);
if (!--remaining_nodes_in_level) {
remaining_nodes_in_level = kNumNodesPerLevel;
++expected_level;
}
}
}
// Test shortest path end to end.
TYPED_TEST(InterpreterTest, ShortestPath) {
const auto test_shortest_path = [this](const bool use_duration) {
const auto get_weight = [use_duration](const auto value) {
return fmt::format(fmt::runtime(use_duration ? "DURATION('PT{}S')" : "{}"), value);
};
this->Interpret(
fmt::format("CREATE (n:A {{x: 1}}), (m:B {{x: 2}}), (l:C {{x: 1}}), (n)-[:r1 {{w: {} "
"}}]->(m)-[:r2 {{w: {}}}]->(l), (n)-[:r3 {{w: {}}}]->(l)",
get_weight(1), get_weight(2), get_weight(4)));
auto stream = this->Interpret("MATCH (n)-[e *wshortest 5 (e, n | e.w) ]->(m) return e");
ASSERT_EQ(stream.GetHeader().size(), 1U);
EXPECT_EQ(stream.GetHeader()[0], "e");
ASSERT_EQ(stream.GetResults().size(), 3U);
std::vector<std::vector<std::string>> expected_results{{"r1"}, {"r2"}, {"r1", "r2"}};
for (const auto &result : stream.GetResults()) {
const auto &edges = ToEdgeList(result[0]);
std::vector<std::string> datum;
datum.reserve(edges.size());
for (const auto &edge : edges) {
datum.push_back(edge.type);
}
bool any_match = false;
for (const auto &expected : expected_results) {
if (expected == datum) {
any_match = true;
break;
}
}
EXPECT_TRUE(any_match);
}
this->Interpret("MATCH (n) DETACH DELETE n");
};
static constexpr bool kUseNumeric{false};
static constexpr bool kUseDuration{true};
{
SCOPED_TRACE("Test with numeric values");
test_shortest_path(kUseNumeric);
}
{
SCOPED_TRACE("Test with Duration values");
test_shortest_path(kUseDuration);
}
}
TYPED_TEST(InterpreterTest, AllShortestById) {
auto stream_init = this->Interpret(
"CREATE (n:A {x: 1}), (m:B {x: 2}), (l:C {x: 3}), (k:D {x: 4}), (n)-[:r1 {w: 1 "
"}]->(m)-[:r2 {w: 2}]->(l), (n)-[:r3 {w: 4}]->(l), (k)-[:r4 {w: 3}]->(l) return id(n), id(l)");
auto id_n = stream_init.GetResults().front()[0].ValueInt();
auto id_l = stream_init.GetResults().front()[1].ValueInt();
auto stream = this->Interpret(
fmt::format("MATCH (n)-[e *allshortest 5 (e, n | e.w) ]->(l) WHERE id(n)={} AND id(l)={} return e", id_n, id_l));
ASSERT_EQ(stream.GetHeader().size(), 1U);
EXPECT_EQ(stream.GetHeader()[0], "e");
ASSERT_EQ(stream.GetResults().size(), 1U);
std::vector<std::string> expected_result = {"r1", "r2"};
const auto &result = stream.GetResults()[0];
const auto &edges = ToEdgeList(result[0]);
std::vector<std::string> datum;
datum.reserve(edges.size());
for (const auto &edge : edges) {
datum.push_back(edge.type);
}
EXPECT_TRUE(expected_result == datum);
this->Interpret("MATCH (n) DETACH DELETE n");
}
TYPED_TEST(InterpreterTest, CreateLabelIndexInMulticommandTransaction) {
this->Interpret("BEGIN");
ASSERT_THROW(this->Interpret("CREATE INDEX ON :X"), memgraph::query::IndexInMulticommandTxException);
this->Interpret("ROLLBACK");
}
TYPED_TEST(InterpreterTest, CreateLabelPropertyIndexInMulticommandTransaction) {
this->Interpret("BEGIN");
ASSERT_THROW(this->Interpret("CREATE INDEX ON :X(y)"), memgraph::query::IndexInMulticommandTxException);
this->Interpret("ROLLBACK");
}
TYPED_TEST(InterpreterTest, CreateExistenceConstraintInMulticommandTransaction) {
this->Interpret("BEGIN");
ASSERT_THROW(this->Interpret("CREATE CONSTRAINT ON (n:A) ASSERT EXISTS (n.a)"),
memgraph::query::ConstraintInMulticommandTxException);
this->Interpret("ROLLBACK");
}
TYPED_TEST(InterpreterTest, CreateUniqueConstraintInMulticommandTransaction) {
this->Interpret("BEGIN");
ASSERT_THROW(this->Interpret("CREATE CONSTRAINT ON (n:A) ASSERT n.a, n.b IS UNIQUE"),
memgraph::query::ConstraintInMulticommandTxException);
this->Interpret("ROLLBACK");
}
TYPED_TEST(InterpreterTest, ShowIndexInfoInMulticommandTransaction) {
this->Interpret("BEGIN");
ASSERT_THROW(this->Interpret("SHOW INDEX INFO"), memgraph::query::InfoInMulticommandTxException);
this->Interpret("ROLLBACK");
}
TYPED_TEST(InterpreterTest, ShowConstraintInfoInMulticommandTransaction) {
this->Interpret("BEGIN");
ASSERT_THROW(this->Interpret("SHOW CONSTRAINT INFO"), memgraph::query::InfoInMulticommandTxException);
this->Interpret("ROLLBACK");
}
TYPED_TEST(InterpreterTest, ShowStorageInfoInMulticommandTransaction) {
this->Interpret("BEGIN");
ASSERT_THROW(this->Interpret("SHOW STORAGE INFO"), memgraph::query::InfoInMulticommandTxException);
this->Interpret("ROLLBACK");
}
// NOLINTNEXTLINE(hicpp-special-member-functions)
TYPED_TEST(InterpreterTest, ExistenceConstraintTest) {
this->Interpret("CREATE CONSTRAINT ON (n:A) ASSERT EXISTS (n.a);");
this->Interpret("CREATE (:A{a:1})");
this->Interpret("CREATE (:A{a:2})");
ASSERT_THROW(this->Interpret("CREATE (:A)"), memgraph::query::QueryException);
this->Interpret("MATCH (n:A{a:2}) SET n.a=3");
this->Interpret("CREATE (:A{a:2})");
this->Interpret("MATCH (n:A{a:2}) DETACH DELETE n");
this->Interpret("CREATE (n:A{a:2})");
ASSERT_THROW(this->Interpret("CREATE CONSTRAINT ON (n:A) ASSERT EXISTS (n.b);"),
memgraph::query::QueryRuntimeException);
}
TYPED_TEST(InterpreterTest, UniqueConstraintTest) {
// Empty property list should result with syntax exception.
