thevar1able/ClickHouse
1
0
Fork 0
mirror of https://github.com/ClickHouse/ClickHouse.git synced 2024-11-21 15:12:02 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'master' into fix-toDayOfWeek-mono

Browse Source
This commit is contained in:
Yakov Olkhovskiy 2024-11-14 14:37:06 +00:00
commit 43c2b69b6c
135 changed files with 2178 additions and 1052 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
README.md
View File

@ -42,7 +42,6 @@ Keep an eye out for upcoming meetups and events around the world. Somewhere else
Upcoming meetups
* [Barcelona Meetup](https://www.meetup.com/clickhouse-spain-user-group/events/303096876/) - November 12
* [Ghent Meetup](https://www.meetup.com/clickhouse-belgium-user-group/events/303049405/) - November 19
* [Dubai Meetup](https://www.meetup.com/clickhouse-dubai-meetup-group/events/303096989/) - November 21
* [Paris Meetup](https://www.meetup.com/clickhouse-france-user-group/events/303096434) - November 26
@ -53,6 +52,7 @@ Upcoming meetups
Recently completed meetups
* [Barcelona Meetup](https://www.meetup.com/clickhouse-spain-user-group/events/303096876/) - November 12
* [Madrid Meetup](https://www.meetup.com/clickhouse-spain-user-group/events/303096564/) - October 22
* [Singapore Meetup](https://www.meetup.com/clickhouse-singapore-meetup-group/events/303212064/) - October 3
* [Jakarta Meetup](https://www.meetup.com/clickhouse-indonesia-user-group/events/303191359/) - October 1

1
base/base/defines.h
View File

@ -145,6 +145,7 @@
#define TSA_TRY_ACQUIRE_SHARED(...) __attribute__((try_acquire_shared_capability(__VA_ARGS__))) /// function tries to acquire a shared capability and returns a boolean value indicating success or failure
#define TSA_RELEASE_SHARED(...) __attribute__((release_shared_capability(__VA_ARGS__))) /// function releases the given shared capability
#define TSA_SCOPED_LOCKABLE __attribute__((scoped_lockable)) /// object of a class has scoped lockable capability
#define TSA_RETURN_CAPABILITY(...) __attribute__((lock_returned(__VA_ARGS__))) /// to return capabilities in functions
/// Macros for suppressing TSA warnings for specific reads/writes (instead of suppressing it for the whole function)
/// They use a lambda function to apply function attribute to a single statement. This enable us to suppress warnings locally instead of