ASSERT_THROW(this->Interpret("CREATE CONSTRAINT ON (n:A) ASSERT IS UNIQUE;"), memgraph::query::SyntaxException);
ASSERT_THROW(this->Interpret("DROP CONSTRAINT ON (n:A) ASSERT IS UNIQUE;"), memgraph::query::SyntaxException);
// Too large list of properties should also result with syntax exception.
{
std::stringstream stream;
stream << " ON (n:A) ASSERT ";
for (size_t i = 0; i < 33; ++i) {
if (i > 0) stream << ", ";
stream << "n.prop" << i;
}
stream << " IS UNIQUE;";
std::string create_query = "CREATE CONSTRAINT" + stream.str();
std::string drop_query = "DROP CONSTRAINT" + stream.str();
ASSERT_THROW(this->Interpret(create_query), memgraph::query::SyntaxException);
ASSERT_THROW(this->Interpret(drop_query), memgraph::query::SyntaxException);
}
// Providing property list with duplicates results with syntax exception.
ASSERT_THROW(this->Interpret("CREATE CONSTRAINT ON (n:A) ASSERT n.a, n.b, n.a IS UNIQUE;"),
memgraph::query::SyntaxException);
ASSERT_THROW(this->Interpret("DROP CONSTRAINT ON (n:A) ASSERT n.a, n.b, n.a IS UNIQUE;"),
memgraph::query::SyntaxException);
// Commit of vertex should fail if a constraint is violated.
this->Interpret("CREATE CONSTRAINT ON (n:A) ASSERT n.a, n.b IS UNIQUE;");
this->Interpret("CREATE (:A{a:1, b:2})");
this->Interpret("CREATE (:A{a:1, b:3})");
ASSERT_THROW(this->Interpret("CREATE (:A{a:1, b:2})"), memgraph::query::QueryException);
// Attempt to create a constraint should fail if it's violated.
this->Interpret("CREATE (:A{a:1, c:2})");
this->Interpret("CREATE (:A{a:1, c:2})");
ASSERT_THROW(this->Interpret("CREATE CONSTRAINT ON (n:A) ASSERT n.a, n.c IS UNIQUE;"),
memgraph::query::QueryRuntimeException);
this->Interpret("MATCH (n:A{a:2, b:2}) SET n.a=1");
this->Interpret("CREATE (:A{a:2})");
this->Interpret("MATCH (n:A{a:2}) DETACH DELETE n");
this->Interpret("CREATE (n:A{a:2})");
// Show constraint info.
{
auto stream = this->Interpret("SHOW CONSTRAINT INFO");
ASSERT_EQ(stream.GetHeader().size(), 3U);
const auto &header = stream.GetHeader();
ASSERT_EQ(header[0], "constraint type");
ASSERT_EQ(header[1], "label");
ASSERT_EQ(header[2], "properties");
ASSERT_EQ(stream.GetResults().size(), 1U);
const auto &result = stream.GetResults().front();
ASSERT_EQ(result.size(), 3U);
ASSERT_EQ(result[0].ValueString(), "unique");
ASSERT_EQ(result[1].ValueString(), "A");
const auto &properties = result[2].ValueList();
ASSERT_EQ(properties.size(), 2U);
ASSERT_EQ(properties[0].ValueString(), "a");
ASSERT_EQ(properties[1].ValueString(), "b");
}
// Drop constraint.
this->Interpret("DROP CONSTRAINT ON (n:A) ASSERT n.a, n.b IS UNIQUE;");
// Removing the same constraint twice should not throw any exception.
this->Interpret("DROP CONSTRAINT ON (n:A) ASSERT n.a, n.b IS UNIQUE;");
}
TYPED_TEST(InterpreterTest, ExplainQuery) {
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 0U);
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 0U);
auto stream = this->Interpret("EXPLAIN MATCH (n) RETURN *;");
ASSERT_EQ(stream.GetHeader().size(), 1U);
EXPECT_EQ(stream.GetHeader().front(), "QUERY PLAN");
std::vector<std::string> expected_rows{" * Produce {n}", " * ScanAll (n)", " * Once"};
ASSERT_EQ(stream.GetResults().size(), expected_rows.size());
auto expected_it = expected_rows.begin();
for (const auto &row : stream.GetResults()) {
ASSERT_EQ(row.size(), 1U);
EXPECT_EQ(row.front().ValueString(), *expected_it);
++expected_it;
}
// We should have a plan cache for MATCH ...
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
// We should have AST cache for EXPLAIN ... and for inner MATCH ...
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 2U);
this->Interpret("MATCH (n) RETURN *;");
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 2U);
}
TYPED_TEST(InterpreterTest, ExplainQueryMultiplePulls) {
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 0U);
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 0U);
auto [stream, qid] = this->Prepare("EXPLAIN MATCH (n) RETURN *;");
ASSERT_EQ(stream.GetHeader().size(), 1U);
EXPECT_EQ(stream.GetHeader().front(), "QUERY PLAN");
std::vector<std::string> expected_rows{" * Produce {n}", " * ScanAll (n)", " * Once"};
this->Pull(&stream, 1);
ASSERT_EQ(stream.GetResults().size(), 1);
auto expected_it = expected_rows.begin();
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
EXPECT_EQ(stream.GetResults()[0].front().ValueString(), *expected_it);
++expected_it;
this->Pull(&stream, 1);
ASSERT_EQ(stream.GetResults().size(), 2);
ASSERT_EQ(stream.GetResults()[1].size(), 1U);
EXPECT_EQ(stream.GetResults()[1].front().ValueString(), *expected_it);
++expected_it;
this->Pull(&stream);
ASSERT_EQ(stream.GetResults().size(), 3);
ASSERT_EQ(stream.GetResults()[2].size(), 1U);
EXPECT_EQ(stream.GetResults()[2].front().ValueString(), *expected_it);
// We should have a plan cache for MATCH ...
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
// We should have AST cache for EXPLAIN ... and for inner MATCH ...
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 2U);
this->Interpret("MATCH (n) RETURN *;");
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 2U);
}
TYPED_TEST(InterpreterTest, ExplainQueryInMulticommandTransaction) {
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 0U);
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 0U);
this->Interpret("BEGIN");
auto stream = this->Interpret("EXPLAIN MATCH (n) RETURN *;");
this->Interpret("COMMIT");
ASSERT_EQ(stream.GetHeader().size(), 1U);
EXPECT_EQ(stream.GetHeader().front(), "QUERY PLAN");
std::vector<std::string> expected_rows{" * Produce {n}", " * ScanAll (n)", " * Once"};
ASSERT_EQ(stream.GetResults().size(), expected_rows.size());
auto expected_it = expected_rows.begin();
for (const auto &row : stream.GetResults()) {
ASSERT_EQ(row.size(), 1U);
EXPECT_EQ(row.front().ValueString(), *expected_it);
++expected_it;
}
// We should have a plan cache for MATCH ...
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
// We should have AST cache for EXPLAIN ... and for inner MATCH ...
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 2U);
this->Interpret("MATCH (n) RETURN *;");
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 2U);
}
TYPED_TEST(InterpreterTest, ExplainQueryWithParams) {
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 0U);
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 0U);
auto stream =
this->Interpret("EXPLAIN MATCH (n) WHERE n.id = $id RETURN *;", {{"id", memgraph::storage::PropertyValue(42)}});
ASSERT_EQ(stream.GetHeader().size(), 1U);
EXPECT_EQ(stream.GetHeader().front(), "QUERY PLAN");
std::vector<std::string> expected_rows{" * Produce {n}", " * Filter {n.id}", " * ScanAll (n)", " * Once"};
ASSERT_EQ(stream.GetResults().size(), expected_rows.size());
auto expected_it = expected_rows.begin();
for (const auto &row : stream.GetResults()) {
ASSERT_EQ(row.size(), 1U);
EXPECT_EQ(row.front().ValueString(), *expected_it);
++expected_it;
}
// We should have a plan cache for MATCH ...
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
// We should have AST cache for EXPLAIN ... and for inner MATCH ...
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 2U);
this->Interpret("MATCH (n) WHERE n.id = $id RETURN *;", {{"id", memgraph::storage::PropertyValue("something else")}});
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 2U);
}
TYPED_TEST(InterpreterTest, ProfileQuery) {
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 0U);
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 0U);
auto stream = this->Interpret("PROFILE MATCH (n) RETURN *;");
std::vector<std::string> expected_header{"OPERATOR", "ACTUAL HITS", "RELATIVE TIME", "ABSOLUTE TIME"};
EXPECT_EQ(stream.GetHeader(), expected_header);
std::vector<std::string> expected_rows{"* Produce {n}", "* ScanAll (n)", "* Once"};
ASSERT_EQ(stream.GetResults().size(), expected_rows.size());
auto expected_it = expected_rows.begin();
for (const auto &row : stream.GetResults()) {
ASSERT_EQ(row.size(), 4U);
EXPECT_EQ(row.front().ValueString(), *expected_it);
++expected_it;
}
// We should have a plan cache for MATCH ...
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
// We should have AST cache for PROFILE ... and for inner MATCH ...
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 2U);
this->Interpret("MATCH (n) RETURN *;");
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 2U);
}
TYPED_TEST(InterpreterTest, ProfileQueryMultiplePulls) {
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 0U);
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 0U);
auto [stream, qid] = this->Prepare("PROFILE MATCH (n) RETURN *;");
std::vector<std::string> expected_header{"OPERATOR", "ACTUAL HITS", "RELATIVE TIME", "ABSOLUTE TIME"};
EXPECT_EQ(stream.GetHeader(), expected_header);
std::vector<std::string> expected_rows{"* Produce {n}", "* ScanAll (n)", "* Once"};
auto expected_it = expected_rows.begin();
this->Pull(&stream, 1);
ASSERT_EQ(stream.GetResults().size(), 1U);
ASSERT_EQ(stream.GetResults()[0].size(), 4U);
ASSERT_EQ(stream.GetResults()[0][0].ValueString(), *expected_it);
++expected_it;
this->Pull(&stream, 1);
ASSERT_EQ(stream.GetResults().size(), 2U);
ASSERT_EQ(stream.GetResults()[1].size(), 4U);
ASSERT_EQ(stream.GetResults()[1][0].ValueString(), *expected_it);
++expected_it;
this->Pull(&stream);
ASSERT_EQ(stream.GetResults().size(), 3U);
ASSERT_EQ(stream.GetResults()[2].size(), 4U);
ASSERT_EQ(stream.GetResults()[2][0].ValueString(), *expected_it);
// We should have a plan cache for MATCH ...
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
// We should have AST cache for PROFILE ... and for inner MATCH ...
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 2U);
this->Interpret("MATCH (n) RETURN *;");
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 2U);
}
TYPED_TEST(InterpreterTest, ProfileQueryInMulticommandTransaction) {
this->Interpret("BEGIN");
ASSERT_THROW(this->Interpret("PROFILE MATCH (n) RETURN *;"), memgraph::query::ProfileInMulticommandTxException);
this->Interpret("ROLLBACK");
}
TYPED_TEST(InterpreterTest, ProfileQueryWithParams) {
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 0U);
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 0U);
auto stream =
this->Interpret("PROFILE MATCH (n) WHERE n.id = $id RETURN *;", {{"id", memgraph::storage::PropertyValue(42)}});
std::vector<std::string> expected_header{"OPERATOR", "ACTUAL HITS", "RELATIVE TIME", "ABSOLUTE TIME"};
EXPECT_EQ(stream.GetHeader(), expected_header);
std::vector<std::string> expected_rows{"* Produce {n}", "* Filter {n.id}", "* ScanAll (n)", "* Once"};
ASSERT_EQ(stream.GetResults().size(), expected_rows.size());
auto expected_it = expected_rows.begin();
for (const auto &row : stream.GetResults()) {
ASSERT_EQ(row.size(), 4U);
EXPECT_EQ(row.front().ValueString(), *expected_it);
++expected_it;
}
// We should have a plan cache for MATCH ...
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
// We should have AST cache for PROFILE ... and for inner MATCH ...