24
docs/en/operations/server-configuration-parameters/settings.md
View File

@ -597,6 +597,30 @@ If number of tables is greater than this value, server will throw an exception.
<max_table_num_to_throw>400</max_table_num_to_throw>
```
## max\_replicated\_table\_num\_to\_throw {#max-replicated-table-num-to-throw}
If number of replicated tables is greater than this value, server will throw an exception. 0 means no limitation. Only count table in Atomic/Ordinary/Replicated/Lazy database engine.
**Example**
```xml
<max_replicated_table_num_to_throw>400</max_replicated_table_num_to_throw>
```
## max\_dictionary\_num\_to\_throw {#max-dictionary-num-to-throw}
If number of dictionaries is greater than this value, server will throw an exception. 0 means no limitation. Only count table in Atomic/Ordinary/Replicated/Lazy database engine.
**Example**
```xml
<max_dictionary_num_to_throw>400</max_dictionary_num_to_throw>
```
## max\_view\_num\_to\_throw {#max-view-num-to-throw}
If number of views is greater than this value, server will throw an exception. 0 means no limitation. Only count table in Atomic/Ordinary/Replicated/Lazy database engine.
**Example**
```xml
<max_view_num_to_throw>400</max_view_num_to_throw>
```
## max\_database\_num\_to\_throw {#max-table-num-to-throw}
If number of _database is greater than this value, server will throw an exception. 0 means no limitation.
Default value: 0

20
programs/compressor/Compressor.cpp
View File

@ -12,9 +12,12 @@
#include <Compression/ParallelCompressedWriteBuffer.h>
#include <Compression/CompressedReadBuffer.h>
#include <Compression/CompressedReadBufferFromFile.h>
#include <Compression/getCompressionCodecForFile.h>
#include <IO/ReadHelpers.h>
#include <IO/WriteHelpers.h>
#include <IO/copyData.h>
#include <Parsers/parseQuery.h>
#include <Parsers/queryToString.h>
#include <Parsers/ExpressionElementParsers.h>
#include <Compression/CompressionFactory.h>
#include <Common/TerminalSize.h>
@ -43,29 +46,24 @@ namespace CurrentMetrics
namespace
{
/// Outputs sizes of uncompressed and compressed blocks for compressed file.
/// Outputs method, sizes of uncompressed and compressed blocks for compressed file.
void checkAndWriteHeader(DB::ReadBuffer & in, DB::WriteBuffer & out)
{
while (!in.eof())
{
in.ignore(16); /// checksum
char header[COMPRESSED_BLOCK_HEADER_SIZE];
in.readStrict(header, COMPRESSED_BLOCK_HEADER_SIZE);
UInt32 size_compressed = unalignedLoad<UInt32>(&header[1]);
UInt32 size_compressed;
UInt32 size_decompressed;
auto codec = DB::getCompressionCodecForFile(in, size_compressed, size_decompressed, true /* skip_to_next_block */);
if (size_compressed > DBMS_MAX_COMPRESSED_SIZE)
throw DB::Exception(DB::ErrorCodes::TOO_LARGE_SIZE_COMPRESSED, "Too large size_compressed. Most likely corrupted data.");
UInt32 size_decompressed = unalignedLoad<UInt32>(&header[5]);
DB::writeText(queryToString(codec->getFullCodecDesc()), out);
DB::writeChar('\t', out);
DB::writeText(size_decompressed, out);
DB::writeChar('\t', out);
DB::writeText(size_compressed, out);
DB::writeChar('\n', out);
in.ignore(size_compressed - COMPRESSED_BLOCK_HEADER_SIZE);
}
}

16
src/Backups/BackupConcurrencyCheck.cpp
View File

@ -14,12 +14,12 @@ namespace ErrorCodes
BackupConcurrencyCheck::BackupConcurrencyCheck(
const UUID & backup_or_restore_uuid_,
bool is_restore_,
bool on_cluster_,
const String & zookeeper_path_,
bool allow_concurrency_,
BackupConcurrencyCounters & counters_)
: is_restore(is_restore_), backup_or_restore_uuid(backup_or_restore_uuid_), on_cluster(on_cluster_), counters(counters_)
: is_restore(is_restore_), on_cluster(on_cluster_), zookeeper_path(zookeeper_path_), counters(counters_)
{
std::lock_guard lock{counters.mutex};
@ -32,7 +32,7 @@ BackupConcurrencyCheck::BackupConcurrencyCheck(
size_t num_on_cluster_restores = counters.on_cluster_restores.size();
if (on_cluster)
{
if (!counters.on_cluster_restores.contains(backup_or_restore_uuid))
if (!counters.on_cluster_restores.contains(zookeeper_path))
++num_on_cluster_restores;
}
else
@ -47,7 +47,7 @@ BackupConcurrencyCheck::BackupConcurrencyCheck(
size_t num_on_cluster_backups = counters.on_cluster_backups.size();
if (on_cluster)
{
if (!counters.on_cluster_backups.contains(backup_or_restore_uuid))
if (!counters.on_cluster_backups.contains(zookeeper_path))
++num_on_cluster_backups;
}
else
@ -64,9 +64,9 @@ BackupConcurrencyCheck::BackupConcurrencyCheck(
if (on_cluster)
{
if (is_restore)
++counters.on_cluster_restores[backup_or_restore_uuid];
++counters.on_cluster_restores[zookeeper_path];
else
++counters.on_cluster_backups[backup_or_restore_uuid];
++counters.on_cluster_backups[zookeeper_path];
}
else
{
@ -86,7 +86,7 @@ BackupConcurrencyCheck::~BackupConcurrencyCheck()
{
if (is_restore)
{
auto it = counters.on_cluster_restores.find(backup_or_restore_uuid);
auto it = counters.on_cluster_restores.find(zookeeper_path);
if (it != counters.on_cluster_restores.end())
{
if (!--it->second)
@ -95,7 +95,7 @@ BackupConcurrencyCheck::~BackupConcurrencyCheck()
}
else
{
auto it = counters.on_cluster_backups.find(backup_or_restore_uuid);
auto it = counters.on_cluster_backups.find(zookeeper_path);
if (it != counters.on_cluster_backups.end())
{
if (!--it->second)

11
src/Backups/BackupConcurrencyCheck.h
View File

@ -1,7 +1,8 @@
#pragma once
#include <Core/UUID.h>
#include <base/defines.h>
#include <base/scope_guard.h>
#include <base/types.h>
#include <mutex>
#include <unordered_map>
@ -19,9 +20,9 @@ public:
/// Checks concurrency of a BACKUP operation or a RESTORE operation.
/// Keep a constructed instance of BackupConcurrencyCheck until the operation is done.
BackupConcurrencyCheck(
const UUID & backup_or_restore_uuid_,
bool is_restore_,
bool on_cluster_,
const String & zookeeper_path_,
bool allow_concurrency_,
BackupConcurrencyCounters & counters_);
@ -31,8 +32,8 @@ public:
private:
const bool is_restore;
const UUID backup_or_restore_uuid;
const bool on_cluster;
const String zookeeper_path;
BackupConcurrencyCounters & counters;
};
@ -47,8 +48,8 @@ private:
friend class BackupConcurrencyCheck;
size_t local_backups TSA_GUARDED_BY(mutex) = 0;
size_t local_restores TSA_GUARDED_BY(mutex) = 0;
std::unordered_map<UUID /* backup_uuid */, size_t /* num_refs */> on_cluster_backups TSA_GUARDED_BY(mutex);
std::unordered_map<UUID /* restore_uuid */, size_t /* num_refs */> on_cluster_restores TSA_GUARDED_BY(mutex);
std::unordered_map<String /* zookeeper_path */, size_t /* num_refs */> on_cluster_backups TSA_GUARDED_BY(mutex);
std::unordered_map<String /* zookeeper_path */, size_t /* num_refs */> on_cluster_restores TSA_GUARDED_BY(mutex);
std::mutex mutex;
};

36
src/Backups/BackupCoordinationCleaner.cpp
View File

@ -4,31 +4,29 @@
namespace DB
{
BackupCoordinationCleaner::BackupCoordinationCleaner(const String & zookeeper_path_, const WithRetries & with_retries_, LoggerPtr log_)
: zookeeper_path(zookeeper_path_), with_retries(with_retries_), log(log_)
BackupCoordinationCleaner::BackupCoordinationCleaner(bool is_restore_, const String & zookeeper_path_, const WithRetries & with_retries_, LoggerPtr log_)
: is_restore(is_restore_), zookeeper_path(zookeeper_path_), with_retries(with_retries_), log(log_)
{
}
void BackupCoordinationCleaner::cleanup()
bool BackupCoordinationCleaner::cleanup(bool throw_if_error)
{
tryRemoveAllNodes(/* throw_if_error = */ true, /* retries_kind = */ WithRetries::kNormal);
WithRetries::Kind retries_kind = throw_if_error ? WithRetries::kNormal : WithRetries::kErrorHandling;
return cleanupImpl(throw_if_error, retries_kind);
}
bool BackupCoordinationCleaner::tryCleanupAfterError() noexcept
{
return tryRemoveAllNodes(/* throw_if_error = */ false, /* retries_kind = */ WithRetries::kNormal);
}
bool BackupCoordinationCleaner::tryRemoveAllNodes(bool throw_if_error, WithRetries::Kind retries_kind)
bool BackupCoordinationCleaner::cleanupImpl(bool throw_if_error, WithRetries::Kind retries_kind)
{
{
std::lock_guard lock{mutex};
if (cleanup_result.succeeded)
return true;
if (cleanup_result.exception)
if (succeeded)
{
if (throw_if_error)
std::rethrow_exception(cleanup_result.exception);
LOG_TRACE(log, "Nodes from ZooKeeper are already removed");
return true;
}
if (tried)
{
LOG_INFO(log, "Skipped removing nodes from ZooKeeper because because earlier we failed to do that");
return false;
}
}
@ -44,16 +42,18 @@ bool BackupCoordinationCleaner::tryRemoveAllNodes(bool throw_if_error, WithRetri
});
std::lock_guard lock{mutex};
cleanup_result.succeeded = true;
tried = true;
succeeded = true;
return true;
}
catch (...)
{
LOG_TRACE(log, "Caught exception while removing nodes from ZooKeeper for this restore: {}",
LOG_TRACE(log, "Caught exception while removing nodes from ZooKeeper for this {}: {}",
is_restore ? "restore" : "backup",
getCurrentExceptionMessage(/* with_stacktrace= */ false, /* check_embedded_stacktrace= */ true));
std::lock_guard lock{mutex};
cleanup_result.exception = std::current_exception();
tried = true;
if (throw_if_error)
throw;

17
src/Backups/BackupCoordinationCleaner.h
View File

@ -12,14 +12,14 @@ namespace DB
class BackupCoordinationCleaner
{
public:
BackupCoordinationCleaner(const String & zookeeper_path_, const WithRetries & with_retries_, LoggerPtr log_);
BackupCoordinationCleaner(bool is_restore_, const String & zookeeper_path_, const WithRetries & with_retries_, LoggerPtr log_);
void cleanup();
bool tryCleanupAfterError() noexcept;
bool cleanup(bool throw_if_error);
private:
bool tryRemoveAllNodes(bool throw_if_error, WithRetries::Kind retries_kind);
bool cleanupImpl(bool throw_if_error, WithRetries::Kind retries_kind);
const bool is_restore;
const String zookeeper_path;
/// A reference to a field of the parent object which is either BackupCoordinationOnCluster or RestoreCoordinationOnCluster.
@ -27,13 +27,8 @@ private:
const LoggerPtr log;
struct CleanupResult
{
bool succeeded = false;
std::exception_ptr exception;
};
CleanupResult cleanup_result TSA_GUARDED_BY(mutex);
bool tried TSA_GUARDED_BY(mutex) = false;
bool succeeded TSA_GUARDED_BY(mutex) = false;
std::mutex mutex;
};

3
src/Backups/BackupCoordinationLocal.cpp
View File

@ -11,12 +11,11 @@ namespace DB
{
BackupCoordinationLocal::BackupCoordinationLocal(
const UUID & backup_uuid_,
bool is_plain_backup_,
bool allow_concurrent_backup_,
BackupConcurrencyCounters & concurrency_counters_)
: log(getLogger("BackupCoordinationLocal"))
, concurrency_check(backup_uuid_, /* is_restore = */ false, /* on_cluster = */ false, allow_concurrent_backup_, concurrency_counters_)
, concurrency_check(/* is_restore = */ false, /* on_cluster = */ false, /* zookeeper_path = */ "", allow_concurrent_backup_, concurrency_counters_)
, file_infos(is_plain_backup_)
{
}

13
src/Backups/BackupCoordinationLocal.h
View File

@ -23,20 +23,19 @@ class BackupCoordinationLocal : public IBackupCoordination
{
public:
explicit BackupCoordinationLocal(
const UUID & backup_uuid_,
bool is_plain_backup_,
bool allow_concurrent_backup_,
BackupConcurrencyCounters & concurrency_counters_);
~BackupCoordinationLocal() override;
void setBackupQueryIsSentToOtherHosts() override {}
bool isBackupQuerySentToOtherHosts() const override { return false; }
Strings setStage(const String &, const String &, bool) override { return {}; }
void setBackupQueryWasSentToOtherHosts() override {}
bool trySetError(std::exception_ptr) override { return true; }
void finish() override {}
bool tryFinishAfterError() noexcept override { return true; }
void waitForOtherHostsToFinish() override {}
bool tryWaitForOtherHostsToFinishAfterError() noexcept override { return true; }
bool setError(std::exception_ptr, bool) override { return true; }
bool waitOtherHostsFinish(bool) const override { return true; }
bool finish(bool) override { return true; }
bool cleanup(bool) override { return true; }
void addReplicatedPartNames(const String & table_zk_path, const String & table_name_for_logs, const String & replica_name,
const std::vector<PartNameAndChecksum> & part_names_and_checksums) override;

95
src/Backups/BackupCoordinationOnCluster.cpp
View File

@ -184,17 +184,21 @@ BackupCoordinationOnCluster::BackupCoordinationOnCluster(
, plain_backup(is_plain_backup_)
, log(getLogger("BackupCoordinationOnCluster"))
, with_retries(log, get_zookeeper_, keeper_settings, process_list_element_, [root_zookeeper_path_](Coordination::ZooKeeperWithFaultInjection::Ptr zk) { zk->sync(root_zookeeper_path_); })
, concurrency_check(backup_uuid_, /* is_restore = */ false, /* on_cluster = */ true, allow_concurrent_backup_, concurrency_counters_)
, stage_sync(/* is_restore = */ false, fs::path{zookeeper_path} / "stage", current_host, all_hosts, allow_concurrent_backup_, with_retries, schedule_, process_list_element_, log)
, cleaner(zookeeper_path, with_retries, log)
, cleaner(/* is_restore = */ false, zookeeper_path, with_retries, log)
, stage_sync(/* is_restore = */ false, fs::path{zookeeper_path} / "stage", current_host, all_hosts, allow_concurrent_backup_, concurrency_counters_, with_retries, schedule_, process_list_element_, log)
{
createRootNodes();
try
{
createRootNodes();
}
catch (...)
{
stage_sync.setError(std::current_exception(), /* throw_if_error = */ false);
throw;
}
}
BackupCoordinationOnCluster::~BackupCoordinationOnCluster()
{
tryFinishImpl();
}
BackupCoordinationOnCluster::~BackupCoordinationOnCluster() = default;
void BackupCoordinationOnCluster::createRootNodes()
{
@ -217,69 +221,52 @@ void BackupCoordinationOnCluster::createRootNodes()
});
}
void BackupCoordinationOnCluster::setBackupQueryIsSentToOtherHosts()
{
stage_sync.setQueryIsSentToOtherHosts();
}
bool BackupCoordinationOnCluster::isBackupQuerySentToOtherHosts() const
{
return stage_sync.isQuerySentToOtherHosts();
}
Strings BackupCoordinationOnCluster::setStage(const String & new_stage, const String & message, bool sync)
{
stage_sync.setStage(new_stage, message);
if (!sync)
return {};
return stage_sync.waitForHostsToReachStage(new_stage, all_hosts_without_initiator);
if (sync)
return stage_sync.waitHostsReachStage(all_hosts_without_initiator, new_stage);
return {};
}
void BackupCoordinationOnCluster::setBackupQueryWasSentToOtherHosts()
bool BackupCoordinationOnCluster::setError(std::exception_ptr exception, bool throw_if_error)
{
backup_query_was_sent_to_other_hosts = true;
return stage_sync.setError(exception, throw_if_error);
}
bool BackupCoordinationOnCluster::trySetError(std::exception_ptr exception)
bool BackupCoordinationOnCluster::waitOtherHostsFinish(bool throw_if_error) const
{
return stage_sync.trySetError(exception);
return stage_sync.waitOtherHostsFinish(throw_if_error);
}
void BackupCoordinationOnCluster::finish()
bool BackupCoordinationOnCluster::finish(bool throw_if_error)
{
bool other_hosts_also_finished = false;
stage_sync.finish(other_hosts_also_finished);
if ((current_host == kInitiator) && (other_hosts_also_finished || !backup_query_was_sent_to_other_hosts))
cleaner.cleanup();
return stage_sync.finish(throw_if_error);
}
bool BackupCoordinationOnCluster::tryFinishAfterError() noexcept
bool BackupCoordinationOnCluster::cleanup(bool throw_if_error)
{
return tryFinishImpl();
}
bool BackupCoordinationOnCluster::tryFinishImpl() noexcept
{
bool other_hosts_also_finished = false;
if (!stage_sync.tryFinishAfterError(other_hosts_also_finished))
return false;
if ((current_host == kInitiator) && (other_hosts_also_finished || !backup_query_was_sent_to_other_hosts))
/// All the hosts must finish before we remove the coordination nodes.
bool expect_other_hosts_finished = stage_sync.isQuerySentToOtherHosts() || !stage_sync.isErrorSet();
bool all_hosts_finished = stage_sync.finished() && (stage_sync.otherHostsFinished() || !expect_other_hosts_finished);
if (!all_hosts_finished)
{
if (!cleaner.tryCleanupAfterError())
return false;
}
return true;
}
void BackupCoordinationOnCluster::waitForOtherHostsToFinish()
{
if ((current_host != kInitiator) || !backup_query_was_sent_to_other_hosts)
return;
stage_sync.waitForOtherHostsToFinish();
}
bool BackupCoordinationOnCluster::tryWaitForOtherHostsToFinishAfterError() noexcept
{
if (current_host != kInitiator)
auto unfinished_hosts = expect_other_hosts_finished ? stage_sync.getUnfinishedHosts() : Strings{current_host};
LOG_INFO(log, "Skipping removing nodes from ZooKeeper because hosts {} didn't finish",
BackupCoordinationStageSync::getHostsDesc(unfinished_hosts));
return false;
if (!backup_query_was_sent_to_other_hosts)
return true;
return stage_sync.tryWaitForOtherHostsToFinishAfterError();
}
return cleaner.cleanup(throw_if_error);
}
ZooKeeperRetriesInfo BackupCoordinationOnCluster::getOnClusterInitializationKeeperRetriesInfo() const

20
src/Backups/BackupCoordinationOnCluster.h
View File

@ -1,7 +1,6 @@
#pragma once
#include <Backups/IBackupCoordination.h>
#include <Backups/BackupConcurrencyCheck.h>
#include <Backups/BackupCoordinationCleaner.h>
#include <Backups/BackupCoordinationFileInfos.h>
#include <Backups/BackupCoordinationReplicatedAccess.h>
@ -20,7 +19,7 @@ class BackupCoordinationOnCluster : public IBackupCoordination
{
public:
/// Empty string as the current host is used to mark the initiator of a BACKUP ON CLUSTER query.
static const constexpr std::string_view kInitiator;
static const constexpr std::string_view kInitiator = BackupCoordinationStageSync::kInitiator;
BackupCoordinationOnCluster(
const UUID & backup_uuid_,
@ -37,13 +36,13 @@ public:
~BackupCoordinationOnCluster() override;
void setBackupQueryIsSentToOtherHosts() override;
bool isBackupQuerySentToOtherHosts() const override;
Strings setStage(const String & new_stage, const String & message, bool sync) override;
void setBackupQueryWasSentToOtherHosts() override;
bool trySetError(std::exception_ptr exception) override;
void finish() override;
bool tryFinishAfterError() noexcept override;
void waitForOtherHostsToFinish() override;
bool tryWaitForOtherHostsToFinishAfterError() noexcept override;
bool setError(std::exception_ptr exception, bool throw_if_error) override;
bool waitOtherHostsFinish(bool throw_if_error) const override;
bool finish(bool throw_if_error) override;
bool cleanup(bool throw_if_error) override;
void addReplicatedPartNames(
const String & table_zk_path,
@ -110,11 +109,10 @@ private:
const bool plain_backup;
LoggerPtr const log;
/// The order is important: `stage_sync` must be initialized after `with_retries` and `cleaner`.
const WithRetries with_retries;
BackupConcurrencyCheck concurrency_check;
BackupCoordinationStageSync stage_sync;
BackupCoordinationCleaner cleaner;
std::atomic<bool> backup_query_was_sent_to_other_hosts = false;
BackupCoordinationStageSync stage_sync;
mutable std::optional<BackupCoordinationReplicatedTables> replicated_tables TSA_GUARDED_BY(replicated_tables_mutex);
mutable std::optional<BackupCoordinationReplicatedAccess> replicated_access TSA_GUARDED_BY(replicated_access_mutex);

895
src/Backups/BackupCoordinationStageSync.cpp
View File

File diff suppressed because it is too large Load Diff

97
src/Backups/BackupCoordinationStageSync.h
View File

@ -1,7 +1,9 @@
#pragma once
#include <Backups/BackupConcurrencyCheck.h>
#include <Backups/WithRetries.h>
namespace DB
{
@ -9,12 +11,16 @@ namespace DB
class BackupCoordinationStageSync
{
public:
/// Empty string as the current host is used to mark the initiator of a BACKUP ON CLUSTER or RESTORE ON CLUSTER query.
static const constexpr std::string_view kInitiator;
BackupCoordinationStageSync(
bool is_restore_, /// true if this is a RESTORE ON CLUSTER command, false if this is a BACKUP ON CLUSTER command
const String & zookeeper_path_, /// path to the "stage" folder in ZooKeeper
const String & current_host_, /// the current host, or an empty string if it's the initiator of the BACKUP/RESTORE ON CLUSTER command
const Strings & all_hosts_, /// all the hosts (including the initiator and the current host) performing the BACKUP/RESTORE ON CLUSTER command
bool allow_concurrency_, /// whether it's allowed to have concurrent backups or restores.
BackupConcurrencyCounters & concurrency_counters_,
const WithRetries & with_retries_,
ThreadPoolCallbackRunnerUnsafe<void> schedule_,
QueryStatusPtr process_list_element_,
@ -22,30 +28,37 @@ public:
~BackupCoordinationStageSync();
/// Sets that the BACKUP or RESTORE query was sent to other hosts.
void setQueryIsSentToOtherHosts();
bool isQuerySentToOtherHosts() const;
/// Sets the stage of the current host and signal other hosts if there were other hosts waiting for that.
void setStage(const String & stage, const String & stage_result = {});
/// Waits until all the specified hosts come to the specified stage.
/// The function returns the results which specified hosts set when they came to the required stage.
/// If it doesn't happen before the timeout then the function will stop waiting and throw an exception.
Strings waitForHostsToReachStage(const String & stage_to_wait, const Strings & hosts, std::optional<std::chrono::milliseconds> timeout = {}) const;
/// Waits until all the other hosts finish their work.
/// Stops waiting and throws an exception if another host encounters an error or if some host gets cancelled.
void waitForOtherHostsToFinish() const;
/// Lets other host know that the current host has finished its work.
void finish(bool & other_hosts_also_finished);
/// Waits until specified hosts come to the specified stage.
/// The function returns the results which the specified hosts set when they came to the required stage.
Strings waitHostsReachStage(const Strings & hosts, const String & stage_to_wait) const;
/// Lets other hosts know that the current host has encountered an error.
bool trySetError(std::exception_ptr exception) noexcept;
/// The function returns true if it successfully created the error node or if the error node was found already exist.
bool setError(std::exception_ptr exception, bool throw_if_error);
bool isErrorSet() const;
/// Waits until all the other hosts finish their work (as a part of error-handling process).
/// Doesn't stops waiting if some host encounters an error or gets cancelled.
bool tryWaitForOtherHostsToFinishAfterError() const noexcept;
/// Waits until the hosts other than the current host finish their work. Must be called before finish().
/// Stops waiting and throws an exception if another host encounters an error or if some host gets cancelled.
bool waitOtherHostsFinish(bool throw_if_error) const;
bool otherHostsFinished() const;
/// Lets other host know that the current host has finished its work (as a part of error-handling process).
bool tryFinishAfterError(bool & other_hosts_also_finished) noexcept;
/// Lets other hosts know that the current host has finished its work.
bool finish(bool throw_if_error);
bool finished() const;
/// Returns true if all the hosts have finished.
bool allHostsFinished() const { return finished() && otherHostsFinished(); }
/// Returns a list of the hosts which haven't finished yet.
Strings getUnfinishedHosts() const;
Strings getUnfinishedOtherHosts() const;
/// Returns a printable name of a specific host. For empty host the function returns "initiator".
static String getHostDesc(const String & host);
@ -78,14 +91,17 @@ private:
/// Reads the current state from ZooKeeper without throwing exceptions.
void readCurrentState(Coordination::ZooKeeperWithFaultInjection::Ptr zookeeper);
/// Creates a stage node to let other hosts know we've reached the specified stage.
void createStageNode(const String & stage, const String & stage_result, Coordination::ZooKeeperWithFaultInjection::Ptr zookeeper);
String getStageNodePath(const String & stage) const;
/// Lets other hosts know that the current host has encountered an error.
bool trySetError(const Exception & exception);
void setError(const Exception & exception);
bool setError(const Exception & exception, bool throw_if_error);
void createErrorNode(const Exception & exception, Coordination::ZooKeeperWithFaultInjection::Ptr zookeeper);
/// Deserializes an error stored in the error node.
static std::pair<std::exception_ptr, String> parseErrorNode(const String & error_node_contents);
std::pair<std::exception_ptr, String> parseErrorNode(const String & error_node_contents) const;
/// Reset the `connected` flag for each host.
void resetConnectedFlag();
@ -102,19 +118,27 @@ private:
void cancelQueryIfDisconnectedTooLong();
/// Used by waitForHostsToReachStage() to check if everything is ready to return.
bool checkIfHostsReachStage(const Strings & hosts, const String & stage_to_wait, bool time_is_out, std::optional<std::chrono::milliseconds> timeout, Strings & results) const TSA_REQUIRES(mutex);
bool checkIfHostsReachStage(const Strings & hosts, const String & stage_to_wait, Strings & results) const TSA_REQUIRES(mutex);
/// Creates the 'finish' node.
bool tryFinishImpl();
bool tryFinishImpl(bool & other_hosts_also_finished, bool throw_if_error, WithRetries::Kind retries_kind);
void createFinishNodeAndRemoveAliveNode(Coordination::ZooKeeperWithFaultInjection::Ptr zookeeper);
bool finishImpl(bool throw_if_error, WithRetries::Kind retries_kind);
void createFinishNodeAndRemoveAliveNode(Coordination::ZooKeeperWithFaultInjection::Ptr zookeeper, bool throw_if_error);
/// Returns the version used by the initiator.
int getInitiatorVersion() const;
/// Waits until all the other hosts finish their work.
bool tryWaitForOtherHostsToFinishImpl(const String & reason, bool throw_if_error, std::optional<std::chrono::seconds> timeout) const;
bool checkIfOtherHostsFinish(const String & reason, bool throw_if_error, bool time_is_out, std::optional<std::chrono::milliseconds> timeout) const TSA_REQUIRES(mutex);
bool waitOtherHostsFinishImpl(const String & reason, std::optional<std::chrono::seconds> timeout, bool throw_if_error) const;
bool checkIfOtherHostsFinish(const String & reason, std::optional<std::chrono::milliseconds> timeout, bool time_is_out, bool & result, bool throw_if_error) const TSA_REQUIRES(mutex);
/// Returns true if all the hosts have finished.
bool allHostsFinishedNoLock() const TSA_REQUIRES(mutex);
bool finishedNoLock() const TSA_REQUIRES(mutex);
bool otherHostsFinishedNoLock() const TSA_REQUIRES(mutex);
/// Returns a list of the hosts which haven't finished yet.
Strings getUnfinishedHostsNoLock() const TSA_REQUIRES(mutex);
Strings getUnfinishedOtherHostsNoLock() const TSA_REQUIRES(mutex);
const bool is_restore;
const String operation_name;
@ -138,15 +162,16 @@ private:
/// Paths in ZooKeeper.
const std::filesystem::path zookeeper_path;
const String root_zookeeper_path;
const String operation_node_path;
const String operation_zookeeper_path;
const String operation_node_name;
const String stage_node_path;
const String start_node_path;
const String finish_node_path;
const String num_hosts_node_path;
const String error_node_path;
const String alive_node_path;
const String alive_tracker_node_path;
const String error_node_path;
std::optional<BackupConcurrencyCheck> concurrency_check;
std::shared_ptr<Poco::Event> zk_nodes_changed;
@ -176,25 +201,21 @@ private:
{
std::map<String /* host */, HostInfo> hosts; /// std::map because we need to compare states
std::optional<String> host_with_error;
bool cancelled = false;
bool operator ==(const State & other) const;
bool operator !=(const State & other) const;
void merge(const State & other);
};
State state TSA_GUARDED_BY(mutex);
mutable std::condition_variable state_changed;
std::future<void> watching_thread_future;
std::atomic<bool> should_stop_watching_thread = false;
bool should_stop_watching_thread TSA_GUARDED_BY(mutex) = false;
struct FinishResult
{
bool succeeded = false;
std::exception_ptr exception;
bool other_hosts_also_finished = false;
};
FinishResult finish_result TSA_GUARDED_BY(mutex);
bool query_is_sent_to_other_hosts TSA_GUARDED_BY(mutex) = false;
bool tried_to_finish TSA_GUARDED_BY(mutex) = false;
bool tried_to_set_error TSA_GUARDED_BY(mutex) = false;
mutable std::mutex mutex;
};

108
src/Backups/BackupsWorker.cpp
View File

@ -329,6 +329,7 @@ std::pair<OperationID, BackupStatus> BackupsWorker::start(const ASTPtr & backup_
struct BackupsWorker::BackupStarter
{
BackupsWorker & backups_worker;
LoggerPtr log;
std::shared_ptr<ASTBackupQuery> backup_query;
ContextPtr query_context; /// We have to keep `query_context` until the end of the operation because a pointer to it is stored inside the ThreadGroup we're using.
ContextMutablePtr backup_context;
@ -345,6 +346,7 @@ struct BackupsWorker::BackupStarter
BackupStarter(BackupsWorker & backups_worker_, const ASTPtr & query_, const ContextPtr & context_)
: backups_worker(backups_worker_)
, log(backups_worker.log)
, backup_query(std::static_pointer_cast<ASTBackupQuery>(query_->clone()))
, query_context(context_)
, backup_context(Context::createCopy(query_context))
@ -399,9 +401,20 @@ struct BackupsWorker::BackupStarter
chassert(!backup);
backup = backups_worker.openBackupForWriting(backup_info, backup_settings, backup_coordination, backup_context);
backups_worker.doBackup(
backup, backup_query, backup_id, backup_name_for_logging, backup_settings, backup_coordination, backup_context,
on_cluster, cluster);
backups_worker.doBackup(backup, backup_query, backup_id, backup_settings, backup_coordination, backup_context,
on_cluster, cluster);
backup_coordination->finish(/* throw_if_error = */ true);
backup.reset();
/// The backup coordination is not needed anymore.
if (!is_internal_backup)
backup_coordination->cleanup(/* throw_if_error = */ true);
backup_coordination.reset();
/// NOTE: setStatus is called after setNumFilesAndSize in order to have actual information in a backup log record
LOG_INFO(log, "{} {} was created successfully", (is_internal_backup ? "Internal backup" : "Backup"), backup_name_for_logging);
backups_worker.setStatus(backup_id, BackupStatus::BACKUP_CREATED);
}
void onException()
@ -416,16 +429,29 @@ struct BackupsWorker::BackupStarter
if (backup && !backup->setIsCorrupted())
should_remove_files_in_backup = false;
if (backup_coordination && backup_coordination->trySetError(std::current_exception()))
bool all_hosts_finished = false;
if (backup_coordination && backup_coordination->setError(std::current_exception(), /* throw_if_error = */ false))
{
bool other_hosts_finished = backup_coordination->tryWaitForOtherHostsToFinishAfterError();
bool other_hosts_finished = !is_internal_backup
&& (!backup_coordination->isBackupQuerySentToOtherHosts() || backup_coordination->waitOtherHostsFinish(/* throw_if_error = */ false));
if (should_remove_files_in_backup && other_hosts_finished)
backup->tryRemoveAllFiles();
backup_coordination->tryFinishAfterError();
all_hosts_finished = backup_coordination->finish(/* throw_if_error = */ false) && other_hosts_finished;
}
if (!all_hosts_finished)
should_remove_files_in_backup = false;
if (backup && should_remove_files_in_backup)
backup->tryRemoveAllFiles();
backup.reset();
if (backup_coordination && all_hosts_finished)
backup_coordination->cleanup(/* throw_if_error = */ false);
backup_coordination.reset();
backups_worker.setStatusSafe(backup_id, getBackupStatusFromCurrentException());
}
};
@ -497,7 +523,6 @@ void BackupsWorker::doBackup(
BackupMutablePtr backup,
const std::shared_ptr<ASTBackupQuery> & backup_query,
const OperationID & backup_id,
const String & backup_name_for_logging,
const BackupSettings & backup_settings,
std::shared_ptr<IBackupCoordination> backup_coordination,
ContextMutablePtr context,
@ -521,10 +546,10 @@ void BackupsWorker::doBackup(
backup_settings.copySettingsToQuery(*backup_query);
sendQueryToOtherHosts(*backup_query, cluster, backup_settings.shard_num, backup_settings.replica_num,
context, required_access, backup_coordination->getOnClusterInitializationKeeperRetriesInfo());
backup_coordination->setBackupQueryWasSentToOtherHosts();
backup_coordination->setBackupQueryIsSentToOtherHosts();
/// Wait until all the hosts have written their backup entries.
backup_coordination->waitForOtherHostsToFinish();
backup_coordination->waitOtherHostsFinish(/* throw_if_error = */ true);
}
else
{
@ -569,18 +594,8 @@ void BackupsWorker::doBackup(
compressed_size = backup->getCompressedSize();
}
/// Close the backup.
backup.reset();
/// The backup coordination is not needed anymore.
backup_coordination->finish();
/// NOTE: we need to update metadata again after backup->finalizeWriting(), because backup metadata is written there.
setNumFilesAndSize(backup_id, num_files, total_size, num_entries, uncompressed_size, compressed_size, 0, 0);
/// NOTE: setStatus is called after setNumFilesAndSize in order to have actual information in a backup log record
LOG_INFO(log, "{} {} was created successfully", (is_internal_backup ? "Internal backup" : "Backup"), backup_name_for_logging);
setStatus(backup_id, BackupStatus::BACKUP_CREATED);
}
@ -687,6 +702,7 @@ void BackupsWorker::writeBackupEntries(
struct BackupsWorker::RestoreStarter
{
BackupsWorker & backups_worker;
LoggerPtr log;
std::shared_ptr<ASTBackupQuery> restore_query;
ContextPtr query_context; /// We have to keep `query_context` until the end of the operation because a pointer to it is stored inside the ThreadGroup we're using.
ContextMutablePtr restore_context;
@ -702,6 +718,7 @@ struct BackupsWorker::RestoreStarter
RestoreStarter(BackupsWorker & backups_worker_, const ASTPtr & query_, const ContextPtr & context_)
: backups_worker(backups_worker_)
, log(backups_worker.log)
, restore_query(std::static_pointer_cast<ASTBackupQuery>(query_->clone()))
, query_context(context_)
, restore_context(Context::createCopy(query_context))
@ -753,16 +770,17 @@ struct BackupsWorker::RestoreStarter
}
restore_coordination = backups_worker.makeRestoreCoordination(on_cluster, restore_settings, restore_context);
backups_worker.doRestore(
restore_query,
restore_id,
backup_name_for_logging,
backup_info,
restore_settings,
restore_coordination,
restore_context,
on_cluster,
cluster);
backups_worker.doRestore(restore_query, restore_id, backup_info, restore_settings, restore_coordination, restore_context,
on_cluster, cluster);
/// The restore coordination is not needed anymore.
restore_coordination->finish(/* throw_if_error = */ true);
if (!is_internal_restore)
restore_coordination->cleanup(/* throw_if_error = */ true);
restore_coordination.reset();
LOG_INFO(log, "Restored from {} {} successfully", (is_internal_restore ? "internal backup" : "backup"), backup_name_for_logging);
backups_worker.setStatus(restore_id, BackupStatus::RESTORED);
}
void onException()
@ -770,12 +788,16 @@ struct BackupsWorker::RestoreStarter
/// Something bad happened, some data were not restored.
tryLogCurrentException(backups_worker.log, fmt::format("Failed to restore from {} {}", (is_internal_restore ? "internal backup" : "backup"), backup_name_for_logging));
if (restore_coordination && restore_coordination->trySetError(std::current_exception()))
if (restore_coordination && restore_coordination->setError(std::current_exception(), /* throw_if_error = */ false))
{
restore_coordination->tryWaitForOtherHostsToFinishAfterError();
restore_coordination->tryFinishAfterError();
bool other_hosts_finished = !is_internal_restore
&& (!restore_coordination->isRestoreQuerySentToOtherHosts() || restore_coordination->waitOtherHostsFinish(/* throw_if_error = */ false));
if (restore_coordination->finish(/* throw_if_error = */ false) && other_hosts_finished)
restore_coordination->cleanup(/* throw_if_error = */ false);
}
restore_coordination.reset();
backups_worker.setStatusSafe(restore_id, getRestoreStatusFromCurrentException());
}
};
@ -838,7 +860,6 @@ BackupPtr BackupsWorker::openBackupForReading(const BackupInfo & backup_info, co
void BackupsWorker::doRestore(
const std::shared_ptr<ASTBackupQuery> & restore_query,
const OperationID & restore_id,
const String & backup_name_for_logging,
const BackupInfo & backup_info,
RestoreSettings restore_settings,
std::shared_ptr<IRestoreCoordination> restore_coordination,
@ -882,10 +903,10 @@ void BackupsWorker::doRestore(
restore_settings.copySettingsToQuery(*restore_query);
sendQueryToOtherHosts(*restore_query, cluster, restore_settings.shard_num, restore_settings.replica_num,
context, {}, restore_coordination->getOnClusterInitializationKeeperRetriesInfo());
restore_coordination->setRestoreQueryWasSentToOtherHosts();
restore_coordination->setRestoreQueryIsSentToOtherHosts();
/// Wait until all the hosts have done with their restoring work.
restore_coordination->waitForOtherHostsToFinish();
restore_coordination->waitOtherHostsFinish(/* throw_if_error = */ true);
}
else
{
@ -905,12 +926,6 @@ void BackupsWorker::doRestore(
backup, context, getThreadPool(ThreadPoolId::RESTORE), after_task_callback};
restorer.run(RestorerFromBackup::RESTORE);
}
/// The restore coordination is not needed anymore.
restore_coordination->finish();
LOG_INFO(log, "Restored from {} {} successfully", (is_internal_restore ? "internal backup" : "backup"), backup_name_for_logging);
setStatus(restore_id, BackupStatus::RESTORED);
}
@ -943,7 +958,7 @@ BackupsWorker::makeBackupCoordination(bool on_cluster, const BackupSettings & ba
if (!on_cluster)
{
return std::make_shared<BackupCoordinationLocal>(
*backup_settings.backup_uuid, !backup_settings.deduplicate_files, allow_concurrent_backups, *concurrency_counters);
!backup_settings.deduplicate_files, allow_concurrent_backups, *concurrency_counters);
}
bool is_internal_backup = backup_settings.internal;
@ -981,8 +996,7 @@ BackupsWorker::makeRestoreCoordination(bool on_cluster, const RestoreSettings &
{
if (!on_cluster)
{
return std::make_shared<RestoreCoordinationLocal>(
*restore_settings.restore_uuid, allow_concurrent_restores, *concurrency_counters);
return std::make_shared<RestoreCoordinationLocal>(allow_concurrent_restores, *concurrency_counters);
}
bool is_internal_restore = restore_settings.internal;

2
src/Backups/BackupsWorker.h
View File

@ -81,7 +81,6 @@ private:
BackupMutablePtr backup,
const std::shared_ptr<ASTBackupQuery> & backup_query,
const BackupOperationID & backup_id,
const String & backup_name_for_logging,
const BackupSettings & backup_settings,
std::shared_ptr<IBackupCoordination> backup_coordination,
ContextMutablePtr context,
@ -102,7 +101,6 @@ private:
void doRestore(
const std::shared_ptr<ASTBackupQuery> & restore_query,
const BackupOperationID & restore_id,
const String & backup_name_for_logging,
const BackupInfo & backup_info,
RestoreSettings restore_settings,
std::shared_ptr<IRestoreCoordination> restore_coordination,

28
src/Backups/IBackupCoordination.h
View File

@ -20,29 +20,27 @@ class IBackupCoordination
public:
virtual ~IBackupCoordination() = default;
/// Sets that the backup query was sent to other hosts.
/// Function waitOtherHostsFinish() will check that to find out if it should really wait or not.
virtual void setBackupQueryIsSentToOtherHosts() = 0;
virtual bool isBackupQuerySentToOtherHosts() const = 0;
/// Sets the current stage and waits for other hosts to come to this stage too.
virtual Strings setStage(const String & new_stage, const String & message, bool sync) = 0;
/// Sets that the backup query was sent to other hosts.
/// Function waitForOtherHostsToFinish() will check that to find out if it should really wait or not.
virtual void setBackupQueryWasSentToOtherHosts() = 0;
/// Lets other hosts know that the current host has encountered an error.
virtual bool trySetError(std::exception_ptr exception) = 0;
/// Lets other hosts know that the current host has finished its work.
virtual void finish() = 0;
/// Lets other hosts know that the current host has finished its work (as a part of error-handling process).
virtual bool tryFinishAfterError() noexcept = 0;
/// Returns true if the information is successfully passed so other hosts can read it.
virtual bool setError(std::exception_ptr exception, bool throw_if_error) = 0;
/// Waits until all the other hosts finish their work.
/// Stops waiting and throws an exception if another host encounters an error or if some host gets cancelled.
virtual void waitForOtherHostsToFinish() = 0;
virtual bool waitOtherHostsFinish(bool throw_if_error) const = 0;
/// Waits until all the other hosts finish their work (as a part of error-handling process).
/// Doesn't stops waiting if some host encounters an error or gets cancelled.
virtual bool tryWaitForOtherHostsToFinishAfterError() noexcept = 0;
/// Lets other hosts know that the current host has finished its work.
virtual bool finish(bool throw_if_error) = 0;
/// Removes temporary nodes in ZooKeeper.
virtual bool cleanup(bool throw_if_error) = 0;
struct PartNameAndChecksum
{

28
src/Backups/IRestoreCoordination.h
View File

@ -18,29 +18,27 @@ class IRestoreCoordination
public:
virtual ~IRestoreCoordination() = default;
/// Sets that the restore query was sent to other hosts.
/// Function waitOtherHostsFinish() will check that to find out if it should really wait or not.
virtual void setRestoreQueryIsSentToOtherHosts() = 0;
virtual bool isRestoreQuerySentToOtherHosts() const = 0;
/// Sets the current stage and waits for other hosts to come to this stage too.
virtual Strings setStage(const String & new_stage, const String & message, bool sync) = 0;
/// Sets that the restore query was sent to other hosts.
/// Function waitForOtherHostsToFinish() will check that to find out if it should really wait or not.
virtual void setRestoreQueryWasSentToOtherHosts() = 0;
/// Lets other hosts know that the current host has encountered an error.
virtual bool trySetError(std::exception_ptr exception) = 0;
/// Lets other hosts know that the current host has finished its work.
virtual void finish() = 0;
/// Lets other hosts know that the current host has finished its work (as a part of error-handling process).
virtual bool tryFinishAfterError() noexcept = 0;
/// Returns true if the information is successfully passed so other hosts can read it.
virtual bool setError(std::exception_ptr exception, bool throw_if_error) = 0;
/// Waits until all the other hosts finish their work.
/// Stops waiting and throws an exception if another host encounters an error or if some host gets cancelled.
virtual void waitForOtherHostsToFinish() = 0;
virtual bool waitOtherHostsFinish(bool throw_if_error) const = 0;
/// Waits until all the other hosts finish their work (as a part of error-handling process).
/// Doesn't stops waiting if some host encounters an error or gets cancelled.
virtual bool tryWaitForOtherHostsToFinishAfterError() noexcept = 0;
/// Lets other hosts know that the current host has finished its work.
virtual bool finish(bool throw_if_error) = 0;
/// Removes temporary nodes in ZooKeeper.
virtual bool cleanup(bool throw_if_error) = 0;
/// Starts creating a table in a replicated database. Returns false if there is another host which is already creating this table.
virtual bool acquireCreatingTableInReplicatedDatabase(const String & database_zk_path, const String & table_name) = 0;

4
src/Backups/RestoreCoordinationLocal.cpp
View File

@ -10,9 +10,9 @@ namespace DB
{
RestoreCoordinationLocal::RestoreCoordinationLocal(
const UUID & restore_uuid, bool allow_concurrent_restore_, BackupConcurrencyCounters & concurrency_counters_)
bool allow_concurrent_restore_, BackupConcurrencyCounters & concurrency_counters_)
: log(getLogger("RestoreCoordinationLocal"))
, concurrency_check(restore_uuid, /* is_restore = */ true, /* on_cluster = */ false, allow_concurrent_restore_, concurrency_counters_)
, concurrency_check(/* is_restore = */ true, /* on_cluster = */ false, /* zookeeper_path = */ "", allow_concurrent_restore_, concurrency_counters_)
{
}

14
src/Backups/RestoreCoordinationLocal.h
View File

@ -17,16 +17,16 @@ class ASTCreateQuery;
class RestoreCoordinationLocal : public IRestoreCoordination
{
public:
RestoreCoordinationLocal(const UUID & restore_uuid_, bool allow_concurrent_restore_, BackupConcurrencyCounters & concurrency_counters_);
RestoreCoordinationLocal(bool allow_concurrent_restore_, BackupConcurrencyCounters & concurrency_counters_);
~RestoreCoordinationLocal() override;
void setRestoreQueryIsSentToOtherHosts() override {}
bool isRestoreQuerySentToOtherHosts() const override { return false; }
Strings setStage(const String &, const String &, bool) override { return {}; }
void setRestoreQueryWasSentToOtherHosts() override {}
bool trySetError(std::exception_ptr) override { return true; }
void finish() override {}
bool tryFinishAfterError() noexcept override { return true; }
void waitForOtherHostsToFinish() override {}
bool tryWaitForOtherHostsToFinishAfterError() noexcept override { return true; }
bool setError(std::exception_ptr, bool) override { return true; }
bool waitOtherHostsFinish(bool) const override { return true; }
bool finish(bool) override { return true; }
bool cleanup(bool) override { return true; }
/// Starts creating a table in a replicated database. Returns false if there is another host which is already creating this table.
bool acquireCreatingTableInReplicatedDatabase(const String & database_zk_path, const String & table_name) override;

95
src/Backups/RestoreCoordinationOnCluster.cpp
View File

@ -35,17 +35,21 @@ RestoreCoordinationOnCluster::RestoreCoordinationOnCluster(
, current_host_index(BackupCoordinationOnCluster::findCurrentHostIndex(current_host, all_hosts))
, log(getLogger("RestoreCoordinationOnCluster"))
, with_retries(log, get_zookeeper_, keeper_settings, process_list_element_, [root_zookeeper_path_](Coordination::ZooKeeperWithFaultInjection::Ptr zk) { zk->sync(root_zookeeper_path_); })
, concurrency_check(restore_uuid_, /* is_restore = */ true, /* on_cluster = */ true, allow_concurrent_restore_, concurrency_counters_)
, stage_sync(/* is_restore = */ true, fs::path{zookeeper_path} / "stage", current_host, all_hosts, allow_concurrent_restore_, with_retries, schedule_, process_list_element_, log)
, cleaner(zookeeper_path, with_retries, log)
, cleaner(/* is_restore = */ true, zookeeper_path, with_retries, log)
, stage_sync(/* is_restore = */ true, fs::path{zookeeper_path} / "stage", current_host, all_hosts, allow_concurrent_restore_, concurrency_counters_, with_retries, schedule_, process_list_element_, log)
{
createRootNodes();
try
{
createRootNodes();
}
catch (...)
{
stage_sync.setError(std::current_exception(), /* throw_if_error = */ false);
throw;
}
}
RestoreCoordinationOnCluster::~RestoreCoordinationOnCluster()
{
tryFinishImpl();
}
RestoreCoordinationOnCluster::~RestoreCoordinationOnCluster() = default;
void RestoreCoordinationOnCluster::createRootNodes()
{
@ -66,69 +70,52 @@ void RestoreCoordinationOnCluster::createRootNodes()
});
}
void RestoreCoordinationOnCluster::setRestoreQueryIsSentToOtherHosts()
{
stage_sync.setQueryIsSentToOtherHosts();
}
bool RestoreCoordinationOnCluster::isRestoreQuerySentToOtherHosts() const
{
return stage_sync.isQuerySentToOtherHosts();
}
Strings RestoreCoordinationOnCluster::setStage(const String & new_stage, const String & message, bool sync)
{
stage_sync.setStage(new_stage, message);
if (!sync)
return {};
return stage_sync.waitForHostsToReachStage(new_stage, all_hosts_without_initiator);
if (sync)
return stage_sync.waitHostsReachStage(all_hosts_without_initiator, new_stage);
return {};
}
void RestoreCoordinationOnCluster::setRestoreQueryWasSentToOtherHosts()
bool RestoreCoordinationOnCluster::setError(std::exception_ptr exception, bool throw_if_error)
{
restore_query_was_sent_to_other_hosts = true;
return stage_sync.setError(exception, throw_if_error);
}
bool RestoreCoordinationOnCluster::trySetError(std::exception_ptr exception)
bool RestoreCoordinationOnCluster::waitOtherHostsFinish(bool throw_if_error) const
{
return stage_sync.trySetError(exception);
return stage_sync.waitOtherHostsFinish(throw_if_error);
}
void RestoreCoordinationOnCluster::finish()
bool RestoreCoordinationOnCluster::finish(bool throw_if_error)
{
bool other_hosts_also_finished = false;
stage_sync.finish(other_hosts_also_finished);
if ((current_host == kInitiator) && (other_hosts_also_finished || !restore_query_was_sent_to_other_hosts))
cleaner.cleanup();
return stage_sync.finish(throw_if_error);
}
bool RestoreCoordinationOnCluster::tryFinishAfterError() noexcept
bool RestoreCoordinationOnCluster::cleanup(bool throw_if_error)
{
return tryFinishImpl();
}
bool RestoreCoordinationOnCluster::tryFinishImpl() noexcept
{
bool other_hosts_also_finished = false;
if (!stage_sync.tryFinishAfterError(other_hosts_also_finished))
return false;
if ((current_host == kInitiator) && (other_hosts_also_finished || !restore_query_was_sent_to_other_hosts))
/// All the hosts must finish before we remove the coordination nodes.
bool expect_other_hosts_finished = stage_sync.isQuerySentToOtherHosts() || !stage_sync.isErrorSet();
bool all_hosts_finished = stage_sync.finished() && (stage_sync.otherHostsFinished() || !expect_other_hosts_finished);
if (!all_hosts_finished)
{
if (!cleaner.tryCleanupAfterError())
return false;
}
return true;
}
void RestoreCoordinationOnCluster::waitForOtherHostsToFinish()
{
if ((current_host != kInitiator) || !restore_query_was_sent_to_other_hosts)
return;
stage_sync.waitForOtherHostsToFinish();
}
bool RestoreCoordinationOnCluster::tryWaitForOtherHostsToFinishAfterError() noexcept
{
if (current_host != kInitiator)
auto unfinished_hosts = expect_other_hosts_finished ? stage_sync.getUnfinishedHosts() : Strings{current_host};
LOG_INFO(log, "Skipping removing nodes from ZooKeeper because hosts {} didn't finish",
BackupCoordinationStageSync::getHostsDesc(unfinished_hosts));
return false;
if (!restore_query_was_sent_to_other_hosts)
return true;
return stage_sync.tryWaitForOtherHostsToFinishAfterError();
}
return cleaner.cleanup(throw_if_error);
}
ZooKeeperRetriesInfo RestoreCoordinationOnCluster::getOnClusterInitializationKeeperRetriesInfo() const

20
src/Backups/RestoreCoordinationOnCluster.h
View File

@ -1,7 +1,6 @@
#pragma once
#include <Backups/IRestoreCoordination.h>
#include <Backups/BackupConcurrencyCheck.h>
#include <Backups/BackupCoordinationCleaner.h>
#include <Backups/BackupCoordinationStageSync.h>
#include <Backups/WithRetries.h>
@ -15,7 +14,7 @@ class RestoreCoordinationOnCluster : public IRestoreCoordination
{
public:
/// Empty string as the current host is used to mark the initiator of a RESTORE ON CLUSTER query.
static const constexpr std::string_view kInitiator;
static const constexpr std::string_view kInitiator = BackupCoordinationStageSync::kInitiator;
RestoreCoordinationOnCluster(
const UUID & restore_uuid_,
@ -31,13 +30,13 @@ public:
~RestoreCoordinationOnCluster() override;
void setRestoreQueryIsSentToOtherHosts() override;
bool isRestoreQuerySentToOtherHosts() const override;
Strings setStage(const String & new_stage, const String & message, bool sync) override;
void setRestoreQueryWasSentToOtherHosts() override;
bool trySetError(std::exception_ptr exception) override;
void finish() override;
bool tryFinishAfterError() noexcept override;
void waitForOtherHostsToFinish() override;
bool tryWaitForOtherHostsToFinishAfterError() noexcept override;
bool setError(std::exception_ptr exception, bool throw_if_error) override;
bool waitOtherHostsFinish(bool throw_if_error) const override;
bool finish(bool throw_if_error) override;
bool cleanup(bool throw_if_error) override;
/// Starts creating a table in a replicated database. Returns false if there is another host which is already creating this table.
bool acquireCreatingTableInReplicatedDatabase(const String & database_zk_path, const String & table_name) override;
@ -78,11 +77,10 @@ private:
const size_t current_host_index;
LoggerPtr const log;
/// The order is important: `stage_sync` must be initialized after `with_retries` and `cleaner`.
const WithRetries with_retries;
BackupConcurrencyCheck concurrency_check;
BackupCoordinationStageSync stage_sync;
BackupCoordinationCleaner cleaner;
std::atomic<bool> restore_query_was_sent_to_other_hosts = false;
BackupCoordinationStageSync stage_sync;
};
}

1
src/Common/CurrentMetrics.cpp
View File

@ -255,6 +255,7 @@
M(PartsActive, "Active data part, used by current and upcoming SELECTs.") \
M(AttachedDatabase, "Active databases.") \
M(AttachedTable, "Active tables.") \
M(AttachedReplicatedTable, "Active replicated tables.") \
M(AttachedView, "Active views.") \
M(AttachedDictionary, "Active dictionaries.") \
M(PartsOutdated, "Not active data part, but could be used by only current SELECTs, could be deleted after SELECTs finishes.") \

19
src/Common/HostResolvePool.cpp
View File

@ -9,6 +9,7 @@
#include <mutex>
#include <algorithm>
#include <Poco/Timespan.h>
namespace ProfileEvents
@ -49,16 +50,18 @@ HostResolver::WeakPtr HostResolver::getWeakFromThis()
}
HostResolver::HostResolver(String host_, Poco::Timespan history_)
: host(std::move(host_))
, history(history_)
, resolve_function([](const String & host_to_resolve) { return DNSResolver::instance().resolveHostAllInOriginOrder(host_to_resolve); })
{
update();
}
: HostResolver(
[](const String & host_to_resolve) { return DNSResolver::instance().resolveHostAllInOriginOrder(host_to_resolve); },
host_,
history_)
{}
HostResolver::HostResolver(
ResolveFunction && resolve_function_, String host_, Poco::Timespan history_)
: host(std::move(host_)), history(history_), resolve_function(std::move(resolve_function_))
: host(std::move(host_))
, history(history_)
, resolve_interval(history_.totalMicroseconds() / 3)
, resolve_function(std::move(resolve_function_))
{
update();
}
@ -203,7 +206,7 @@ bool HostResolver::isUpdateNeeded()
Poco::Timestamp now;
std::lock_guard lock(mutex);
return last_resolve_time + history < now || records.empty();
return last_resolve_time + resolve_interval < now || records.empty();
}
void HostResolver::updateImpl(Poco::Timestamp now, std::vector<Poco::Net::IPAddress> & next_gen)

4
src/Common/HostResolvePool.h
View File

@ -26,7 +26,7 @@
// a) it still occurs in resolve set after `history_` time or b) all other addresses are pessimized as well.
// - resolve schedule
// Addresses are resolved through `DB::DNSResolver::instance()`.
// Usually it does not happen more often than once in `history_` time.
// Usually it does not happen more often than 3 times in `history_` period.
// But also new resolve performed each `setFail()` call.
namespace DB
@ -212,6 +212,7 @@ protected:
const String host;
const Poco::Timespan history;
const Poco::Timespan resolve_interval;
const HostResolverMetrics metrics = getMetrics();
// for tests purpose
@ -245,4 +246,3 @@ private:
};
}

32
src/Common/LockGuard.h
View File

@ -1,23 +1,47 @@
#pragma once
#include <Common/OvercommitTracker.h>
#include <base/defines.h>
#include <Common/Exception.h>
#include <Common/OvercommitTracker.h>
namespace DB
{
namespace ErrorCodes
{
extern const int LOGICAL_ERROR;
};
/** LockGuard provides RAII-style locking mechanism for a mutex.
** It's intended to be used like std::unique_ptr but with TSA annotations
** It's intended to be used like std::unique_lock but with TSA annotations
*/
template <typename Mutex>
class TSA_SCOPED_LOCKABLE LockGuard
{
public:
explicit LockGuard(Mutex & mutex_) TSA_ACQUIRE(mutex_) : mutex(mutex_) { mutex.lock(); }
~LockGuard() TSA_RELEASE() { mutex.unlock(); }
explicit LockGuard(Mutex & mutex_) TSA_ACQUIRE(mutex_) : mutex(mutex_) { lock(); }
~LockGuard() TSA_RELEASE() { if (locked) unlock(); }
void lock() TSA_ACQUIRE()
{
/// Don't allow recursive_mutex for now.
if (locked)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Can't lock twice the same mutex");
mutex.lock();
locked = true;
}
void unlock() TSA_RELEASE()
{
if (!locked)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Can't unlock the mutex without locking it first");
mutex.unlock();
locked = false;
}
private:
Mutex & mutex;
bool locked = false;
};
template <template<typename> typename TLockGuard, typename Mutex>

31
src/Compression/getCompressionCodecForFile.cpp
View File

@ -10,33 +10,50 @@
namespace DB
{
using Checksum = CityHash_v1_0_2::uint128;
CompressionCodecPtr getCompressionCodecForFile(const IDataPartStorage & data_part_storage, const String & relative_path)
CompressionCodecPtr
getCompressionCodecForFile(ReadBuffer & read_buffer, UInt32 & size_compressed, UInt32 & size_decompressed, bool skip_to_next_block)
{
auto read_buffer = data_part_storage.readFile(relative_path, {}, std::nullopt, std::nullopt);
read_buffer->ignore(sizeof(Checksum));
read_buffer.ignore(sizeof(Checksum));
UInt8 header_size = ICompressionCodec::getHeaderSize();
size_t starting_bytes = read_buffer.count();
PODArray<char> compressed_buffer;
compressed_buffer.resize(header_size);
read_buffer->readStrict(compressed_buffer.data(), header_size);
read_buffer.readStrict(compressed_buffer.data(), header_size);
uint8_t method = ICompressionCodec::readMethod(compressed_buffer.data());
size_compressed = unalignedLoad<UInt32>(&compressed_buffer[1]);
size_decompressed = unalignedLoad<UInt32>(&compressed_buffer[5]);
if (method == static_cast<uint8_t>(CompressionMethodByte::Multiple))
{
compressed_buffer.resize(1);
read_buffer->readStrict(compressed_buffer.data(), 1);
read_buffer.readStrict(compressed_buffer.data(), 1);
compressed_buffer.resize(1 + compressed_buffer[0]);
read_buffer->readStrict(compressed_buffer.data() + 1, compressed_buffer[0]);
read_buffer.readStrict(compressed_buffer.data() + 1, compressed_buffer[0]);
auto codecs_bytes = CompressionCodecMultiple::getCodecsBytesFromData(compressed_buffer.data());
Codecs codecs;
for (auto byte : codecs_bytes)
codecs.push_back(CompressionCodecFactory::instance().get(byte));
if (skip_to_next_block)
read_buffer.ignore(size_compressed - (read_buffer.count() - starting_bytes));
return std::make_shared<CompressionCodecMultiple>(codecs);
}
if (skip_to_next_block)
read_buffer.ignore(size_compressed - (read_buffer.count() - starting_bytes));
return CompressionCodecFactory::instance().get(method);
}
CompressionCodecPtr getCompressionCodecForFile(const IDataPartStorage & data_part_storage, const String & relative_path)
{
auto read_buffer = data_part_storage.readFile(relative_path, {}, std::nullopt, std::nullopt);
UInt32 size_compressed;
UInt32 size_decompressed;
return getCompressionCodecForFile(*read_buffer, size_compressed, size_decompressed, false);
}
}

4
src/Compression/getCompressionCodecForFile.h
View File

@ -13,4 +13,8 @@ namespace DB
/// from metadata.
CompressionCodecPtr getCompressionCodecForFile(const IDataPartStorage & data_part_storage, const String & relative_path);
/// Same as above which is used by clickhouse-compressor to print compression statistics of each data block.
CompressionCodecPtr
getCompressionCodecForFile(ReadBuffer & read_buffer, UInt32 & size_compressed, UInt32 & size_decompressed, bool skip_to_next_block);
}

3
src/Core/ServerSettings.cpp
View File

@ -131,6 +131,9 @@ namespace DB
DECLARE(UInt64, max_database_num_to_warn, 1000lu, "If the number of databases is greater than this value, the server will create a warning that will displayed to user.", 0) \
DECLARE(UInt64, max_part_num_to_warn, 100000lu, "If the number of parts is greater than this value, the server will create a warning that will displayed to user.", 0) \
DECLARE(UInt64, max_table_num_to_throw, 0lu, "If number of tables is greater than this value, server will throw an exception. 0 means no limitation. View, remote tables, dictionary, system tables are not counted. Only count table in Atomic/Ordinary/Replicated/Lazy database engine.", 0) \
DECLARE(UInt64, max_replicated_table_num_to_throw, 0lu, "If number of replicated tables is greater than this value, server will throw an exception. 0 means no limitation. Only count table in Atomic/Ordinary/Replicated/Lazy database engine.", 0) \
DECLARE(UInt64, max_dictionary_num_to_throw, 0lu, "If number of dictionaries is greater than this value, server will throw an exception. 0 means no limitation. Only count table in Atomic/Ordinary/Replicated/Lazy database engine.", 0) \
DECLARE(UInt64, max_view_num_to_throw, 0lu, "If number of views is greater than this value, server will throw an exception. 0 means no limitation. Only count table in Atomic/Ordinary/Replicated/Lazy database engine.", 0) \
DECLARE(UInt64, max_database_num_to_throw, 0lu, "If number of databases is greater than this value, server will throw an exception. 0 means no limitation.", 0) \
DECLARE(UInt64, max_authentication_methods_per_user, 100, "The maximum number of authentication methods a user can be created with or altered. Changing this setting does not affect existing users. Zero means unlimited", 0) \
DECLARE(UInt64, concurrent_threads_soft_limit_num, 0, "Sets how many concurrent thread can be allocated before applying CPU pressure. Zero means unlimited.", 0) \

9
src/Core/Settings.cpp
View File

@ -3669,6 +3669,11 @@ Given that, for example, dictionaries, can be out of sync across nodes, mutation
</profiles>
```
)", 0) \
DECLARE(Bool, validate_mutation_query, true, R"(
Validate mutation queries before accepting them. Mutations are executed in the background, and running an invalid query will cause mutations to get stuck, requiring manual intervention.
Only change this setting if you encounter a backward-incompatible bug.
)", 0) \
DECLARE(Seconds, lock_acquire_timeout, DBMS_DEFAULT_LOCK_ACQUIRE_TIMEOUT_SEC, R"(
Defines how many seconds a locking request waits before failing.
@ -4861,9 +4866,9 @@ Allows to record the filesystem caching log for each query
DECLARE(Bool, read_from_filesystem_cache_if_exists_otherwise_bypass_cache, false, R"(
Allow to use the filesystem cache in passive mode - benefit from the existing cache entries, but don't put more entries into the cache. If you set this setting for heavy ad-hoc queries and leave it disabled for short real-time queries, this will allows to avoid cache threshing by too heavy queries and to improve the overall system efficiency.
)", 0) \
DECLARE(Bool, skip_download_if_exceeds_query_cache, true, R"(
DECLARE(Bool, filesystem_cache_skip_download_if_exceeds_per_query_cache_write_limit, true, R"(
Skip download from remote filesystem if exceeds query cache size
)", 0) \
)", 0) ALIAS(skip_download_if_exceeds_query_cache) \
DECLARE(UInt64, filesystem_cache_max_download_size, (128UL * 1024 * 1024 * 1024), R"(
Max remote filesystem cache size that can be downloaded by a single query
)", 0) \