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 2U);
this->Interpret("MATCH (n) WHERE n.id = $id RETURN *;", {{"id", memgraph::storage::PropertyValue("something else")}});
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 2U);
}
TYPED_TEST(InterpreterTest, ProfileQueryWithLiterals) {
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 0U);
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 0U);
auto stream = this->Interpret("PROFILE UNWIND range(1, 1000) AS x CREATE (:Node {id: x});", {});
std::vector<std::string> expected_header{"OPERATOR", "ACTUAL HITS", "RELATIVE TIME", "ABSOLUTE TIME"};
EXPECT_EQ(stream.GetHeader(), expected_header);
std::vector<std::string> expected_rows{"* EmptyResult", "* CreateNode", "* Unwind", "* Once"};
ASSERT_EQ(stream.GetResults().size(), expected_rows.size());
auto expected_it = expected_rows.begin();
for (const auto &row : stream.GetResults()) {
ASSERT_EQ(row.size(), 4U);
EXPECT_EQ(row.front().ValueString(), *expected_it);
++expected_it;
}
// We should have a plan cache for UNWIND ...
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
// We should have AST cache for PROFILE ... and for inner UNWIND ...
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 2U);
this->Interpret("UNWIND range(42, 4242) AS x CREATE (:Node {id: x});", {});
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 2U);
}
TYPED_TEST(InterpreterTest, Transactions) {
auto &interpreter = this->default_interpreter.interpreter;
{
ASSERT_THROW(interpreter.CommitTransaction(), memgraph::query::ExplicitTransactionUsageException);
ASSERT_THROW(interpreter.RollbackTransaction(), memgraph::query::ExplicitTransactionUsageException);
interpreter.BeginTransaction();
ASSERT_THROW(interpreter.BeginTransaction(), memgraph::query::ExplicitTransactionUsageException);
auto [stream, qid] = this->Prepare("RETURN 2");
ASSERT_EQ(stream.GetHeader().size(), 1U);
EXPECT_EQ(stream.GetHeader()[0], "2");
this->Pull(&stream, 1);
ASSERT_EQ(stream.GetSummary().count("has_more"), 1);
ASSERT_FALSE(stream.GetSummary().at("has_more").ValueBool());
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
ASSERT_EQ(stream.GetResults()[0][0].ValueInt(), 2);
interpreter.CommitTransaction();
}
{
interpreter.BeginTransaction();
auto [stream, qid] = this->Prepare("RETURN 2");
ASSERT_EQ(stream.GetHeader().size(), 1U);
EXPECT_EQ(stream.GetHeader()[0], "2");
this->Pull(&stream, 1);
ASSERT_EQ(stream.GetSummary().count("has_more"), 1);
ASSERT_FALSE(stream.GetSummary().at("has_more").ValueBool());
ASSERT_EQ(stream.GetResults()[0].size(), 1U);
ASSERT_EQ(stream.GetResults()[0][0].ValueInt(), 2);
interpreter.RollbackTransaction();
}
}
TYPED_TEST(InterpreterTest, Qid) {
auto &interpreter = this->default_interpreter.interpreter;
{
interpreter.BeginTransaction();
auto [stream, qid] = this->Prepare("RETURN 2");
ASSERT_TRUE(qid);
ASSERT_THROW(this->Pull(&stream, {}, *qid + 1), memgraph::query::InvalidArgumentsException);
interpreter.RollbackTransaction();
}
{
interpreter.BeginTransaction();
auto [stream1, qid1] = this->Prepare("UNWIND(range(1,3)) as n RETURN n");
ASSERT_TRUE(qid1);
ASSERT_EQ(stream1.GetHeader().size(), 1U);
EXPECT_EQ(stream1.GetHeader()[0], "n");
auto [stream2, qid2] = this->Prepare("UNWIND(range(4,6)) as n RETURN n");
ASSERT_TRUE(qid2);
ASSERT_EQ(stream2.GetHeader().size(), 1U);
EXPECT_EQ(stream2.GetHeader()[0], "n");
this->Pull(&stream1, 1, qid1);
ASSERT_EQ(stream1.GetSummary().count("has_more"), 1);
ASSERT_TRUE(stream1.GetSummary().at("has_more").ValueBool());
ASSERT_EQ(stream1.GetResults().size(), 1U);
ASSERT_EQ(stream1.GetResults()[0].size(), 1U);
ASSERT_EQ(stream1.GetResults()[0][0].ValueInt(), 1);
auto [stream3, qid3] = this->Prepare("UNWIND(range(7,9)) as n RETURN n");
ASSERT_TRUE(qid3);
ASSERT_EQ(stream3.GetHeader().size(), 1U);
EXPECT_EQ(stream3.GetHeader()[0], "n");
this->Pull(&stream2, {}, qid2);
ASSERT_EQ(stream2.GetSummary().count("has_more"), 1);
ASSERT_FALSE(stream2.GetSummary().at("has_more").ValueBool());
ASSERT_EQ(stream2.GetResults().size(), 3U);
ASSERT_EQ(stream2.GetResults()[0].size(), 1U);
ASSERT_EQ(stream2.GetResults()[0][0].ValueInt(), 4);
ASSERT_EQ(stream2.GetResults()[1][0].ValueInt(), 5);
ASSERT_EQ(stream2.GetResults()[2][0].ValueInt(), 6);
this->Pull(&stream3, 1, qid3);
ASSERT_EQ(stream3.GetSummary().count("has_more"), 1);
ASSERT_TRUE(stream3.GetSummary().at("has_more").ValueBool());
ASSERT_EQ(stream3.GetResults().size(), 1U);
ASSERT_EQ(stream3.GetResults()[0].size(), 1U);
ASSERT_EQ(stream3.GetResults()[0][0].ValueInt(), 7);
this->Pull(&stream1, {}, qid1);
ASSERT_EQ(stream1.GetSummary().count("has_more"), 1);
ASSERT_FALSE(stream1.GetSummary().at("has_more").ValueBool());
ASSERT_EQ(stream1.GetResults().size(), 3U);
ASSERT_EQ(stream1.GetResults()[1].size(), 1U);
ASSERT_EQ(stream1.GetResults()[1][0].ValueInt(), 2);
ASSERT_EQ(stream1.GetResults()[2][0].ValueInt(), 3);
this->Pull(&stream3);
ASSERT_EQ(stream3.GetSummary().count("has_more"), 1);
ASSERT_FALSE(stream3.GetSummary().at("has_more").ValueBool());
ASSERT_EQ(stream3.GetResults().size(), 3U);
ASSERT_EQ(stream3.GetResults()[1].size(), 1U);
ASSERT_EQ(stream3.GetResults()[1][0].ValueInt(), 8);
ASSERT_EQ(stream3.GetResults()[2][0].ValueInt(), 9);
interpreter.CommitTransaction();
}
}
namespace {
// copied from utils_csv_parsing.cpp - tmp dir management and csv file writer
class TmpDirManager final {
public:
explicit TmpDirManager(const std::string_view directory)
: tmp_dir_{std::filesystem::temp_directory_path() / directory} {
CreateDir();
}
~TmpDirManager() { Clear(); }
const std::filesystem::path &Path() const { return tmp_dir_; }
private:
std::filesystem::path tmp_dir_;
void CreateDir() {
if (!std::filesystem::exists(tmp_dir_)) {
std::filesystem::create_directory(tmp_dir_);
}
}
void Clear() {
if (!std::filesystem::exists(tmp_dir_)) return;
std::filesystem::remove_all(tmp_dir_);
}
};
class FileWriter {
public:
explicit FileWriter(const std::filesystem::path path) { stream_.open(path); }
FileWriter(const FileWriter &) = delete;
FileWriter &operator=(const FileWriter &) = delete;
FileWriter(FileWriter &&) = delete;
FileWriter &operator=(FileWriter &&) = delete;
void Close() { stream_.close(); }
size_t WriteLine(const std::string_view line) {
if (!stream_.is_open()) {
return 0;
}
stream_ << line << std::endl;
// including the newline character
return line.size() + 1;
}
private:
std::ofstream stream_;
};
std::string CreateRow(const std::vector<std::string> &columns, const std::string_view delim) {
return memgraph::utils::Join(columns, delim);
}
} // namespace
TYPED_TEST(InterpreterTest, LoadCsvClause) {
auto dir_manager = TmpDirManager("csv_directory");
const auto csv_path = dir_manager.Path() / "file.csv";
auto writer = FileWriter(csv_path);
const std::string delimiter{"|"};
const std::vector<std::string> header{"A", "B", "C"};
writer.WriteLine(CreateRow(header, delimiter));
const std::vector<std::string> good_columns_1{"a", "b", "c"};
writer.WriteLine(CreateRow(good_columns_1, delimiter));
const std::vector<std::string> bad_columns{"\"\"1", "2", "3"};
writer.WriteLine(CreateRow(bad_columns, delimiter));
const std::vector<std::string> good_columns_2{"d", "e", "f"};
writer.WriteLine(CreateRow(good_columns_2, delimiter));
writer.Close();
{
const std::string query = fmt::format(R"(LOAD CSV FROM "{}" WITH HEADER IGNORE BAD DELIMITER "{}" AS x RETURN x.A)",
csv_path.string(), delimiter);
auto [stream, qid] = this->Prepare(query);
ASSERT_EQ(stream.GetHeader().size(), 1U);
EXPECT_EQ(stream.GetHeader()[0], "x.A");
this->Pull(&stream, 1);
ASSERT_EQ(stream.GetSummary().count("has_more"), 1);
ASSERT_TRUE(stream.GetSummary().at("has_more").ValueBool());
ASSERT_EQ(stream.GetResults().size(), 1U);
ASSERT_EQ(stream.GetResults()[0][0].ValueString(), "a");
this->Pull(&stream, 1);
ASSERT_EQ(stream.GetSummary().count("has_more"), 1);
ASSERT_FALSE(stream.GetSummary().at("has_more").ValueBool());
ASSERT_EQ(stream.GetResults().size(), 2U);
ASSERT_EQ(stream.GetResults()[1][0].ValueString(), "d");
}
{
const std::string query = fmt::format(R"(LOAD CSV FROM "{}" WITH HEADER IGNORE BAD DELIMITER "{}" AS x RETURN x.C)",
csv_path.string(), delimiter);
auto [stream, qid] = this->Prepare(query);
ASSERT_EQ(stream.GetHeader().size(), 1U);
EXPECT_EQ(stream.GetHeader()[0], "x.C");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("has_more"), 1);
ASSERT_FALSE(stream.GetSummary().at("has_more").ValueBool());
ASSERT_EQ(stream.GetResults().size(), 2U);
ASSERT_EQ(stream.GetResults()[0][0].ValueString(), "c");
ASSERT_EQ(stream.GetResults()[1][0].ValueString(), "f");
}
}
TYPED_TEST(InterpreterTest, CacheableQueries) {
// This should be cached
{
SCOPED_TRACE("Cacheable query");
this->Interpret("RETURN 1");
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 1U);
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
}
{
SCOPED_TRACE("Uncacheable query");
// Queries which are calling procedure should not be cached because the
// result signature could be changed
this->Interpret("CALL mg.load_all()");
EXPECT_EQ(this->interpreter_context.ast_cache.size(), 1U);
EXPECT_EQ(this->db->plan_cache()->WithLock([&](auto &cache) { return cache.size(); }), 1U);
}
}
TYPED_TEST(InterpreterTest, AllowLoadCsvConfig) {
const auto check_load_csv_queries = [&](const bool allow_load_csv) {
TmpDirManager directory_manager{"allow_load_csv"};
const auto csv_path = directory_manager.Path() / "file.csv";
auto writer = FileWriter(csv_path);
const std::vector<std::string> data{"A", "B", "C"};
writer.WriteLine(CreateRow(data, ","));
writer.Close();
const std::array<std::string, 2> queries = {
fmt::format("LOAD CSV FROM \"{}\" WITH HEADER AS row RETURN row", csv_path.string()),
"CREATE TRIGGER trigger ON CREATE BEFORE COMMIT EXECUTE LOAD CSV FROM 'file.csv' WITH HEADER AS row RETURN "
"row"};
memgraph::storage::Config config2{};
config2.durability.storage_directory = directory_manager.Path();
config2.disk.main_storage_directory = config2.durability.storage_directory / "disk";
if constexpr (std::is_same_v<TypeParam, memgraph::storage::DiskStorage>) {
config2.disk = disk_test_utils::GenerateOnDiskConfig(this->testSuiteCsv).disk;
config2.force_on_disk = true;
}
memgraph::replication::ReplicationState repl_state2{memgraph::storage::ReplicationStateRootPath(config2)};
memgraph::utils::Gatekeeper<memgraph::dbms::Database> db_gk2(config2, repl_state2);
auto db_acc_opt = db_gk2.access();
ASSERT_TRUE(db_acc_opt) << "Failed to access db2";
auto &db_acc = *db_acc_opt;
ASSERT_TRUE(db_acc->GetStorageMode() == (std::is_same_v<TypeParam, memgraph::storage::DiskStorage>
? memgraph::storage::StorageMode::ON_DISK_TRANSACTIONAL
: memgraph::storage::StorageMode::IN_MEMORY_TRANSACTIONAL))
<< "Wrong storage mode!";
memgraph::replication::ReplicationState repl_state{std::nullopt};
memgraph::query::InterpreterContext csv_interpreter_context{
{.query = {.allow_load_csv = allow_load_csv}}, nullptr, &repl_state};
InterpreterFaker interpreter_faker{&csv_interpreter_context, db_acc};