4
src/Core/SettingsChangesHistory.cpp
View File

@ -64,6 +64,7 @@ static std::initializer_list<std::pair<ClickHouseVersion, SettingsChangesHistory
},
{"24.11",
{
{"validate_mutation_query", false, true, "New setting to validate mutation queries by default."},
{"enable_job_stack_trace", false, true, "Enable by default collecting stack traces from job's scheduling."},
{"allow_suspicious_types_in_group_by", true, false, "Don't allow Variant/Dynamic types in GROUP BY by default"},
{"allow_suspicious_types_in_order_by", true, false, "Don't allow Variant/Dynamic types in ORDER BY by default"},
@ -77,6 +78,7 @@ static std::initializer_list<std::pair<ClickHouseVersion, SettingsChangesHistory
{"backup_restore_keeper_max_retries_while_handling_error", 0, 20, "New setting."},
{"backup_restore_finish_timeout_after_error_sec", 0, 180, "New setting."},
{"parallel_replicas_local_plan", false, true, "Use local plan for local replica in a query with parallel replicas"},
{"filesystem_cache_skip_download_if_exceeds_per_query_cache_write_limit", 1, 1, "Rename of setting skip_download_if_exceeds_query_cache_limit"},
{"filesystem_cache_prefer_bigger_buffer_size", true, true, "New setting"},
{"read_in_order_use_virtual_row", false, false, "Use virtual row while reading in order of primary key or its monotonic function fashion. It is useful when searching over multiple parts as only relevant ones are touched."},
}
@ -124,7 +126,7 @@ static std::initializer_list<std::pair<ClickHouseVersion, SettingsChangesHistory
{"allow_experimental_refreshable_materialized_view", false, true, "Not experimental anymore"},
{"max_parts_to_move", 0, 1000, "New setting"},
{"hnsw_candidate_list_size_for_search", 64, 256, "New setting. Previously, the value was optionally specified in CREATE INDEX and 64 by default."},
{"allow_reorder_prewhere_conditions", false, true, "New setting"},
{"allow_reorder_prewhere_conditions", true, true, "New setting"},
{"input_format_parquet_bloom_filter_push_down", false, true, "When reading Parquet files, skip whole row groups based on the WHERE/PREWHERE expressions and bloom filter in the Parquet metadata."},
{"date_time_64_output_format_cut_trailing_zeros_align_to_groups_of_thousands", false, false, "Dynamically trim the trailing zeros of datetime64 values to adjust the output scale to (0, 3, 6), corresponding to 'seconds', 'milliseconds', and 'microseconds'."},
}

6
src/Databases/DatabasesCommon.cpp
View File

@ -382,7 +382,8 @@ StoragePtr DatabaseWithOwnTablesBase::detachTableUnlocked(const String & table_n
if (!table_storage->isSystemStorage() && !DatabaseCatalog::isPredefinedDatabase(database_name))
{
LOG_TEST(log, "Counting detached table {} to database {}", table_name, database_name);
CurrentMetrics::sub(getAttachedCounterForStorage(table_storage));
for (auto metric : getAttachedCountersForStorage(table_storage))
CurrentMetrics::sub(metric);
}
auto table_id = table_storage->getStorageID();
@ -430,7 +431,8 @@ void DatabaseWithOwnTablesBase::attachTableUnlocked(const String & table_name, c
if (!table->isSystemStorage() && !DatabaseCatalog::isPredefinedDatabase(database_name))
{
LOG_TEST(log, "Counting attached table {} to database {}", table_name, database_name);
CurrentMetrics::add(getAttachedCounterForStorage(table));
for (auto metric : getAttachedCountersForStorage(table))
CurrentMetrics::add(metric);
}
}

13
src/Disks/ObjectStorages/AzureBlobStorage/AzureObjectStorage.cpp
View File

@ -277,19 +277,6 @@ void AzureObjectStorage::removeObjectImpl(const StoredObject & object, const Sha
}
}
/// Remove file. Throws exception if file doesn't exists or it's a directory.
void AzureObjectStorage::removeObject(const StoredObject & object)
{
removeObjectImpl(object, client.get(), false);
}
void AzureObjectStorage::removeObjects(const StoredObjects & objects)
{
auto client_ptr = client.get();
for (const auto & object : objects)
removeObjectImpl(object, client_ptr, false);
}
void AzureObjectStorage::removeObjectIfExists(const StoredObject & object)
{
removeObjectImpl(object, client.get(), true);

5
src/Disks/ObjectStorages/AzureBlobStorage/AzureObjectStorage.h
View File

@ -59,11 +59,6 @@ public:
size_t buf_size = DBMS_DEFAULT_BUFFER_SIZE,
const WriteSettings & write_settings = {}) override;
/// Remove file. Throws exception if file doesn't exists or it's a directory.
void removeObject(const StoredObject & object) override;
void removeObjects(const StoredObjects & objects) override;
void removeObjectIfExists(const StoredObject & object) override;
void removeObjectsIfExist(const StoredObjects & objects) override;

14
src/Disks/ObjectStorages/Cached/CachedObjectStorage.cpp
View File

@ -148,20 +148,6 @@ void CachedObjectStorage::removeCacheIfExists(const std::string & path_key_for_c
cache->removeKeyIfExists(getCacheKey(path_key_for_cache), FileCache::getCommonUser().user_id);
}
void CachedObjectStorage::removeObject(const StoredObject & object)
{
removeCacheIfExists(object.remote_path);
object_storage->removeObject(object);
}
void CachedObjectStorage::removeObjects(const StoredObjects & objects)
{
for (const auto & object : objects)
removeCacheIfExists(object.remote_path);
object_storage->removeObjects(objects);
}
void CachedObjectStorage::removeObjectIfExists(const StoredObject & object)
{
removeCacheIfExists(object.remote_path);

4
src/Disks/ObjectStorages/Cached/CachedObjectStorage.h
View File

@ -45,10 +45,6 @@ public:
size_t buf_size = DBMS_DEFAULT_BUFFER_SIZE,
const WriteSettings & write_settings = {}) override;
void removeObject(const StoredObject & object) override;
void removeObjects(const StoredObjects & objects) override;
void removeObjectIfExists(const StoredObject & object) override;
void removeObjectsIfExist(const StoredObjects & objects) override;

6
src/Disks/ObjectStorages/DiskObjectStorageTransaction.cpp
View File

@ -480,8 +480,7 @@ struct WriteFileObjectStorageOperation final : public IDiskObjectStorageOperatio
void undo() override
{
if (object_storage.exists(object))
object_storage.removeObject(object);
object_storage.removeObjectIfExists(object);
}
void finalize() override
@ -543,8 +542,7 @@ struct CopyFileObjectStorageOperation final : public IDiskObjectStorageOperation
void undo() override
{
for (const auto & object : created_objects)
destination_object_storage.removeObject(object);
destination_object_storage.removeObjectsIfExist(created_objects);
}
void finalize() override

10
src/Disks/ObjectStorages/HDFS/HDFSObjectStorage.h
View File

@ -77,11 +77,6 @@ public:
size_t buf_size = DBMS_DEFAULT_BUFFER_SIZE,
const WriteSettings & write_settings = {}) override;
/// Remove file. Throws exception if file doesn't exists or it's a directory.
void removeObject(const StoredObject & object) override;
void removeObjects(const StoredObjects & objects) override;
void removeObjectIfExists(const StoredObject & object) override;
void removeObjectsIfExist(const StoredObjects & objects) override;
@ -117,6 +112,11 @@ private:
void initializeHDFSFS() const;
std::string extractObjectKeyFromURL(const StoredObject & object) const;
/// Remove file. Throws exception if file doesn't exists or it's a directory.
void removeObject(const StoredObject & object);
void removeObjects(const StoredObjects & objects);
const Poco::Util::AbstractConfiguration & config;
mutable HDFSBuilderWrapper hdfs_builder;

4
src/Disks/ObjectStorages/IObjectStorage.h
View File

@ -161,11 +161,11 @@ public:
virtual bool isRemote() const = 0;
/// Remove object. Throws exception if object doesn't exists.
virtual void removeObject(const StoredObject & object) = 0;
// virtual void removeObject(const StoredObject & object) = 0;
/// Remove multiple objects. Some object storages can do batch remove in a more
/// optimal way.
virtual void removeObjects(const StoredObjects & objects) = 0;
// virtual void removeObjects(const StoredObjects & objects) = 0;
/// Remove object on path if exists
virtual void removeObjectIfExists(const StoredObject & object) = 0;

4
src/Disks/ObjectStorages/Local/LocalObjectStorage.cpp
View File

@ -81,7 +81,7 @@ std::unique_ptr<WriteBufferFromFileBase> LocalObjectStorage::writeObject( /// NO
return std::make_unique<WriteBufferFromFile>(object.remote_path, buf_size);
}
void LocalObjectStorage::removeObject(const StoredObject & object)
void LocalObjectStorage::removeObject(const StoredObject & object) const
{
/// For local object storage files are actually removed when "metadata" is removed.
if (!exists(object))
@ -91,7 +91,7 @@ void LocalObjectStorage::removeObject(const StoredObject & object)
ErrnoException::throwFromPath(ErrorCodes::CANNOT_UNLINK, object.remote_path, "Cannot unlink file {}", object.remote_path);
}
void LocalObjectStorage::removeObjects(const StoredObjects & objects)
void LocalObjectStorage::removeObjects(const StoredObjects & objects) const
{
for (const auto & object : objects)
removeObject(object);

8
src/Disks/ObjectStorages/Local/LocalObjectStorage.h
View File

@ -42,10 +42,6 @@ public:
size_t buf_size = DBMS_DEFAULT_BUFFER_SIZE,
const WriteSettings & write_settings = {}) override;
void removeObject(const StoredObject & object) override;
void removeObjects(const StoredObjects & objects) override;
void removeObjectIfExists(const StoredObject & object) override;
void removeObjectsIfExist(const StoredObjects & objects) override;
@ -82,6 +78,10 @@ public:
ReadSettings patchSettings(const ReadSettings & read_settings) const override;
private:
void removeObject(const StoredObject & object) const;
void removeObjects(const StoredObjects & objects) const;
String key_prefix;
LoggerPtr log;
std::string description;

4
src/Disks/ObjectStorages/MetadataStorageFromPlainObjectStorage.cpp
View File

@ -203,7 +203,7 @@ void MetadataStorageFromPlainObjectStorageTransaction::unlinkFile(const std::str
{
auto object_key = metadata_storage.object_storage->generateObjectKeyForPath(path, std::nullopt /* key_prefix */);
auto object = StoredObject(object_key.serialize());
metadata_storage.object_storage->removeObject(object);
metadata_storage.object_storage->removeObjectIfExists(object);
}
void MetadataStorageFromPlainObjectStorageTransaction::removeDirectory(const std::string & path)
@ -211,7 +211,7 @@ void MetadataStorageFromPlainObjectStorageTransaction::removeDirectory(const std
if (metadata_storage.object_storage->isWriteOnce())
{
for (auto it = metadata_storage.iterateDirectory(path); it->isValid(); it->next())
metadata_storage.object_storage->removeObject(StoredObject(it->path()));
metadata_storage.object_storage->removeObjectIfExists(StoredObject(it->path()));
}
else
{

4
src/Disks/ObjectStorages/MetadataStorageFromPlainObjectStorageOperations.cpp
View File

@ -107,7 +107,7 @@ void MetadataStorageFromPlainObjectStorageCreateDirectoryOperation::undo(std::un
auto metric = object_storage->getMetadataStorageMetrics().directory_map_size;
CurrentMetrics::sub(metric, 1);
object_storage->removeObject(StoredObject(metadata_object_key.serialize(), path / PREFIX_PATH_FILE_NAME));
object_storage->removeObjectIfExists(StoredObject(metadata_object_key.serialize(), path / PREFIX_PATH_FILE_NAME));
}
else if (write_created)
object_storage->removeObjectIfExists(StoredObject(metadata_object_key.serialize(), path / PREFIX_PATH_FILE_NAME));
@ -247,7 +247,7 @@ void MetadataStorageFromPlainObjectStorageRemoveDirectoryOperation::execute(std:
auto metadata_object_key = createMetadataObjectKey(key_prefix, metadata_key_prefix);
auto metadata_object = StoredObject(/*remote_path*/ metadata_object_key.serialize(), /*local_path*/ path / PREFIX_PATH_FILE_NAME);
object_storage->removeObject(metadata_object);
object_storage->removeObjectIfExists(metadata_object);
{
std::lock_guard lock(path_map.mutex);