for (const auto &query : queries) {
if (allow_load_csv) {
SCOPED_TRACE(fmt::format("'{}' should not throw because LOAD CSV is allowed", query));
ASSERT_NO_THROW(interpreter_faker.Interpret(query));
} else {
SCOPED_TRACE(fmt::format("'{}' should throw becuase LOAD CSV is not allowed", query));
ASSERT_THROW(interpreter_faker.Interpret(query), memgraph::utils::BasicException);
}
SCOPED_TRACE(fmt::format("Normal query should not throw (allow_load_csv: {})", allow_load_csv));
ASSERT_NO_THROW(interpreter_faker.Interpret("RETURN 1"));
}
};
check_load_csv_queries(true);
check_load_csv_queries(false);
}
void AssertAllValuesAreZero(const std::map<std::string, memgraph::communication::bolt::Value> &map,
const std::vector<std::string> &exceptions) {
for (const auto &[key, value] : map) {
if (const auto it = std::find(exceptions.begin(), exceptions.end(), key); it != exceptions.end()) continue;
ASSERT_EQ(value.ValueInt(), 0);
}
}
TYPED_TEST(InterpreterTest, ExecutionStatsIsValid) {
{
auto [stream, qid] = this->Prepare("MATCH (n) DELETE n;");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("stats"), 0);
}
{
std::array stats_keys{"nodes-created", "nodes-deleted", "relationships-created", "relationships-deleted",
"properties-set", "labels-added", "labels-removed"};
auto [stream, qid] = this->Prepare("CREATE ();");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("stats"), 1);
ASSERT_TRUE(stream.GetSummary().at("stats").IsMap());
auto stats = stream.GetSummary().at("stats").ValueMap();
ASSERT_TRUE(
std::all_of(stats_keys.begin(), stats_keys.end(), [&stats](const auto &key) { return stats.contains(key); }));
AssertAllValuesAreZero(stats, {"nodes-created"});
}
}
TYPED_TEST(InterpreterTest, ExecutionStatsValues) {
{
auto [stream, qid] = this->Prepare("CREATE (),(),(),();");
this->Pull(&stream);
auto stats = stream.GetSummary().at("stats").ValueMap();
ASSERT_EQ(stats["nodes-created"].ValueInt(), 4);
AssertAllValuesAreZero(stats, {"nodes-created"});
}
{
auto [stream, qid] = this->Prepare("MATCH (n) DELETE n;");
this->Pull(&stream);
auto stats = stream.GetSummary().at("stats").ValueMap();
ASSERT_EQ(stats["nodes-deleted"].ValueInt(), 4);
AssertAllValuesAreZero(stats, {"nodes-deleted"});
}
{
auto [stream, qid] = this->Prepare("CREATE (n)-[:TO]->(m), (n)-[:TO]->(m), (n)-[:TO]->(m);");
this->Pull(&stream);
auto stats = stream.GetSummary().at("stats").ValueMap();
ASSERT_EQ(stats["nodes-created"].ValueInt(), 2);
ASSERT_EQ(stats["relationships-created"].ValueInt(), 3);
AssertAllValuesAreZero(stats, {"nodes-created", "relationships-created"});
}
{
auto [stream, qid] = this->Prepare("MATCH (n) DETACH DELETE n;");
this->Pull(&stream);
auto stats = stream.GetSummary().at("stats").ValueMap();
ASSERT_EQ(stats["nodes-deleted"].ValueInt(), 2);
ASSERT_EQ(stats["relationships-deleted"].ValueInt(), 3);
AssertAllValuesAreZero(stats, {"nodes-deleted", "relationships-deleted"});
}
{
auto [stream, qid] = this->Prepare("CREATE (n)-[:TO]->(m);");
this->Pull(&stream);
auto stats = stream.GetSummary().at("stats").ValueMap();
ASSERT_EQ(stats["nodes-created"].ValueInt(), 2);
ASSERT_EQ(stats["relationships-created"].ValueInt(), 1);
AssertAllValuesAreZero(stats, {"nodes-created", "relationships-created"});
}
{
auto [stream, qid] = this->Prepare("MATCH (n)-[r]->(m) DELETE r;");
this->Pull(&stream);
auto stats = stream.GetSummary().at("stats").ValueMap();
ASSERT_EQ(stats["relationships-deleted"].ValueInt(), 1);
AssertAllValuesAreZero(stats, {"nodes-deleted", "relationships-deleted"});
}
{
auto [stream, qid] = this->Prepare("MATCH (n) DELETE n;");
this->Pull(&stream);
auto stats = stream.GetSummary().at("stats").ValueMap();
ASSERT_EQ(stats["nodes-deleted"].ValueInt(), 2);
AssertAllValuesAreZero(stats, {"nodes-deleted", ""});
}
{
auto [stream, qid] = this->Prepare("CREATE (:L1:L2:L3), (:L1), (:L1), (:L2);");
this->Pull(&stream);
auto stats = stream.GetSummary().at("stats").ValueMap();
ASSERT_EQ(stats["nodes-created"].ValueInt(), 4);
ASSERT_EQ(stats["labels-added"].ValueInt(), 6);
AssertAllValuesAreZero(stats, {"nodes-created", "labels-added"});
}
{
auto [stream, qid] = this->Prepare("MATCH (n:L1) SET n.name='test';");
this->Pull(&stream);
auto stats = stream.GetSummary().at("stats").ValueMap();
ASSERT_EQ(stats["properties-set"].ValueInt(), 3);
AssertAllValuesAreZero(stats, {"properties-set"});
}
}
TYPED_TEST(InterpreterTest, ExecutionStatsValuesPropertiesSet) {
{
auto [stream, qid] = this->Prepare(
"CREATE (u:Employee {Uid: 'EMP_AAAAA', FirstName: 'Bong', LastName: 'Revilla'}) RETURN u.name AS name;");
this->Pull(&stream);
}
{
auto [stream, qid] = this->Prepare(
"MATCH (node:Employee) WHERE node.Uid='EMP_AAAAA' SET node={FirstName: 'James', LastName: 'Revilla', Uid: "
"'EMP_AAAAA', CreatedOn: 'null', CreatedBy: 'null', LastModifiedOn: '1698226931701', LastModifiedBy: 'null', "
"Description: 'null'};");
this->Pull(&stream);
auto stats = stream.GetSummary().at("stats").ValueMap();
auto key = memgraph::query::ExecutionStatsKeyToString(memgraph::query::ExecutionStats::Key::UPDATED_PROPERTIES);
ASSERT_EQ(stats[key].ValueInt(), 8);
}
}
TYPED_TEST(InterpreterTest, NotificationsValidStructure) {
{
auto [stream, qid] = this->Prepare("MATCH (n) DELETE n;");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("notifications"), 0);
}
{
auto [stream, qid] = this->Prepare("CREATE INDEX ON :Person(id);");
this->Pull(&stream);
// Assert notifications list
ASSERT_EQ(stream.GetSummary().count("notifications"), 1);
ASSERT_TRUE(stream.GetSummary().at("notifications").IsList());
auto notifications = stream.GetSummary().at("notifications").ValueList();
// Assert one notification structure
ASSERT_EQ(notifications.size(), 1);
ASSERT_TRUE(notifications[0].IsMap());
auto notification = notifications[0].ValueMap();
ASSERT_TRUE(notification.contains("severity"));
ASSERT_TRUE(notification.contains("code"));
ASSERT_TRUE(notification.contains("title"));
ASSERT_TRUE(notification.contains("description"));
ASSERT_TRUE(notification["severity"].IsString());
ASSERT_TRUE(notification["code"].IsString());
ASSERT_TRUE(notification["title"].IsString());
ASSERT_TRUE(notification["description"].IsString());
}
}
TYPED_TEST(InterpreterTest, IndexInfoNotifications) {
{
auto [stream, qid] = this->Prepare("CREATE INDEX ON :Person;");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("notifications"), 1);
auto notifications = stream.GetSummary().at("notifications").ValueList();
auto notification = notifications[0].ValueMap();
ASSERT_EQ(notification["severity"].ValueString(), "INFO");
ASSERT_EQ(notification["code"].ValueString(), "CreateIndex");
ASSERT_EQ(notification["title"].ValueString(), "Created index on label Person on properties .");
ASSERT_EQ(notification["description"].ValueString(), "");
}
{
auto [stream, qid] = this->Prepare("CREATE INDEX ON :Person(id);");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("notifications"), 1);
auto notifications = stream.GetSummary().at("notifications").ValueList();
auto notification = notifications[0].ValueMap();
ASSERT_EQ(notification["severity"].ValueString(), "INFO");
ASSERT_EQ(notification["code"].ValueString(), "CreateIndex");
ASSERT_EQ(notification["title"].ValueString(), "Created index on label Person on properties id.");
ASSERT_EQ(notification["description"].ValueString(), "");
}
{
auto [stream, qid] = this->Prepare("CREATE INDEX ON :Person(id);");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("notifications"), 1);
auto notifications = stream.GetSummary().at("notifications").ValueList();
auto notification = notifications[0].ValueMap();
ASSERT_EQ(notification["severity"].ValueString(), "INFO");
ASSERT_EQ(notification["code"].ValueString(), "IndexAlreadyExists");
ASSERT_EQ(notification["title"].ValueString(), "Index on label Person on properties id already exists.");
ASSERT_EQ(notification["description"].ValueString(), "");
}
{
auto [stream, qid] = this->Prepare("DROP INDEX ON :Person(id);");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("notifications"), 1);
auto notifications = stream.GetSummary().at("notifications").ValueList();
auto notification = notifications[0].ValueMap();
ASSERT_EQ(notification["severity"].ValueString(), "INFO");
ASSERT_EQ(notification["code"].ValueString(), "DropIndex");
ASSERT_EQ(notification["title"].ValueString(), "Dropped index on label Person on properties id.");
ASSERT_EQ(notification["description"].ValueString(), "");
}
{
auto [stream, qid] = this->Prepare("DROP INDEX ON :Person(id);");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("notifications"), 1);
auto notifications = stream.GetSummary().at("notifications").ValueList();
auto notification = notifications[0].ValueMap();
ASSERT_EQ(notification["severity"].ValueString(), "INFO");
ASSERT_EQ(notification["code"].ValueString(), "IndexDoesNotExist");
ASSERT_EQ(notification["title"].ValueString(), "Index on label Person on properties id doesn't exist.");
ASSERT_EQ(notification["description"].ValueString(), "");
}
}
TYPED_TEST(InterpreterTest, ConstraintUniqueInfoNotifications) {
{
auto [stream, qid] = this->Prepare("CREATE CONSTRAINT ON (n:Person) ASSERT n.email, n.id IS UNIQUE;");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("notifications"), 1);
auto notifications = stream.GetSummary().at("notifications").ValueList();
auto notification = notifications[0].ValueMap();
ASSERT_EQ(notification["severity"].ValueString(), "INFO");
ASSERT_EQ(notification["code"].ValueString(), "CreateConstraint");
ASSERT_EQ(notification["title"].ValueString(),
"Created UNIQUE constraint on label Person on properties email, id.");
ASSERT_EQ(notification["description"].ValueString(), "");
}
{
auto [stream, qid] = this->Prepare("CREATE CONSTRAINT ON (n:Person) ASSERT n.email, n.id IS UNIQUE;");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("notifications"), 1);
auto notifications = stream.GetSummary().at("notifications").ValueList();
auto notification = notifications[0].ValueMap();
ASSERT_EQ(notification["severity"].ValueString(), "INFO");
ASSERT_EQ(notification["code"].ValueString(), "ConstraintAlreadyExists");
ASSERT_EQ(notification["title"].ValueString(),
"Constraint UNIQUE on label Person on properties email, id already exists.");
ASSERT_EQ(notification["description"].ValueString(), "");
}
{
auto [stream, qid] = this->Prepare("DROP CONSTRAINT ON (n:Person) ASSERT n.email, n.id IS UNIQUE;");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("notifications"), 1);
auto notifications = stream.GetSummary().at("notifications").ValueList();
auto notification = notifications[0].ValueMap();
ASSERT_EQ(notification["severity"].ValueString(), "INFO");
ASSERT_EQ(notification["code"].ValueString(), "DropConstraint");
ASSERT_EQ(notification["title"].ValueString(),
"Dropped UNIQUE constraint on label Person on properties email, id.");
ASSERT_EQ(notification["description"].ValueString(), "");
}
{
auto [stream, qid] = this->Prepare("DROP CONSTRAINT ON (n:Person) ASSERT n.email, n.id IS UNIQUE;");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("notifications"), 1);