10
src/Disks/ObjectStorages/S3/S3ObjectStorage.cpp
View File

@ -326,21 +326,11 @@ void S3ObjectStorage::removeObjectsImpl(const StoredObjects & objects, bool if_e
ProfileEvents::DiskS3DeleteObjects);
}
void S3ObjectStorage::removeObject(const StoredObject & object)
{
removeObjectImpl(object, false);
}
void S3ObjectStorage::removeObjectIfExists(const StoredObject & object)
{
removeObjectImpl(object, true);
}
void S3ObjectStorage::removeObjects(const StoredObjects & objects)
{
removeObjectsImpl(objects, false);
}
void S3ObjectStorage::removeObjectsIfExist(const StoredObjects & objects)
{
removeObjectsImpl(objects, true);

7
src/Disks/ObjectStorages/S3/S3ObjectStorage.h
View File

@ -101,13 +101,6 @@ public:
ObjectStorageIteratorPtr iterate(const std::string & path_prefix, size_t max_keys) const override;
/// Uses `DeleteObjectRequest`.
void removeObject(const StoredObject & object) override;
/// Uses `DeleteObjectsRequest` if it is allowed by `s3_capabilities`, otherwise `DeleteObjectRequest`.
/// `DeleteObjectsRequest` is not supported on GCS, see https://issuetracker.google.com/issues/162653700 .
void removeObjects(const StoredObjects & objects) override;
/// Uses `DeleteObjectRequest`.
void removeObjectIfExists(const StoredObject & object) override;

10
src/Disks/ObjectStorages/Web/WebObjectStorage.cpp
View File

@ -254,16 +254,6 @@ std::unique_ptr<WriteBufferFromFileBase> WebObjectStorage::writeObject( /// NOLI
throwNotAllowed();
}
void WebObjectStorage::removeObject(const StoredObject &)
{
throwNotAllowed();
}
void WebObjectStorage::removeObjects(const StoredObjects &)
{
throwNotAllowed();
}
void WebObjectStorage::removeObjectIfExists(const StoredObject &)
{
throwNotAllowed();

4
src/Disks/ObjectStorages/Web/WebObjectStorage.h
View File

@ -47,10 +47,6 @@ public:
size_t buf_size = DBMS_DEFAULT_BUFFER_SIZE,
const WriteSettings & write_settings = {}) override;
void removeObject(const StoredObject & object) override;
void removeObjects(const StoredObjects & objects) override;
void removeObjectIfExists(const StoredObject & object) override;
void removeObjectsIfExist(const StoredObjects & objects) override;

3
src/Functions/FunctionsComparison.h
View File

@ -1033,6 +1033,9 @@ private:
size_t tuple_size,
size_t input_rows_count) const
{
if (0 == tuple_size)
throw Exception(ErrorCodes::NOT_IMPLEMENTED, "Comparison of zero-sized tuples is not implemented");
ColumnsWithTypeAndName less_columns(tuple_size);
ColumnsWithTypeAndName equal_columns(tuple_size - 1);
ColumnsWithTypeAndName tmp_columns(2);

3
src/Functions/if.cpp
View File

@ -668,6 +668,9 @@ private:
temporary_columns[0] = arguments[0];
size_t tuple_size = type1.getElements().size();
if (tuple_size == 0)
return ColumnTuple::create(input_rows_count);
Columns tuple_columns(tuple_size);
for (size_t i = 0; i < tuple_size; ++i)

2
src/IO/ReadSettings.h
View File

@ -69,7 +69,7 @@ struct ReadSettings
std::shared_ptr<PageCache> page_cache;
size_t filesystem_cache_max_download_size = (128UL * 1024 * 1024 * 1024);
bool skip_download_if_exceeds_query_cache = true;
bool filesystem_cache_skip_download_if_exceeds_per_query_cache_write_limit = true;
size_t remote_read_min_bytes_for_seek = DBMS_DEFAULT_BUFFER_SIZE;

12
src/IO/S3/URI.cpp
View File

@ -37,7 +37,7 @@ URI::URI(const std::string & uri_, bool allow_archive_path_syntax)
/// Case when bucket name represented in domain name of S3 URL.
/// E.g. (https://bucket-name.s3.region.amazonaws.com/key)
/// https://docs.aws.amazon.com/AmazonS3/latest/dev/VirtualHosting.html#virtual-hosted-style-access
static const RE2 virtual_hosted_style_pattern(R"((.+)\.(s3express[\-a-z0-9]+|s3|cos|obs|oss|eos)([.\-][a-z0-9\-.:]+))");
static const RE2 virtual_hosted_style_pattern(R"((.+)\.(s3express[\-a-z0-9]+|s3|cos|obs|oss-data-acc|oss|eos)([.\-][a-z0-9\-.:]+))");
/// Case when AWS Private Link Interface is being used
/// E.g. (bucket.vpce-07a1cd78f1bd55c5f-j3a3vg6w.s3.us-east-1.vpce.amazonaws.com/bucket-name/key)
@ -115,7 +115,15 @@ URI::URI(const std::string & uri_, bool allow_archive_path_syntax)
&& re2::RE2::FullMatch(uri.getAuthority(), virtual_hosted_style_pattern, &bucket, &name, &endpoint_authority_from_uri))
{
is_virtual_hosted_style = true;
endpoint = uri.getScheme() + "://" + name + endpoint_authority_from_uri;
if (name == "oss-data-acc")
{
bucket = bucket.substr(0, bucket.find('.'));
endpoint = uri.getScheme() + "://" + uri.getHost().substr(bucket.length() + 1);
}
else
{
endpoint = uri.getScheme() + "://" + name + endpoint_authority_from_uri;
}
validateBucket(bucket, uri);
if (!uri.getPath().empty())

16
src/IO/tests/gtest_s3_uri.cpp
View File

@ -212,6 +212,22 @@ TEST(S3UriTest, validPatterns)
ASSERT_EQ("", uri.version_id);
ASSERT_EQ(true, uri.is_virtual_hosted_style);
}
{
S3::URI uri("https://bucket-test1.oss-cn-beijing-internal.aliyuncs.com/ab-test");
ASSERT_EQ("https://oss-cn-beijing-internal.aliyuncs.com", uri.endpoint);
ASSERT_EQ("bucket-test1", uri.bucket);
ASSERT_EQ("ab-test", uri.key);
ASSERT_EQ("", uri.version_id);
ASSERT_EQ(true, uri.is_virtual_hosted_style);
}
{
S3::URI uri("https://bucket-test.cn-beijing-internal.oss-data-acc.aliyuncs.com/ab-test");
ASSERT_EQ("https://cn-beijing-internal.oss-data-acc.aliyuncs.com", uri.endpoint);
ASSERT_EQ("bucket-test", uri.bucket);
ASSERT_EQ("ab-test", uri.key);
ASSERT_EQ("", uri.version_id);
ASSERT_EQ(true, uri.is_virtual_hosted_style);
}
}
TEST(S3UriTest, versionIdChecks)

32
src/Interpreters/Cache/FileCache.cpp
View File

@ -8,6 +8,7 @@
#include <Interpreters/Cache/FileCacheSettings.h>
#include <Interpreters/Cache/LRUFileCachePriority.h>
#include <Interpreters/Cache/SLRUFileCachePriority.h>
#include <Interpreters/Cache/FileCacheUtils.h>
#include <Interpreters/Cache/EvictionCandidates.h>
#include <Interpreters/Context.h>
#include <base/hex.h>
@ -53,16 +54,6 @@ namespace ErrorCodes
namespace
{
size_t roundDownToMultiple(size_t num, size_t multiple)
{
return (num / multiple) * multiple;
}
size_t roundUpToMultiple(size_t num, size_t multiple)
{
return roundDownToMultiple(num + multiple - 1, multiple);
}
std::string getCommonUserID()
{
auto user_from_context = DB::Context::getGlobalContextInstance()->getFilesystemCacheUser();
@ -96,6 +87,7 @@ FileCache::FileCache(const std::string & cache_name, const FileCacheSettings & s
: max_file_segment_size(settings.max_file_segment_size)
, bypass_cache_threshold(settings.enable_bypass_cache_with_threshold ? settings.bypass_cache_threshold : 0)
, boundary_alignment(settings.boundary_alignment)
, background_download_max_file_segment_size(settings.background_download_max_file_segment_size)
, load_metadata_threads(settings.load_metadata_threads)
, load_metadata_asynchronously(settings.load_metadata_asynchronously)
, write_cache_per_user_directory(settings.write_cache_per_user_id_directory)
@ -103,7 +95,10 @@ FileCache::FileCache(const std::string & cache_name, const FileCacheSettings & s
, keep_current_elements_to_max_ratio(1 - settings.keep_free_space_elements_ratio)
, keep_up_free_space_remove_batch(settings.keep_free_space_remove_batch)
, log(getLogger("FileCache(" + cache_name + ")"))
, metadata(settings.base_path, settings.background_download_queue_size_limit, settings.background_download_threads, write_cache_per_user_directory)
, metadata(settings.base_path,
settings.background_download_queue_size_limit,
settings.background_download_threads,
write_cache_per_user_directory)
{
if (settings.cache_policy == "LRU")
{
@ -601,8 +596,8 @@ FileCache::getOrSet(
/// 2. max_file_segments_limit
FileSegment::Range result_range = initial_range;
const auto aligned_offset = roundDownToMultiple(initial_range.left, boundary_alignment);
auto aligned_end_offset = std::min(roundUpToMultiple(initial_range.right + 1, boundary_alignment), file_size) - 1;
const auto aligned_offset = FileCacheUtils::roundDownToMultiple(initial_range.left, boundary_alignment);
auto aligned_end_offset = std::min(FileCacheUtils::roundUpToMultiple(initial_range.right + 1, boundary_alignment), file_size) - 1;
chassert(aligned_offset <= initial_range.left);
chassert(aligned_end_offset >= initial_range.right);
@ -1600,6 +1595,17 @@ void FileCache::applySettingsIfPossible(const FileCacheSettings & new_settings,
}
}
if (new_settings.background_download_max_file_segment_size != actual_settings.background_download_max_file_segment_size)
{
background_download_max_file_segment_size = new_settings.background_download_max_file_segment_size;
LOG_INFO(log, "Changed background_download_max_file_segment_size from {} to {}",
actual_settings.background_download_max_file_segment_size,
new_settings.background_download_max_file_segment_size);
actual_settings.background_download_max_file_segment_size = new_settings.background_download_max_file_segment_size;
}
if (new_settings.max_size != actual_settings.max_size
|| new_settings.max_elements != actual_settings.max_elements)
{

5
src/Interpreters/Cache/FileCache.h
View File

@ -161,6 +161,10 @@ public:
size_t getMaxFileSegmentSize() const { return max_file_segment_size; }
size_t getBackgroundDownloadMaxFileSegmentSize() const { return background_download_max_file_segment_size.load(); }
size_t getBoundaryAlignment() const { return boundary_alignment; }
bool tryReserve(
FileSegment & file_segment,
size_t size,
@ -199,6 +203,7 @@ private:
std::atomic<size_t> max_file_segment_size;
const size_t bypass_cache_threshold;
const size_t boundary_alignment;
std::atomic<size_t> background_download_max_file_segment_size;
size_t load_metadata_threads;
const bool load_metadata_asynchronously;
std::atomic<bool> stop_loading_metadata = false;

3
src/Interpreters/Cache/FileCacheSettings.cpp
View File

@ -62,6 +62,9 @@ void FileCacheSettings::loadImpl(FuncHas has, FuncGetUInt get_uint, FuncGetStrin
if (has("background_download_queue_size_limit"))
background_download_queue_size_limit = get_uint("background_download_queue_size_limit");
if (has("background_download_max_file_segment_size"))
background_download_max_file_segment_size = get_uint("background_download_max_file_segment_size");
if (has("load_metadata_threads"))
load_metadata_threads = get_uint("load_metadata_threads");

2
src/Interpreters/Cache/FileCacheSettings.h
View File

@ -43,6 +43,8 @@ struct FileCacheSettings
double keep_free_space_elements_ratio = FILECACHE_DEFAULT_FREE_SPACE_ELEMENTS_RATIO;
size_t keep_free_space_remove_batch = FILECACHE_DEFAULT_FREE_SPACE_REMOVE_BATCH;
size_t background_download_max_file_segment_size = FILECACHE_DEFAULT_MAX_FILE_SEGMENT_SIZE_WITH_BACKGROUND_DOWLOAD;
void loadFromConfig(const Poco::Util::AbstractConfiguration & config, const std::string & config_prefix);
void loadFromCollection(const NamedCollection & collection);

17
src/Interpreters/Cache/FileCacheUtils.h Normal file
View File

@ -0,0 +1,17 @@
#pragma once
#include <Core/Types.h>
namespace FileCacheUtils
{
static size_t roundDownToMultiple(size_t num, size_t multiple)
{
return (num / multiple) * multiple;
}
static size_t roundUpToMultiple(size_t num, size_t multiple)
{
return roundDownToMultiple(num + multiple - 1, multiple);
}
}

1
src/Interpreters/Cache/FileCache_fwd.h
View File

@ -6,6 +6,7 @@ namespace DB
static constexpr int FILECACHE_DEFAULT_MAX_FILE_SEGMENT_SIZE = 32 * 1024 * 1024; /// 32Mi
static constexpr int FILECACHE_DEFAULT_FILE_SEGMENT_ALIGNMENT = 4 * 1024 * 1024; /// 4Mi
static constexpr int FILECACHE_DEFAULT_MAX_FILE_SEGMENT_SIZE_WITH_BACKGROUND_DOWLOAD = 4 * 1024 * 1024; /// 4Mi
static constexpr int FILECACHE_DEFAULT_BACKGROUND_DOWNLOAD_THREADS = 5;
static constexpr int FILECACHE_DEFAULT_BACKGROUND_DOWNLOAD_QUEUE_SIZE_LIMIT = 5000;
static constexpr int FILECACHE_DEFAULT_LOAD_METADATA_THREADS = 16;

67
src/Interpreters/Cache/FileSegment.cpp
View File

@ -4,6 +4,7 @@
#include <IO/Operators.h>
#include <IO/WriteBufferFromString.h>
#include <Interpreters/Cache/FileCache.h>
#include <Interpreters/Cache/FileCacheUtils.h>
#include <base/getThreadId.h>
#include <base/hex.h>
#include <Common/CurrentThread.h>
@ -360,11 +361,14 @@ void FileSegment::write(char * from, size_t size, size_t offset_in_file)
"Expected DOWNLOADING state, got {}", stateToString(download_state));
const size_t first_non_downloaded_offset = getCurrentWriteOffset();
if (offset_in_file != first_non_downloaded_offset)
{
throw Exception(
ErrorCodes::LOGICAL_ERROR,
"Attempt to write {} bytes to offset: {}, but current write offset is {}",
size, offset_in_file, first_non_downloaded_offset);
}
const size_t current_downloaded_size = getDownloadedSize();
chassert(reserved_size >= current_downloaded_size);
@ -375,8 +379,19 @@ void FileSegment::write(char * from, size_t size, size_t offset_in_file)
ErrorCodes::LOGICAL_ERROR,
"Not enough space is reserved. Available: {}, expected: {}", free_reserved_size, size);
if (!is_unbound && current_downloaded_size == range().size())
throw Exception(ErrorCodes::LOGICAL_ERROR, "File segment is already fully downloaded");
if (!is_unbound)
{
if (current_downloaded_size == range().size())
throw Exception(ErrorCodes::LOGICAL_ERROR, "File segment is already fully downloaded");
if (current_downloaded_size + size > range().size())
{
throw Exception(
ErrorCodes::LOGICAL_ERROR,
"Cannot download beyond file segment boundaries: {}. Write offset: {}, size: {}, downloaded size: {}",
range().size(), first_non_downloaded_offset, size, current_downloaded_size);
}
}
if (!cache_writer && current_downloaded_size > 0)
throw Exception(
@ -629,6 +644,36 @@ void FileSegment::completePartAndResetDownloader()
LOG_TEST(log, "Complete batch. ({})", getInfoForLogUnlocked(lk));
}
size_t FileSegment::getSizeForBackgroundDownload() const
{
auto lk = lock();
return getSizeForBackgroundDownloadUnlocked(lk);
}
size_t FileSegment::getSizeForBackgroundDownloadUnlocked(const FileSegmentGuard::Lock &) const
{
if (!background_download_enabled
|| !downloaded_size
|| !remote_file_reader)
{
return 0;
}
chassert(downloaded_size <= range().size());
const size_t background_download_max_file_segment_size = cache->getBackgroundDownloadMaxFileSegmentSize();
size_t desired_size;
if (downloaded_size >= background_download_max_file_segment_size)
desired_size = FileCacheUtils::roundUpToMultiple(downloaded_size, cache->getBoundaryAlignment());
else
desired_size = FileCacheUtils::roundUpToMultiple(background_download_max_file_segment_size, cache->getBoundaryAlignment());
desired_size = std::min(desired_size, range().size());
chassert(desired_size >= downloaded_size);
return desired_size - downloaded_size;
}
void FileSegment::complete(bool allow_background_download)
{
ProfileEventTimeIncrement<Microseconds> watch(ProfileEvents::FileSegmentCompleteMicroseconds);
@ -708,7 +753,8 @@ void FileSegment::complete(bool allow_background_download)
if (is_last_holder)
{
bool added_to_download_queue = false;
if (allow_background_download && background_download_enabled && remote_file_reader)
size_t background_download_size = allow_background_download ? getSizeForBackgroundDownloadUnlocked(segment_lock) : 0;
if (background_download_size)
{
ProfileEvents::increment(ProfileEvents::FilesystemCacheBackgroundDownloadQueuePush);
added_to_download_queue = locked_key->addToDownloadQueue(offset(), segment_lock); /// Finish download in background.
@ -862,7 +908,12 @@ bool FileSegment::assertCorrectnessUnlocked(const FileSegmentGuard::Lock & lock)
chassert(downloaded_size == reserved_size);
chassert(downloaded_size == range().size());
chassert(downloaded_size > 0);
chassert(fs::file_size(getPath()) > 0);
auto file_size = fs::file_size(getPath());
UNUSED(file_size);
chassert(file_size == range().size());
chassert(downloaded_size == range().size());
chassert(queue_iterator || on_delayed_removal);
check_iterator(queue_iterator);
@ -884,7 +935,13 @@ bool FileSegment::assertCorrectnessUnlocked(const FileSegmentGuard::Lock & lock)
chassert(reserved_size >= downloaded_size);
chassert(downloaded_size > 0);
chassert(fs::file_size(getPath()) > 0);
auto file_size = fs::file_size(getPath());
UNUSED(file_size);
chassert(file_size > 0);
chassert(file_size <= range().size());
chassert(downloaded_size <= range().size());
chassert(queue_iterator);
check_iterator(queue_iterator);

3
src/Interpreters/Cache/FileSegment.h
View File

@ -185,6 +185,8 @@ public:
bool assertCorrectness() const;
size_t getSizeForBackgroundDownload() const;
/**
* ========== Methods that must do cv.notify() ==================
*/
@ -230,6 +232,7 @@ private:
String getDownloaderUnlocked(const FileSegmentGuard::Lock &) const;
bool isDownloaderUnlocked(const FileSegmentGuard::Lock & segment_lock) const;
void resetDownloaderUnlocked(const FileSegmentGuard::Lock &);
size_t getSizeForBackgroundDownloadUnlocked(const FileSegmentGuard::Lock &) const;
void setDownloadState(State state, const FileSegmentGuard::Lock &);
void resetDownloadingStateUnlocked(const FileSegmentGuard::Lock &);

30
src/Interpreters/Cache/Metadata.cpp
View File

@ -676,13 +676,17 @@ void CacheMetadata::downloadImpl(FileSegment & file_segment, std::optional<Memor
log, "Downloading {} bytes for file segment {}",
file_segment.range().size() - file_segment.getDownloadedSize(), file_segment.getInfoForLog());
size_t size_to_download = file_segment.getSizeForBackgroundDownload();
if (!size_to_download)
return;
auto reader = file_segment.getRemoteFileReader();
if (!reader)
{
throw Exception(
ErrorCodes::LOGICAL_ERROR, "No reader. "
"File segment should not have been submitted for background download ({})",
file_segment.getInfoForLog());
LOG_TEST(log, "No reader in {}:{} (state: {}, range: {}, downloaded size: {})",
file_segment.key(), file_segment.offset(), file_segment.state(),
file_segment.range().toString(), file_segment.getDownloadedSize());
return;
}
/// If remote_fs_read_method == 'threadpool',
@ -690,7 +694,7 @@ void CacheMetadata::downloadImpl(FileSegment & file_segment, std::optional<Memor
if (reader->internalBuffer().empty())
{
if (!memory)
memory.emplace(DBMS_DEFAULT_BUFFER_SIZE);
memory.emplace(std::min(size_t(DBMS_DEFAULT_BUFFER_SIZE), size_to_download));
reader->set(memory->data(), memory->size());
}
@ -701,9 +705,13 @@ void CacheMetadata::downloadImpl(FileSegment & file_segment, std::optional<Memor
if (offset != static_cast<size_t>(reader->getPosition()))
reader->seek(offset, SEEK_SET);
while (!reader->eof())
while (size_to_download && !reader->eof())
{
auto size = reader->available();
const auto available = reader->available();
chassert(available);
const auto size = std::min(available, size_to_download);
size_to_download -= size;
std::string failure_reason;
if (!file_segment.reserve(size, reserve_space_lock_wait_timeout_milliseconds, failure_reason))
@ -713,7 +721,7 @@ void CacheMetadata::downloadImpl(FileSegment & file_segment, std::optional<Memor
"for {}:{} (downloaded size: {}/{})",
file_segment.key(), file_segment.offset(),
file_segment.getDownloadedSize(), file_segment.range().size());
return;
break;
}
try
@ -728,12 +736,14 @@ void CacheMetadata::downloadImpl(FileSegment & file_segment, std::optional<Memor
if (code == /* No space left on device */28 || code == /* Quota exceeded */122)
{
LOG_INFO(log, "Insert into cache is skipped due to insufficient disk space. ({})", e.displayText());
return;
break;
}
throw;
}
}
file_segment.resetRemoteFileReader();
LOG_TEST(log, "Downloaded file segment: {}", file_segment.getInfoForLog());
}
@ -1155,7 +1165,7 @@ std::vector<FileSegment::Info> LockedKey::sync()
actual_size, expected_size, file_segment->getInfoForLog());
broken.push_back(FileSegment::getInfo(file_segment));
it = removeFileSegment(file_segment->offset(), file_segment->lock(), /* can_be_broken */false);
it = removeFileSegment(file_segment->offset(), file_segment->lock(), /* can_be_broken */true);
}
return broken;
}

1
src/Interpreters/Cache/Metadata.h
View File

@ -210,6 +210,7 @@ public:
bool setBackgroundDownloadThreads(size_t threads_num);
size_t getBackgroundDownloadThreads() const { return download_threads.size(); }
bool setBackgroundDownloadQueueSizeLimit(size_t size);
bool isBackgroundDownloadEnabled();

2
src/Interpreters/Cache/QueryLimit.cpp
View File

@ -53,7 +53,7 @@ FileCacheQueryLimit::QueryContextPtr FileCacheQueryLimit::getOrSetQueryContext(
{
it->second = std::make_shared<QueryContext>(
settings.filesystem_cache_max_download_size,
!settings.skip_download_if_exceeds_query_cache);
!settings.filesystem_cache_skip_download_if_exceeds_per_query_cache_write_limit);
}
return it->second;

4
src/Interpreters/Context.cpp
View File

@ -237,7 +237,7 @@ namespace Setting
extern const SettingsUInt64 remote_fs_read_backoff_max_tries;
extern const SettingsUInt64 remote_read_min_bytes_for_seek;
extern const SettingsBool throw_on_error_from_cache_on_write_operations;
extern const SettingsBool skip_download_if_exceeds_query_cache;
extern const SettingsBool filesystem_cache_skip_download_if_exceeds_per_query_cache_write_limit;
extern const SettingsBool s3_allow_parallel_part_upload;
extern const SettingsBool use_page_cache_for_disks_without_file_cache;
extern const SettingsUInt64 use_structure_from_insertion_table_in_table_functions;
@ -5755,7 +5755,7 @@ ReadSettings Context::getReadSettings() const
res.filesystem_cache_prefer_bigger_buffer_size = settings_ref[Setting::filesystem_cache_prefer_bigger_buffer_size];
res.filesystem_cache_max_download_size = settings_ref[Setting::filesystem_cache_max_download_size];
res.skip_download_if_exceeds_query_cache = settings_ref[Setting::skip_download_if_exceeds_query_cache];
res.filesystem_cache_skip_download_if_exceeds_per_query_cache_write_limit = settings_ref[Setting::filesystem_cache_skip_download_if_exceeds_per_query_cache_write_limit];
res.page_cache = getPageCache();
res.use_page_cache_for_disks_without_file_cache = settings_ref[Setting::use_page_cache_for_disks_without_file_cache];

42
src/Interpreters/InterpreterCreateQuery.cpp
View File

@ -98,6 +98,9 @@
namespace CurrentMetrics
{
extern const Metric AttachedTable;
extern const Metric AttachedReplicatedTable;
extern const Metric AttachedDictionary;
extern const Metric AttachedView;
}
namespace DB
@ -145,7 +148,10 @@ namespace ServerSetting
{
extern const ServerSettingsBool ignore_empty_sql_security_in_create_view_query;
extern const ServerSettingsUInt64 max_database_num_to_throw;
extern const ServerSettingsUInt64 max_dictionary_num_to_throw;
extern const ServerSettingsUInt64 max_table_num_to_throw;
extern const ServerSettingsUInt64 max_replicated_table_num_to_throw;
extern const ServerSettingsUInt64 max_view_num_to_throw;
}
namespace ErrorCodes
@ -1912,16 +1918,8 @@ bool InterpreterCreateQuery::doCreateTable(ASTCreateQuery & create,
}
}
UInt64 table_num_limit = getContext()->getGlobalContext()->getServerSettings()[ServerSetting::max_table_num_to_throw];
if (table_num_limit > 0 && !internal)
{
UInt64 table_count = CurrentMetrics::get(CurrentMetrics::AttachedTable);
if (table_count >= table_num_limit)
throw Exception(ErrorCodes::TOO_MANY_TABLES,
"Too many tables. "
"The limit (server configuration parameter `max_table_num_to_throw`) is set to {}, the current number of tables is {}",
table_num_limit, table_count);
}
if (!internal)
throwIfTooManyEntities(create, res);
database->createTable(getContext(), create.getTable(), res, query_ptr);
@ -1948,6 +1946,30 @@ bool InterpreterCreateQuery::doCreateTable(ASTCreateQuery & create,
}
void InterpreterCreateQuery::throwIfTooManyEntities(ASTCreateQuery & create, StoragePtr storage) const
{
auto check_and_throw = [&](auto setting, CurrentMetrics::Metric metric, String setting_name, String entity_name)
{
UInt64 num_limit = getContext()->getGlobalContext()->getServerSettings()[setting];
UInt64 attached_count = CurrentMetrics::get(metric);
if (num_limit > 0 && attached_count >= num_limit)
throw Exception(ErrorCodes::TOO_MANY_TABLES,
"Too many {}. "
"The limit (server configuration parameter `{}`) is set to {}, the current number is {}",
entity_name, setting_name, num_limit, attached_count);
};
if (auto * replicated_storage = typeid_cast<StorageReplicatedMergeTree *>(storage.get()))
check_and_throw(ServerSetting::max_replicated_table_num_to_throw, CurrentMetrics::AttachedReplicatedTable, "max_replicated_table_num_to_throw", "replicated tables");
else if (create.is_dictionary)
check_and_throw(ServerSetting::max_dictionary_num_to_throw, CurrentMetrics::AttachedDictionary, "max_dictionary_num_to_throw", "dictionaries");
else if (create.isView())
check_and_throw(ServerSetting::max_view_num_to_throw, CurrentMetrics::AttachedView, "max_view_num_to_throw", "views");
else
check_and_throw(ServerSetting::max_table_num_to_throw, CurrentMetrics::AttachedTable, "max_table_num_to_throw", "tables");
}
BlockIO InterpreterCreateQuery::doCreateOrReplaceTable(ASTCreateQuery & create,
const InterpreterCreateQuery::TableProperties & properties, LoadingStrictnessLevel mode)
{

2
src/Interpreters/InterpreterCreateQuery.h
View File

@ -122,6 +122,8 @@ private:
BlockIO executeQueryOnCluster(ASTCreateQuery & create);
void throwIfTooManyEntities(ASTCreateQuery & create, StoragePtr storage) const;
ASTPtr query_ptr;
/// Skip safety threshold when loading tables.

13
src/Interpreters/MutationsInterpreter.cpp
View File

@ -53,6 +53,7 @@ namespace Setting
extern const SettingsBool allow_nondeterministic_mutations;
extern const SettingsUInt64 max_block_size;
extern const SettingsBool use_concurrency_control;
extern const SettingsBool validate_mutation_query;
}
namespace MergeTreeSetting
@ -1386,6 +1387,18 @@ void MutationsInterpreter::validate()
}
}
// Make sure the mutation query is valid
if (context->getSettingsRef()[Setting::validate_mutation_query])
{
if (context->getSettingsRef()[Setting::allow_experimental_analyzer])
prepareQueryAffectedQueryTree(commands, source.getStorage(), context);
else
{
ASTPtr select_query = prepareQueryAffectedAST(commands, source.getStorage(), context);
InterpreterSelectQuery(select_query, context, source.getStorage(), metadata_snapshot);
}
}
QueryPlan plan;
initQueryPlan(stages.front(), plan);

164
src/Interpreters/QueryMetricLog.cpp
View File

@ -1,6 +1,7 @@
#include <base/getFQDNOrHostName.h>
#include <Common/DateLUT.h>
#include <Common/DateLUTImpl.h>
#include <Common/LockGuard.h>
#include <DataTypes/DataTypeDate.h>
#include <DataTypes/DataTypeDateTime.h>
#include <DataTypes/DataTypeDateTime64.h>
@ -16,7 +17,6 @@
#include <chrono>
#include <fmt/chrono.h>
#include <mutex>
namespace DB
@ -24,6 +24,15 @@ namespace DB
static auto logger = getLogger("QueryMetricLog");
String timePointToString(QueryMetricLog::TimePoint time)
{
/// fmtlib supports subsecond formatting in 10.0.0. We're in 9.1.0, so we need to add the milliseconds ourselves.
auto seconds = std::chrono::time_point_cast<std::chrono::seconds>(time);
auto microseconds = std::chrono::duration_cast<std::chrono::microseconds>(time - seconds).count();
return fmt::format("{:%Y.%m.%d %H:%M:%S}.{:06}", seconds, microseconds);
}
ColumnsDescription QueryMetricLogElement::getColumnsDescription()
{
ColumnsDescription result;
@ -87,36 +96,73 @@ void QueryMetricLog::shutdown()
Base::shutdown();
}
void QueryMetricLog::startQuery(const String & query_id, TimePoint start_time, UInt64 interval_milliseconds)
void QueryMetricLog::collectMetric(const ProcessList & process_list, String query_id)
{
QueryMetricLogStatus status;
status.interval_milliseconds = interval_milliseconds;
status.next_collect_time = start_time + std::chrono::milliseconds(interval_milliseconds);
auto current_time = std::chrono::system_clock::now();
const auto query_info = process_list.getQueryInfo(query_id, false, true, false);
if (!query_info)
{
/// TODO: remove trace before 24.11 release after checking everything is fine on the CI
LOG_TRACE(logger, "Query {} is not running anymore, so we couldn't get its QueryStatusInfo", query_id);
return;
}
LockGuard global_lock(queries_mutex);
auto it = queries.find(query_id);
/// The query might have finished while the scheduled task is running.
if (it == queries.end())
{
global_lock.unlock();
/// TODO: remove trace before 24.11 release after checking everything is fine on the CI
LOG_TRACE(logger, "Query {} not found in the list. Finished while this collecting task was running", query_id);
return;
}
auto & query_status = it->second;
if (!query_status.mutex)
{
global_lock.unlock();
/// TODO: remove trace before 24.11 release after checking everything is fine on the CI
LOG_TRACE(logger, "Query {} finished while this collecting task was running", query_id);
return;
}
LockGuard query_lock(query_status.getMutex());
global_lock.unlock();
auto elem = query_status.createLogMetricElement(query_id, *query_info, current_time);
if (elem)
add(std::move(elem.value()));
}
/// We use TSA_NO_THREAD_SAFETY_ANALYSIS to prevent TSA complaining that we're modifying the query_status fields
/// without locking the mutex. Since we're building it from scratch, there's no harm in not holding it.
/// If we locked it to make TSA happy, TSAN build would falsely complain about
/// lock-order-inversion (potential deadlock)
/// which is not a real issue since QueryMetricLogStatus's mutex cannot be locked by anything else
/// until we add it to the queries map.
void QueryMetricLog::startQuery(const String & query_id, TimePoint start_time, UInt64 interval_milliseconds) TSA_NO_THREAD_SAFETY_ANALYSIS
{
QueryMetricLogStatus query_status;
QueryMetricLogStatusInfo & info = query_status.info;
info.interval_milliseconds = interval_milliseconds;
info.next_collect_time = start_time;
auto context = getContext();
const auto & process_list = context->getProcessList();
status.task = context->getSchedulePool().createTask("QueryMetricLog", [this, &process_list, query_id] {
auto current_time = std::chrono::system_clock::now();
const auto query_info = process_list.getQueryInfo(query_id, false, true, false);
if (!query_info)
{
LOG_TRACE(logger, "Query {} is not running anymore, so we couldn't get its QueryStatusInfo", query_id);
return;
}
auto elem = createLogMetricElement(query_id, *query_info, current_time);
if (elem)
add(std::move(elem.value()));
info.task = context->getSchedulePool().createTask("QueryMetricLog", [this, &process_list, query_id] {
collectMetric(process_list, query_id);
});
std::lock_guard lock(queries_mutex);
status.task->scheduleAfter(interval_milliseconds);
queries.emplace(query_id, std::move(status));
LockGuard global_lock(queries_mutex);
query_status.scheduleNext(query_id);
queries.emplace(query_id, std::move(query_status));
}
void QueryMetricLog::finishQuery(const String & query_id, TimePoint finish_time, QueryStatusInfoPtr query_info)
{
std::unique_lock lock(queries_mutex);
LockGuard global_lock(queries_mutex);
auto it = queries.find(query_id);
/// finishQuery may be called from logExceptionBeforeStart when the query has not even started
@ -124,9 +170,19 @@ void QueryMetricLog::finishQuery(const String & query_id, TimePoint finish_time,
if (it == queries.end())
return;
auto & query_status = it->second;
decltype(query_status.mutex) query_mutex;
LockGuard query_lock(query_status.getMutex());
/// Move the query mutex here so that we hold it until the end, after removing the query from queries.
query_mutex = std::move(query_status.mutex);
query_status.mutex = {};
global_lock.unlock();
if (query_info)
{
auto elem = createLogMetricElement(query_id, *query_info, finish_time, false);
auto elem = query_status.createLogMetricElement(query_id, *query_info, finish_time, false);
if (elem)
add(std::move(elem.value()));
}
@ -139,51 +195,58 @@ void QueryMetricLog::finishQuery(const String & query_id, TimePoint finish_time,
/// that order.
{
/// Take ownership of the task so that we can destroy it in this scope after unlocking `queries_mutex`.
auto task = std::move(it->second.task);
auto task = std::move(query_status.info.task);
/// Build an empty task for the old task to make sure it does not lock any mutex on its destruction.
it->second.task = {};
query_status.info.task = {};
query_lock.unlock();
global_lock.lock();
queries.erase(query_id);
/// Ensure `queries_mutex` is unlocked before calling task's destructor at the end of this
/// scope which will lock `exec_mutex`.
lock.unlock();
global_lock.unlock();
}
}
std::optional<QueryMetricLogElement> QueryMetricLog::createLogMetricElement(const String & query_id, const QueryStatusInfo & query_info, TimePoint query_info_time, bool schedule_next)
void QueryMetricLogStatus::scheduleNext(String query_id)
{
/// fmtlib supports subsecond formatting in 10.0.0. We're in 9.1.0, so we need to add the milliseconds ourselves.
auto seconds = std::chrono::time_point_cast<std::chrono::seconds>(query_info_time);
auto microseconds = std::chrono::duration_cast<std::chrono::microseconds>(query_info_time - seconds).count();
LOG_DEBUG(logger, "Collecting query_metric_log for query {} with QueryStatusInfo from {:%Y.%m.%d %H:%M:%S}.{:06}. Schedule next: {}", query_id, seconds, microseconds, schedule_next);
std::unique_lock lock(queries_mutex);
auto query_status_it = queries.find(query_id);
/// The query might have finished while the scheduled task is running.
if (query_status_it == queries.end())
info.next_collect_time += std::chrono::milliseconds(info.interval_milliseconds);
const auto now = std::chrono::system_clock::now();
if (info.next_collect_time > now)
{
lock.unlock();
LOG_TRACE(logger, "Query {} finished already while this collecting task was running", query_id);
return {};
const auto wait_time = std::chrono::duration_cast<std::chrono::milliseconds>(info.next_collect_time - now).count();
info.task->scheduleAfter(wait_time);
}
auto & query_status = query_status_it->second;
if (query_info_time <= query_status.last_collect_time)
else
{
lock.unlock();
LOG_TRACE(logger, "The next collecting task for query {} should have already run at {}. Scheduling it right now",
query_id, timePointToString(info.next_collect_time));
info.task->schedule();
}
}
std::optional<QueryMetricLogElement> QueryMetricLogStatus::createLogMetricElement(const String & query_id, const QueryStatusInfo & query_info, TimePoint query_info_time, bool schedule_next)
{
/// TODO: remove trace before 24.11 release after checking everything is fine on the CI
LOG_TRACE(logger, "Collecting query_metric_log for query {} and interval {} ms with QueryStatusInfo from {}. Next collection time: {}",
query_id, info.interval_milliseconds, timePointToString(query_info_time),
schedule_next ? timePointToString(info.next_collect_time + std::chrono::milliseconds(info.interval_milliseconds)) : "finished");
if (query_info_time <= info.last_collect_time)
{
/// TODO: remove trace before 24.11 release after checking everything is fine on the CI
LOG_TRACE(logger, "Query {} has a more recent metrics collected. Skipping this one", query_id);
return {};
}
query_status.last_collect_time = query_info_time;
info.last_collect_time = query_info_time;
QueryMetricLogElement elem;
elem.event_time = timeInSeconds(query_info_time);
elem.event_time_microseconds = timeInMicroseconds(query_info_time);
elem.query_id = query_status_it->first;
elem.query_id = query_id;
elem.memory_usage = query_info.memory_usage > 0 ? query_info.memory_usage : 0;
elem.peak_memory_usage = query_info.peak_memory_usage > 0 ? query_info.peak_memory_usage : 0;
@ -192,7 +255,7 @@ std::optional<QueryMetricLogElement> QueryMetricLog::createLogMetricElement(cons
for (ProfileEvents::Event i = ProfileEvents::Event(0), end = ProfileEvents::end(); i < end; ++i)
{
const auto & new_value = (*(query_info.profile_counters))[i];
auto & old_value = query_status.last_profile_events[i];
auto & old_value = info.last_profile_events[i];
/// Profile event counters are supposed to be monotonic. However, at least the `NetworkReceiveBytes` can be inaccurate.
/// So, since in the future the counter should always have a bigger value than in the past, we skip this event.
@ -208,16 +271,13 @@ std::optional<QueryMetricLogElement> QueryMetricLog::createLogMetricElement(cons
}
else
{
LOG_TRACE(logger, "Query {} has no profile counters", query_id);
/// TODO: remove trace before 24.11 release after checking everything is fine on the CI
LOG_DEBUG(logger, "Query {} has no profile counters", query_id);
elem.profile_events = std::vector<ProfileEvents::Count>(ProfileEvents::end());
}
if (schedule_next)
{
query_status.next_collect_time += std::chrono::milliseconds(query_status.interval_milliseconds);
const auto wait_time = std::chrono::duration_cast<std::chrono::milliseconds>(query_status.next_collect_time - std::chrono::system_clock::now()).count();
query_status.task->scheduleAfter(wait_time);
}
scheduleNext(query_id);
return elem;
}