auto notifications = stream.GetSummary().at("notifications").ValueList();
auto notification = notifications[0].ValueMap();
ASSERT_EQ(notification["severity"].ValueString(), "INFO");
ASSERT_EQ(notification["code"].ValueString(), "ConstraintDoesNotExist");
ASSERT_EQ(notification["title"].ValueString(),
"Constraint UNIQUE on label Person on properties email, id doesn't exist.");
ASSERT_EQ(notification["description"].ValueString(), "");
}
}
TYPED_TEST(InterpreterTest, ConstraintExistsInfoNotifications) {
{
auto [stream, qid] = this->Prepare("CREATE CONSTRAINT ON (n:L1) ASSERT EXISTS (n.name);");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("notifications"), 1);
auto notifications = stream.GetSummary().at("notifications").ValueList();
auto notification = notifications[0].ValueMap();
ASSERT_EQ(notification["severity"].ValueString(), "INFO");
ASSERT_EQ(notification["code"].ValueString(), "CreateConstraint");
ASSERT_EQ(notification["title"].ValueString(), "Created EXISTS constraint on label L1 on properties name.");
ASSERT_EQ(notification["description"].ValueString(), "");
}
{
auto [stream, qid] = this->Prepare("CREATE CONSTRAINT ON (n:L1) ASSERT EXISTS (n.name);");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("notifications"), 1);
auto notifications = stream.GetSummary().at("notifications").ValueList();
auto notification = notifications[0].ValueMap();
ASSERT_EQ(notification["severity"].ValueString(), "INFO");
ASSERT_EQ(notification["code"].ValueString(), "ConstraintAlreadyExists");
ASSERT_EQ(notification["title"].ValueString(), "Constraint EXISTS on label L1 on properties name already exists.");
ASSERT_EQ(notification["description"].ValueString(), "");
}
{
auto [stream, qid] = this->Prepare("DROP CONSTRAINT ON (n:L1) ASSERT EXISTS (n.name);");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("notifications"), 1);
auto notifications = stream.GetSummary().at("notifications").ValueList();
auto notification = notifications[0].ValueMap();
ASSERT_EQ(notification["severity"].ValueString(), "INFO");
ASSERT_EQ(notification["code"].ValueString(), "DropConstraint");
ASSERT_EQ(notification["title"].ValueString(), "Dropped EXISTS constraint on label L1 on properties name.");
ASSERT_EQ(notification["description"].ValueString(), "");
}
{
auto [stream, qid] = this->Prepare("DROP CONSTRAINT ON (n:L1) ASSERT EXISTS (n.name);");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("notifications"), 1);
auto notifications = stream.GetSummary().at("notifications").ValueList();
auto notification = notifications[0].ValueMap();
ASSERT_EQ(notification["severity"].ValueString(), "INFO");
ASSERT_EQ(notification["code"].ValueString(), "ConstraintDoesNotExist");
ASSERT_EQ(notification["title"].ValueString(), "Constraint EXISTS on label L1 on properties name doesn't exist.");
ASSERT_EQ(notification["description"].ValueString(), "");
}
}
TYPED_TEST(InterpreterTest, TriggerInfoNotifications) {
{
auto [stream, qid] = this->Prepare(
"CREATE TRIGGER bestTriggerEver ON CREATE AFTER COMMIT EXECUTE "
"CREATE ();");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("notifications"), 1);
auto notifications = stream.GetSummary().at("notifications").ValueList();
auto notification = notifications[0].ValueMap();
ASSERT_EQ(notification["severity"].ValueString(), "INFO");
ASSERT_EQ(notification["code"].ValueString(), "CreateTrigger");
ASSERT_EQ(notification["title"].ValueString(), "Created trigger bestTriggerEver.");
ASSERT_EQ(notification["description"].ValueString(), "");
}
{
auto [stream, qid] = this->Prepare("DROP TRIGGER bestTriggerEver;");
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("notifications"), 1);
auto notifications = stream.GetSummary().at("notifications").ValueList();
auto notification = notifications[0].ValueMap();
ASSERT_EQ(notification["severity"].ValueString(), "INFO");
ASSERT_EQ(notification["code"].ValueString(), "DropTrigger");
ASSERT_EQ(notification["title"].ValueString(), "Dropped trigger bestTriggerEver.");
ASSERT_EQ(notification["description"].ValueString(), "");
}
}
TYPED_TEST(InterpreterTest, LoadCsvClauseNotification) {
auto dir_manager = TmpDirManager("csv_directory");
const auto csv_path = dir_manager.Path() / "file.csv";
auto writer = FileWriter(csv_path);
const std::string delimiter{"|"};
const std::vector<std::string> header{"A", "B", "C"};
writer.WriteLine(CreateRow(header, delimiter));
const std::vector<std::string> good_columns_1{"a", "b", "c"};
writer.WriteLine(CreateRow(good_columns_1, delimiter));
writer.Close();
const std::string query = fmt::format(R"(LOAD CSV FROM "{}" WITH HEADER IGNORE BAD DELIMITER "{}" AS x RETURN x;)",
csv_path.string(), delimiter);
auto [stream, qid] = this->Prepare(query);
this->Pull(&stream);
ASSERT_EQ(stream.GetSummary().count("notifications"), 1);
auto notifications = stream.GetSummary().at("notifications").ValueList();
auto notification = notifications[0].ValueMap();
ASSERT_EQ(notification["severity"].ValueString(), "INFO");
ASSERT_EQ(notification["code"].ValueString(), "LoadCSVTip");
ASSERT_EQ(notification["title"].ValueString(),
"It's important to note that the parser parses the values as strings. It's up to the user to "
"convert the parsed row values to the appropriate type. This can be done using the built-in "
"conversion functions such as ToInteger, ToFloat, ToBoolean etc.");
ASSERT_EQ(notification["description"].ValueString(), "");
}