42
src/Interpreters/QueryMetricLog.h
View File

@ -1,5 +1,6 @@
#pragma once
#include <base/defines.h>
#include <Common/ProfileEvents.h>
#include <Common/CurrentMetrics.h>
#include <Core/BackgroundSchedulePool.h>
@ -11,11 +12,17 @@
#include <chrono>
#include <ctime>
#include <mutex>
namespace DB
{
namespace ErrorCodes
{
extern const int LOGICAL_ERROR;
};
/** QueryMetricLogElement is a log of query metric values measured at regular time interval.
*/
@ -34,7 +41,7 @@ struct QueryMetricLogElement
void appendToBlock(MutableColumns & columns) const;
};
struct QueryMetricLogStatus
struct QueryMetricLogStatusInfo
{
UInt64 interval_milliseconds;
std::chrono::system_clock::time_point last_collect_time;
@ -43,24 +50,47 @@ struct QueryMetricLogStatus
BackgroundSchedulePool::TaskHolder task;
};
struct QueryMetricLogStatus
{
using TimePoint = std::chrono::system_clock::time_point;
using Mutex = std::mutex;
QueryMetricLogStatusInfo info TSA_GUARDED_BY(getMutex());
/// We need to be able to move it for the hash map, so we need to add an indirection here.
std::unique_ptr<Mutex> mutex = std::make_unique<Mutex>();
/// Return a reference to the mutex, used for Thread Sanitizer annotations.
Mutex & getMutex() const TSA_RETURN_CAPABILITY(mutex)
{
if (!mutex)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Mutex cannot be NULL");
return *mutex;
}
void scheduleNext(String query_id) TSA_REQUIRES(getMutex());
std::optional<QueryMetricLogElement> createLogMetricElement(const String & query_id, const QueryStatusInfo & query_info, TimePoint query_info_time, bool schedule_next = true) TSA_REQUIRES(getMutex());
};
class QueryMetricLog : public SystemLog<QueryMetricLogElement>
{
using SystemLog<QueryMetricLogElement>::SystemLog;
using TimePoint = std::chrono::system_clock::time_point;
using Base = SystemLog<QueryMetricLogElement>;
public:
using TimePoint = std::chrono::system_clock::time_point;
void shutdown() final;
// Both startQuery and finishQuery are called from the thread that executes the query
/// Both startQuery and finishQuery are called from the thread that executes the query.
void startQuery(const String & query_id, TimePoint start_time, UInt64 interval_milliseconds);
void finishQuery(const String & query_id, TimePoint finish_time, QueryStatusInfoPtr query_info = nullptr);
private:
std::optional<QueryMetricLogElement> createLogMetricElement(const String & query_id, const QueryStatusInfo & query_info, TimePoint query_info_time, bool schedule_next = true);
void collectMetric(const ProcessList & process_list, String query_id);
std::recursive_mutex queries_mutex;
std::unordered_map<String, QueryMetricLogStatus> queries;
std::mutex queries_mutex;
std::unordered_map<String, QueryMetricLogStatus> queries TSA_GUARDED_BY(queries_mutex);
};
}

8
src/Interpreters/executeQuery.cpp
View File

@ -505,6 +505,7 @@ void logQueryFinish(
auto time_now = std::chrono::system_clock::now();
QueryStatusInfo info = process_list_elem->getInfo(true, settings[Setting::log_profile_events]);
logQueryMetricLogFinish(context, internal, elem.client_info.current_query_id, time_now, std::make_shared<QueryStatusInfo>(info));
elem.type = QueryLogElementType::QUERY_FINISH;
addStatusInfoToQueryLogElement(elem, info, query_ast, context);
@ -623,6 +624,7 @@ void logQueryException(
{
elem.query_duration_ms = start_watch.elapsedMilliseconds();
}
logQueryMetricLogFinish(context, internal, elem.client_info.current_query_id, time_now, info);
elem.query_cache_usage = QueryCache::Usage::None;
@ -652,8 +654,6 @@ void logQueryException(
query_span->addAttribute("clickhouse.exception_code", elem.exception_code);
query_span->finish();
}
logQueryMetricLogFinish(context, internal, elem.client_info.current_query_id, time_now, info);
}
void logExceptionBeforeStart(
@ -707,6 +707,8 @@ void logExceptionBeforeStart(
elem.client_info = context->getClientInfo();
logQueryMetricLogFinish(context, false, elem.client_info.current_query_id, std::chrono::system_clock::now(), nullptr);
elem.log_comment = settings[Setting::log_comment];
if (elem.log_comment.size() > settings[Setting::max_query_size])
elem.log_comment.resize(settings[Setting::max_query_size]);
@ -751,8 +753,6 @@ void logExceptionBeforeStart(
ProfileEvents::increment(ProfileEvents::FailedInsertQuery);
}
}
logQueryMetricLogFinish(context, false, elem.client_info.current_query_id, std::chrono::system_clock::now(), nullptr);
}
void validateAnalyzerSettings(ASTPtr ast, bool context_value)

13
src/Processors/QueryPlan/BuildQueryPipelineSettings.cpp
View File

@ -6,12 +6,23 @@
namespace DB
{
namespace Setting
{
extern const SettingsBool query_plan_merge_filters;
}
BuildQueryPipelineSettings BuildQueryPipelineSettings::fromContext(ContextPtr from)
{
const auto & query_settings = from->getSettingsRef();
BuildQueryPipelineSettings settings;
settings.actions_settings = ExpressionActionsSettings::fromSettings(from->getSettingsRef(), CompileExpressions::yes);
settings.actions_settings = ExpressionActionsSettings::fromSettings(query_settings, CompileExpressions::yes);
settings.process_list_element = from->getProcessListElement();
settings.progress_callback = from->getProgressCallback();
/// Setting query_plan_merge_filters is enabled by default.
/// But it can brake short-circuit without splitting filter step into smaller steps.
/// So, enable and disable this optimizations together.
settings.enable_multiple_filters_transforms_for_and_chain = query_settings[Setting::query_plan_merge_filters];
return settings;
}

2
src/Processors/QueryPlan/BuildQueryPipelineSettings.h
View File

@ -17,6 +17,8 @@ using TemporaryFileLookupPtr = std::shared_ptr<ITemporaryFileLookup>;
struct BuildQueryPipelineSettings
{
bool enable_multiple_filters_transforms_for_and_chain = true;
ExpressionActionsSettings actions_settings;
QueryStatusPtr process_list_element;
ProgressCallback progress_callback = nullptr;

137
src/Processors/QueryPlan/FilterStep.cpp
View File

@ -5,6 +5,11 @@
#include <Interpreters/ExpressionActions.h>
#include <IO/Operators.h>
#include <Common/JSONBuilder.h>
#include <DataTypes/DataTypeLowCardinality.h>
#include <DataTypes/DataTypesNumber.h>
#include <Functions/IFunction.h>
#include <stack>
#include <ranges>
namespace DB
{
@ -24,6 +29,92 @@ static ITransformingStep::Traits getTraits()
};
}
static bool isTrivialSubtree(const ActionsDAG::Node * node)
{
while (node->type == ActionsDAG::ActionType::ALIAS)
node = node->children.at(0);
return node->type != ActionsDAG::ActionType::FUNCTION && node->type != ActionsDAG::ActionType::ARRAY_JOIN;
}
struct ActionsAndName
{
ActionsDAG dag;
std::string name;
};
static ActionsAndName splitSingleAndFilter(ActionsDAG & dag, const ActionsDAG::Node * filter_node)
{
auto split_result = dag.split({filter_node}, true);
dag = std::move(split_result.second);
const auto * split_filter_node = split_result.split_nodes_mapping[filter_node];
auto filter_type = removeLowCardinality(split_filter_node->result_type);
if (!filter_type->onlyNull() && !isUInt8(removeNullable(filter_type)))
{
DataTypePtr cast_type = std::make_shared<DataTypeUInt8>();
if (filter_type->isNullable())
cast_type = std::make_shared<DataTypeNullable>(std::move(cast_type));
split_filter_node = &split_result.first.addCast(*split_filter_node, cast_type, {});
}
split_result.first.getOutputs().emplace(split_result.first.getOutputs().begin(), split_filter_node);
auto name = split_filter_node->result_name;
return ActionsAndName{std::move(split_result.first), std::move(name)};
}
/// Try to split the left most AND atom to a separate DAG.
static std::optional<ActionsAndName> trySplitSingleAndFilter(ActionsDAG & dag, const std::string & filter_name)
{
const auto * filter = &dag.findInOutputs(filter_name);
while (filter->type == ActionsDAG::ActionType::ALIAS)
filter = filter->children.at(0);
if (filter->type != ActionsDAG::ActionType::FUNCTION || filter->function_base->getName() != "and")
return {};
const ActionsDAG::Node * condition_to_split = nullptr;
std::stack<const ActionsDAG::Node *> nodes;
nodes.push(filter);
while (!nodes.empty())
{
const auto * node = nodes.top();
nodes.pop();
if (node->type == ActionsDAG::ActionType::FUNCTION && node->function_base->getName() == "and")
{
/// The order is important. We should take the left-most atom, so put conditions on stack in reverse order.
for (const auto * child : node->children | std::ranges::views::reverse)
nodes.push(child);
continue;
}
if (isTrivialSubtree(node))
continue;
/// Do not split subtree if it's the last non-trivial one.
/// So, split the first found condition only when there is a another one found.
if (condition_to_split)
return splitSingleAndFilter(dag, condition_to_split);
condition_to_split = node;
}
return {};
}
std::vector<ActionsAndName> splitAndChainIntoMultipleFilters(ActionsDAG & dag, const std::string & filter_name)
{
std::vector<ActionsAndName> res;
while (auto condition = trySplitSingleAndFilter(dag, filter_name))
res.push_back(std::move(*condition));
return res;
}
FilterStep::FilterStep(
const Header & input_header_,
ActionsDAG actions_dag_,
@ -50,6 +141,23 @@ FilterStep::FilterStep(
void FilterStep::transformPipeline(QueryPipelineBuilder & pipeline, const BuildQueryPipelineSettings & settings)
{
std::vector<ActionsAndName> and_atoms;
/// Splitting AND filter condition to steps under the setting, which is enabled with merge_filters optimization.
/// This is needed to support short-circuit properly.
if (settings.enable_multiple_filters_transforms_for_and_chain && !actions_dag.hasStatefulFunctions())
and_atoms = splitAndChainIntoMultipleFilters(actions_dag, filter_column_name);
for (auto & and_atom : and_atoms)
{
auto expression = std::make_shared<ExpressionActions>(std::move(and_atom.dag), settings.getActionsSettings());
pipeline.addSimpleTransform([&](const Block & header, QueryPipelineBuilder::StreamType stream_type)
{
bool on_totals = stream_type == QueryPipelineBuilder::StreamType::Totals;
return std::make_shared<FilterTransform>(header, expression, and_atom.name, true, on_totals);
});
}
auto expression = std::make_shared<ExpressionActions>(std::move(actions_dag), settings.getActionsSettings());
pipeline.addSimpleTransform([&](const Block & header, QueryPipelineBuilder::StreamType stream_type)
@ -76,18 +184,45 @@ void FilterStep::transformPipeline(QueryPipelineBuilder & pipeline, const BuildQ
void FilterStep::describeActions(FormatSettings & settings) const
{
String prefix(settings.offset, settings.indent_char);
auto cloned_dag = actions_dag.clone();
std::vector<ActionsAndName> and_atoms;
if (!actions_dag.hasStatefulFunctions())
and_atoms = splitAndChainIntoMultipleFilters(cloned_dag, filter_column_name);
for (auto & and_atom : and_atoms)
{
auto expression = std::make_shared<ExpressionActions>(std::move(and_atom.dag));
settings.out << prefix << "AND column: " << and_atom.name << '\n';
expression->describeActions(settings.out, prefix);
}
settings.out << prefix << "Filter column: " << filter_column_name;
if (remove_filter_column)
settings.out << " (removed)";
settings.out << '\n';
auto expression = std::make_shared<ExpressionActions>(actions_dag.clone());
auto expression = std::make_shared<ExpressionActions>(std::move(cloned_dag));
expression->describeActions(settings.out, prefix);
}
void FilterStep::describeActions(JSONBuilder::JSONMap & map) const
{
auto cloned_dag = actions_dag.clone();
std::vector<ActionsAndName> and_atoms;
if (!actions_dag.hasStatefulFunctions())
and_atoms = splitAndChainIntoMultipleFilters(cloned_dag, filter_column_name);
for (auto & and_atom : and_atoms)
{
auto expression = std::make_shared<ExpressionActions>(std::move(and_atom.dag));
map.add("AND column", and_atom.name);
map.add("Expression", expression->toTree());
}
map.add("Filter Column", filter_column_name);
map.add("Removes Filter", remove_filter_column);

2
src/Processors/QueryPlan/Optimizations/QueryPlanOptimizationSettings.h
View File

@ -32,7 +32,7 @@ struct QueryPlanOptimizationSettings
bool merge_expressions = true;
/// If merge-filters optimization is enabled.
bool merge_filters = false;
bool merge_filters = true;
/// If filter push down optimization is enabled.
bool filter_push_down = true;

2
src/Processors/QueryPlan/ReadFromMergeTree.cpp
View File

@ -176,6 +176,7 @@ namespace Setting
extern const SettingsBool use_skip_indexes;
extern const SettingsBool use_skip_indexes_if_final;
extern const SettingsBool use_uncompressed_cache;
extern const SettingsBool query_plan_merge_filters;
extern const SettingsUInt64 merge_tree_min_read_task_size;
extern const SettingsBool read_in_order_use_virtual_row;
}
@ -208,6 +209,7 @@ static MergeTreeReaderSettings getMergeTreeReaderSettings(
.use_asynchronous_read_from_pool = settings[Setting::allow_asynchronous_read_from_io_pool_for_merge_tree]
&& (settings[Setting::max_streams_to_max_threads_ratio] > 1 || settings[Setting::max_streams_for_merge_tree_reading] > 1),
.enable_multiple_prewhere_read_steps = settings[Setting::enable_multiple_prewhere_read_steps],
.force_short_circuit_execution = settings[Setting::query_plan_merge_filters]
};
}

6
src/Processors/Transforms/AddingDefaultsTransform.cpp
View File

@ -9,6 +9,7 @@
#include <Columns/ColumnsCommon.h>
#include <Columns/ColumnDecimal.h>
#include <Columns/ColumnConst.h>
#include <Columns/ColumnSparse.h>
#include <Columns/FilterDescription.h>
#include <DataTypes/DataTypesNumber.h>
@ -184,7 +185,7 @@ void AddingDefaultsTransform::transform(Chunk & chunk)
std::unordered_map<size_t, MutableColumnPtr> mixed_columns;
for (const ColumnWithTypeAndName & column_def : evaluate_block)
for (auto & column_def : evaluate_block)
{
const String & column_name = column_def.name;
@ -199,6 +200,9 @@ void AddingDefaultsTransform::transform(Chunk & chunk)
if (!defaults_mask.empty())
{
column_read.column = recursiveRemoveSparse(column_read.column);
column_def.column = recursiveRemoveSparse(column_def.column);
/// TODO: FixedString
if (isColumnedAsNumber(column_read.type) || isDecimal(column_read.type))
{

16
src/Storages/MergeTree/IMergeTreeDataPart.cpp
View File

@ -2263,18 +2263,18 @@ void IMergeTreeDataPart::checkConsistencyWithProjections(bool require_part_metad
proj_part->checkConsistency(require_part_metadata);
}
void IMergeTreeDataPart::calculateColumnsAndSecondaryIndicesSizesOnDisk()
void IMergeTreeDataPart::calculateColumnsAndSecondaryIndicesSizesOnDisk(std::optional<Block> columns_sample)
{
calculateColumnsSizesOnDisk();
calculateColumnsSizesOnDisk(columns_sample);
calculateSecondaryIndicesSizesOnDisk();
}
void IMergeTreeDataPart::calculateColumnsSizesOnDisk()
void IMergeTreeDataPart::calculateColumnsSizesOnDisk(std::optional<Block> columns_sample)
{
if (getColumns().empty() || checksums.empty())
throw Exception(ErrorCodes::LOGICAL_ERROR, "Cannot calculate columns sizes when columns or checksums are not initialized");
calculateEachColumnSizes(columns_sizes, total_columns_size);
calculateEachColumnSizes(columns_sizes, total_columns_size, columns_sample);
}
void IMergeTreeDataPart::calculateSecondaryIndicesSizesOnDisk()
@ -2521,22 +2521,24 @@ ColumnPtr IMergeTreeDataPart::getColumnSample(const NameAndTypePair & column) co
StorageMetadataPtr metadata_ptr = storage.getInMemoryMetadataPtr();
StorageSnapshotPtr storage_snapshot_ptr = std::make_shared<StorageSnapshot>(storage, metadata_ptr);
MergeTreeReaderSettings settings;
settings.can_read_part_without_marks = true;
MergeTreeReaderPtr reader = getReader(
cols,
storage_snapshot_ptr,
MarkRanges{MarkRange(0, 1)},
MarkRanges{MarkRange(0, total_mark)},
/*virtual_fields=*/ {},
/*uncompressed_cache=*/{},
storage.getContext()->getMarkCache().get(),
std::make_shared<AlterConversions>(),
MergeTreeReaderSettings{},
settings,
ValueSizeMap{},
ReadBufferFromFileBase::ProfileCallback{});
Columns result;
result.resize(1);
reader->readRows(0, 1, false, 0, result);
reader->readRows(0, total_mark, false, 0, result);
return result[0];
}

6
src/Storages/MergeTree/IMergeTreeDataPart.h
View File

@ -428,7 +428,7 @@ public:
bool shallParticipateInMerges(const StoragePolicyPtr & storage_policy) const;
/// Calculate column and secondary indices sizes on disk.
void calculateColumnsAndSecondaryIndicesSizesOnDisk();
void calculateColumnsAndSecondaryIndicesSizesOnDisk(std::optional<Block> columns_sample = std::nullopt);
std::optional<String> getRelativePathForPrefix(const String & prefix, bool detached = false, bool broken = false) const;
@ -633,7 +633,7 @@ protected:
/// Fill each_columns_size and total_size with sizes from columns files on
/// disk using columns and checksums.
virtual void calculateEachColumnSizes(ColumnSizeByName & each_columns_size, ColumnSize & total_size) const = 0;
virtual void calculateEachColumnSizes(ColumnSizeByName & each_columns_size, ColumnSize & total_size, std::optional<Block> columns_sample) const = 0;
std::optional<String> getRelativePathForDetachedPart(const String & prefix, bool broken) const;
@ -715,7 +715,7 @@ private:
void loadPartitionAndMinMaxIndex();
void calculateColumnsSizesOnDisk();
void calculateColumnsSizesOnDisk(std::optional<Block> columns_sample = std::nullopt);
void calculateSecondaryIndicesSizesOnDisk();

2
src/Storages/MergeTree/IMergeTreeDataPartWriter.h
View File

@ -54,6 +54,8 @@ public:
const MergeTreeIndexGranularity & getIndexGranularity() const { return index_granularity; }
virtual Block getColumnsSample() const = 0;
protected:
SerializationPtr getSerialization(const String & column_name) const;

2
src/Storages/MergeTree/MergeTreeBlockReadUtils.cpp
View File

@ -330,7 +330,7 @@ MergeTreeReadTaskColumns getReadTaskColumns(
auto prewhere_actions = MergeTreeSelectProcessor::getPrewhereActions(
prewhere_info,
actions_settings,
reader_settings.enable_multiple_prewhere_read_steps);
reader_settings.enable_multiple_prewhere_read_steps, reader_settings.force_short_circuit_execution);
for (const auto & step : prewhere_actions.steps)
add_step(*step);

2
src/Storages/MergeTree/MergeTreeDataPartCompact.cpp
View File

@ -80,7 +80,7 @@ MergeTreeDataPartWriterPtr createMergeTreeDataPartCompactWriter(
}
void MergeTreeDataPartCompact::calculateEachColumnSizes(ColumnSizeByName & /*each_columns_size*/, ColumnSize & total_size) const
void MergeTreeDataPartCompact::calculateEachColumnSizes(ColumnSizeByName & /*each_columns_size*/, ColumnSize & total_size, std::optional<Block> /*columns_sample*/) const
{
auto bin_checksum = checksums.files.find(DATA_FILE_NAME_WITH_EXTENSION);
if (bin_checksum != checksums.files.end())

2
src/Storages/MergeTree/MergeTreeDataPartCompact.h
View File

@ -70,7 +70,7 @@ private:
void loadIndexGranularity() override;
/// Compact parts don't support per column size, only total size
void calculateEachColumnSizes(ColumnSizeByName & each_columns_size, ColumnSize & total_size) const override;
void calculateEachColumnSizes(ColumnSizeByName & each_columns_size, ColumnSize & total_size, std::optional<Block> columns_sample) const override;
};
}

8
src/Storages/MergeTree/MergeTreeDataPartWide.cpp
View File

@ -82,7 +82,7 @@ MergeTreeDataPartWriterPtr createMergeTreeDataPartWideWriter(
/// Takes into account the fact that several columns can e.g. share their .size substreams.
/// When calculating totals these should be counted only once.
ColumnSize MergeTreeDataPartWide::getColumnSizeImpl(
const NameAndTypePair & column, std::unordered_set<String> * processed_substreams) const
const NameAndTypePair & column, std::unordered_set<String> * processed_substreams, std::optional<Block> columns_sample) const
{
ColumnSize size;
if (checksums.empty())
@ -108,7 +108,7 @@ ColumnSize MergeTreeDataPartWide::getColumnSizeImpl(
auto mrk_checksum = checksums.files.find(*stream_name + getMarksFileExtension());
if (mrk_checksum != checksums.files.end())
size.marks += mrk_checksum->second.file_size;
});
}, column.type, columns_sample && columns_sample->has(column.name) ? columns_sample->getByName(column.name).column : getColumnSample(column));
return size;
}
@ -374,12 +374,12 @@ std::optional<String> MergeTreeDataPartWide::getFileNameForColumn(const NameAndT
return filename;
}
void MergeTreeDataPartWide::calculateEachColumnSizes(ColumnSizeByName & each_columns_size, ColumnSize & total_size) const
void MergeTreeDataPartWide::calculateEachColumnSizes(ColumnSizeByName & each_columns_size, ColumnSize & total_size, std::optional<Block> columns_sample) const
{
std::unordered_set<String> processed_substreams;
for (const auto & column : columns)
{
ColumnSize size = getColumnSizeImpl(column, &processed_substreams);
ColumnSize size = getColumnSizeImpl(column, &processed_substreams, columns_sample);
each_columns_size[column.name] = size;
total_size.add(size);

4
src/Storages/MergeTree/MergeTreeDataPartWide.h
View File

@ -64,9 +64,9 @@ private:
/// Loads marks index granularity into memory
void loadIndexGranularity() override;
ColumnSize getColumnSizeImpl(const NameAndTypePair & column, std::unordered_set<String> * processed_substreams) const;
ColumnSize getColumnSizeImpl(const NameAndTypePair & column, std::unordered_set<String> * processed_substreams, std::optional<Block> columns_sample) const;
void calculateEachColumnSizes(ColumnSizeByName & each_columns_size, ColumnSize & total_size) const override;
void calculateEachColumnSizes(ColumnSizeByName & each_columns_size, ColumnSize & total_size, std::optional<Block> columns_sample) const override;
};

2
src/Storages/MergeTree/MergeTreeDataPartWriterOnDisk.h
View File

@ -123,6 +123,8 @@ public:
written_offset_columns = written_offset_columns_;
}
Block getColumnsSample() const override { return block_sample; }
protected:
/// Count index_granularity for block and store in `index_granularity`
size_t computeIndexGranularity(const Block & block) const;

18
src/Storages/MergeTree/MergeTreeDataSelectExecutor.cpp
View File

@ -1045,7 +1045,6 @@ MarkRanges MergeTreeDataSelectExecutor::markRangesFromPKRange(
MarkRanges res;
size_t marks_count = part->index_granularity.getMarksCount();
const auto & index = part->getIndex();
if (marks_count == 0)
return res;
@ -1073,14 +1072,19 @@ MarkRanges MergeTreeDataSelectExecutor::markRangesFromPKRange(
auto index_columns = std::make_shared<ColumnsWithTypeAndName>();
const auto & key_indices = key_condition.getKeyIndices();
DataTypes key_types;
for (size_t i : key_indices)
if (!key_indices.empty())
{
if (i < index->size())
index_columns->emplace_back(index->at(i), primary_key.data_types[i], primary_key.column_names[i]);
else
index_columns->emplace_back(); /// The column of the primary key was not loaded in memory - we'll skip it.
const auto & index = part->getIndex();
key_types.emplace_back(primary_key.data_types[i]);
for (size_t i : key_indices)
{
if (i < index->size())
index_columns->emplace_back(index->at(i), primary_key.data_types[i], primary_key.column_names[i]);
else
index_columns->emplace_back(); /// The column of the primary key was not loaded in memory - we'll skip it.
key_types.emplace_back(primary_key.data_types[i]);
}
}
/// If there are no monotonic functions, there is no need to save block reference.

2
src/Storages/MergeTree/MergeTreeIOSettings.h
View File

@ -45,6 +45,8 @@ struct MergeTreeReaderSettings
bool use_asynchronous_read_from_pool = false;
/// If PREWHERE has multiple conditions combined with AND, execute them in separate read/filtering steps.
bool enable_multiple_prewhere_read_steps = false;
/// In case of multiple prewhere steps, execute filtering earlier to support short-circuit properly.
bool force_short_circuit_execution = false;
/// If true, try to lower size of read buffer according to granule size and compressed block size.
bool adjust_read_buffer_size = true;
/// If true, it's allowed to read the whole part without reading marks.

2
src/Storages/MergeTree/MergeTreeReaderWide.cpp
View File

@ -172,7 +172,7 @@ size_t MergeTreeReaderWide::readRows(
throw;
}
if (column->empty())
if (column->empty() && max_rows_to_read > 0)
res_columns[pos] = nullptr;
}

8
src/Storages/MergeTree/MergeTreeSelectProcessor.cpp
View File

@ -91,7 +91,7 @@ MergeTreeSelectProcessor::MergeTreeSelectProcessor(
, algorithm(std::move(algorithm_))
, prewhere_info(prewhere_info_)
, actions_settings(actions_settings_)
, prewhere_actions(getPrewhereActions(prewhere_info, actions_settings, reader_settings_.enable_multiple_prewhere_read_steps))
, prewhere_actions(getPrewhereActions(prewhere_info, actions_settings, reader_settings_.enable_multiple_prewhere_read_steps, reader_settings_.force_short_circuit_execution))
, reader_settings(reader_settings_)
, result_header(transformHeader(pool->getHeader(), prewhere_info))
{
@ -124,9 +124,9 @@ String MergeTreeSelectProcessor::getName() const
return fmt::format("MergeTreeSelect(pool: {}, algorithm: {})", pool->getName(), algorithm->getName());
}
bool tryBuildPrewhereSteps(PrewhereInfoPtr prewhere_info, const ExpressionActionsSettings & actions_settings, PrewhereExprInfo & prewhere);
bool tryBuildPrewhereSteps(PrewhereInfoPtr prewhere_info, const ExpressionActionsSettings & actions_settings, PrewhereExprInfo & prewhere, bool force_short_circuit_execution);
PrewhereExprInfo MergeTreeSelectProcessor::getPrewhereActions(PrewhereInfoPtr prewhere_info, const ExpressionActionsSettings & actions_settings, bool enable_multiple_prewhere_read_steps)
PrewhereExprInfo MergeTreeSelectProcessor::getPrewhereActions(PrewhereInfoPtr prewhere_info, const ExpressionActionsSettings & actions_settings, bool enable_multiple_prewhere_read_steps, bool force_short_circuit_execution)
{
PrewhereExprInfo prewhere_actions;
if (prewhere_info)
@ -147,7 +147,7 @@ PrewhereExprInfo MergeTreeSelectProcessor::getPrewhereActions(PrewhereInfoPtr pr
}
if (!enable_multiple_prewhere_read_steps ||
!tryBuildPrewhereSteps(prewhere_info, actions_settings, prewhere_actions))
!tryBuildPrewhereSteps(prewhere_info, actions_settings, prewhere_actions, force_short_circuit_execution))
{
PrewhereExprStep prewhere_step
{

3
src/Storages/MergeTree/MergeTreeSelectProcessor.h
View File

@ -73,7 +73,8 @@ public:
static PrewhereExprInfo getPrewhereActions(
PrewhereInfoPtr prewhere_info,
const ExpressionActionsSettings & actions_settings,
bool enable_multiple_prewhere_read_steps);
bool enable_multiple_prewhere_read_steps,
bool force_short_circuit_execution);
void addPartLevelToChunk(bool add_part_level_) { add_part_level = add_part_level_; }

112
src/Storages/MergeTree/MergeTreeSplitPrewhereIntoReadSteps.cpp
View File

@ -4,6 +4,7 @@
#include <Storages/SelectQueryInfo.h>
#include <Storages/MergeTree/MergeTreeRangeReader.h>
#include <DataTypes/DataTypeString.h>
#include <DataTypes/DataTypeLowCardinality.h>
#include <Interpreters/ExpressionActions.h>
@ -57,9 +58,9 @@ struct DAGNodeRef
const ActionsDAG::Node * node;
};
/// Result name -> DAGNodeRef
using OriginalToNewNodeMap = std::unordered_map<String, DAGNodeRef>;
using NodeNameToLastUsedStepMap = std::unordered_map<String, size_t>;
/// ResultNode -> DAGNodeRef
using OriginalToNewNodeMap = std::unordered_map<const ActionsDAG::Node *, DAGNodeRef>;
using NodeNameToLastUsedStepMap = std::unordered_map<const ActionsDAG::Node *, size_t>;
/// Clones the part of original DAG responsible for computing the original_dag_node and adds it to the new DAG.
const ActionsDAG::Node & addClonedDAGToDAG(
@ -69,25 +70,28 @@ const ActionsDAG::Node & addClonedDAGToDAG(
OriginalToNewNodeMap & node_remap,
NodeNameToLastUsedStepMap & node_to_step_map)
{
const String & node_name = original_dag_node->result_name;
/// Look for the node in the map of already known nodes
if (node_remap.contains(node_name))
if (node_remap.contains(original_dag_node))
{
/// If the node is already in the new DAG, return it
const auto & node_ref = node_remap.at(node_name);
const auto & node_ref = node_remap.at(original_dag_node);
if (node_ref.dag == new_dag.get())
return *node_ref.node;
/// If the node is known from the previous steps, add it as an input, except for constants
if (original_dag_node->type != ActionsDAG::ActionType::COLUMN)
{
node_ref.dag->addOrReplaceInOutputs(*node_ref.node);
/// If the node was found in node_remap, it was not added to outputs yet.
/// The only exception is the filter node, which is always the first one.
if (node_ref.dag->getOutputs().at(0) != node_ref.node)
node_ref.dag->getOutputs().push_back(node_ref.node);
const auto & new_node = new_dag->addInput(node_ref.node->result_name, node_ref.node->result_type);
node_remap[node_name] = {new_dag.get(), &new_node}; /// TODO: here we update the node reference. Is it always correct?
node_remap[original_dag_node] = {new_dag.get(), &new_node};
/// Remember the index of the last step which reuses this node.
/// We cannot remove this node from the outputs before that step.
node_to_step_map[node_name] = step;
node_to_step_map[original_dag_node] = step;
return new_node;
}
}
@ -96,7 +100,7 @@ const ActionsDAG::Node & addClonedDAGToDAG(
if (original_dag_node->type == ActionsDAG::ActionType::INPUT)
{
const auto & new_node = new_dag->addInput(original_dag_node->result_name, original_dag_node->result_type);
node_remap[node_name] = {new_dag.get(), &new_node};
node_remap[original_dag_node] = {new_dag.get(), &new_node};
return new_node;
}
@ -105,7 +109,7 @@ const ActionsDAG::Node & addClonedDAGToDAG(
{
const auto & new_node = new_dag->addColumn(
ColumnWithTypeAndName(original_dag_node->column, original_dag_node->result_type, original_dag_node->result_name));
node_remap[node_name] = {new_dag.get(), &new_node};
node_remap[original_dag_node] = {new_dag.get(), &new_node};
return new_node;
}
@ -113,7 +117,7 @@ const ActionsDAG::Node & addClonedDAGToDAG(
{
const auto & alias_child = addClonedDAGToDAG(step, original_dag_node->children[0], new_dag, node_remap, node_to_step_map);
const auto & new_node = new_dag->addAlias(alias_child, original_dag_node->result_name);
node_remap[node_name] = {new_dag.get(), &new_node};
node_remap[original_dag_node] = {new_dag.get(), &new_node};
return new_node;
}
@ -128,7 +132,7 @@ const ActionsDAG::Node & addClonedDAGToDAG(
}
const auto & new_node = new_dag->addFunction(original_dag_node->function_base, new_children, original_dag_node->result_name);
node_remap[node_name] = {new_dag.get(), &new_node};
node_remap[original_dag_node] = {new_dag.get(), &new_node};
return new_node;
}
@ -138,11 +142,9 @@ const ActionsDAG::Node & addClonedDAGToDAG(
const ActionsDAG::Node & addFunction(
const ActionsDAGPtr & new_dag,
const FunctionOverloadResolverPtr & function,
ActionsDAG::NodeRawConstPtrs children,
OriginalToNewNodeMap & node_remap)
ActionsDAG::NodeRawConstPtrs children)
{
const auto & new_node = new_dag->addFunction(function, children, "");
node_remap[new_node.result_name] = {new_dag.get(), &new_node};
return new_node;
}
@ -152,14 +154,12 @@ const ActionsDAG::Node & addFunction(
const ActionsDAG::Node & addCast(
const ActionsDAGPtr & dag,
const ActionsDAG::Node & node_to_cast,
const DataTypePtr & to_type,
OriginalToNewNodeMap & node_remap)
const DataTypePtr & to_type)
{
if (!node_to_cast.result_type->equals(*to_type))
return node_to_cast;
const auto & new_node = dag->addCast(node_to_cast, to_type, {});
node_remap[new_node.result_name] = {dag.get(), &new_node};
return new_node;
}
@ -169,8 +169,7 @@ const ActionsDAG::Node & addCast(
/// 2. makes sure that the result contains only 0 or 1 values even if the source column contains non-boolean values.
const ActionsDAG::Node & addAndTrue(
const ActionsDAGPtr & dag,
const ActionsDAG::Node & filter_node_to_normalize,
OriginalToNewNodeMap & node_remap)
const ActionsDAG::Node & filter_node_to_normalize)
{
Field const_true_value(true);
@ -181,7 +180,7 @@ const ActionsDAG::Node & addAndTrue(
const auto * const_true_node = &dag->addColumn(std::move(const_true_column));
ActionsDAG::NodeRawConstPtrs children = {&filter_node_to_normalize, const_true_node};
FunctionOverloadResolverPtr func_builder_and = std::make_unique<FunctionToOverloadResolverAdaptor>(std::make_shared<FunctionAnd>());
return addFunction(dag, func_builder_and, children, node_remap);
return addFunction(dag, func_builder_and, children);
}
}
@ -206,7 +205,11 @@ const ActionsDAG::Node & addAndTrue(
/// 6. Find all outputs of the original DAG
/// 7. Find all outputs that were computed in the already built DAGs, mark these nodes as outputs in the steps where they were computed
/// 8. Add computation of the remaining outputs to the last step with the procedure similar to 4
bool tryBuildPrewhereSteps(PrewhereInfoPtr prewhere_info, const ExpressionActionsSettings & actions_settings, PrewhereExprInfo & prewhere)
bool tryBuildPrewhereSteps(
PrewhereInfoPtr prewhere_info,
const ExpressionActionsSettings & actions_settings,
PrewhereExprInfo & prewhere,
bool force_short_circuit_execution)
{
if (!prewhere_info)
return true;
@ -243,7 +246,10 @@ bool tryBuildPrewhereSteps(PrewhereInfoPtr prewhere_info, const ExpressionAction
struct Step
{
ActionsDAGPtr actions;
String column_name;
/// Original condition, in case if we have only one condition, and it was not casted
const ActionsDAG::Node * original_node;
/// Result condition node
const ActionsDAG::Node * result_node;
};
std::vector<Step> steps;
@ -254,7 +260,8 @@ bool tryBuildPrewhereSteps(PrewhereInfoPtr prewhere_info, const ExpressionAction
{
const auto & condition_group = condition_groups[step_index];
ActionsDAGPtr step_dag = std::make_unique<ActionsDAG>();
String result_name;
const ActionsDAG::Node * original_node = nullptr;
const ActionsDAG::Node * result_node;
std::vector<const ActionsDAG::Node *> new_condition_nodes;
for (const auto * node : condition_group)
@ -267,48 +274,37 @@ bool tryBuildPrewhereSteps(PrewhereInfoPtr prewhere_info, const ExpressionAction
{
/// Add AND function to combine the conditions
FunctionOverloadResolverPtr func_builder_and = std::make_unique<FunctionToOverloadResolverAdaptor>(std::make_shared<FunctionAnd>());
const auto & and_function_node = addFunction(step_dag, func_builder_and, new_condition_nodes, node_remap);
step_dag->addOrReplaceInOutputs(and_function_node);
result_name = and_function_node.result_name;
const auto & and_function_node = addFunction(step_dag, func_builder_and, new_condition_nodes);
result_node = &and_function_node;
}
else
{
const auto & result_node = *new_condition_nodes.front();
result_node = new_condition_nodes.front();
/// Check if explicit cast is needed for the condition to serve as a filter.
const auto result_type_name = result_node.result_type->getName();
if (result_type_name == "UInt8" ||
result_type_name == "Nullable(UInt8)" ||
result_type_name == "LowCardinality(UInt8)" ||
result_type_name == "LowCardinality(Nullable(UInt8))")
{
/// No need to cast
step_dag->addOrReplaceInOutputs(result_node);
result_name = result_node.result_name;
}
else
if (!isUInt8(removeNullable(removeLowCardinality(result_node->result_type))))
{
/// Build "condition AND True" expression to "cast" the condition to UInt8 or Nullable(UInt8) depending on its type.
const auto & cast_node = addAndTrue(step_dag, result_node, node_remap);
step_dag->addOrReplaceInOutputs(cast_node);
result_name = cast_node.result_name;
result_node = &addAndTrue(step_dag, *result_node);
}
}
steps.push_back({std::move(step_dag), result_name});
step_dag->getOutputs().insert(step_dag->getOutputs().begin(), result_node);
steps.push_back({std::move(step_dag), original_node, result_node});
}
/// 6. Find all outputs of the original DAG
auto original_outputs = prewhere_info->prewhere_actions.getOutputs();
steps.back().actions->getOutputs().clear();
/// 7. Find all outputs that were computed in the already built DAGs, mark these nodes as outputs in the steps where they were computed
/// 8. Add computation of the remaining outputs to the last step with the procedure similar to 4
NameSet all_output_names;
std::unordered_set<const ActionsDAG::Node *> all_outputs;
for (const auto * output : original_outputs)
{
all_output_names.insert(output->result_name);
if (node_remap.contains(output->result_name))
all_outputs.insert(output);
if (node_remap.contains(output))
{
const auto & new_node_info = node_remap[output->result_name];
new_node_info.dag->addOrReplaceInOutputs(*new_node_info.node);
const auto & new_node_info = node_remap[output];
new_node_info.dag->getOutputs().push_back(new_node_info.node);
}
else if (output->result_name == prewhere_info->prewhere_column_name)
{
@ -319,20 +315,21 @@ bool tryBuildPrewhereSteps(PrewhereInfoPtr prewhere_info, const ExpressionAction
/// 1. AND the last condition with constant True. This is needed to make sure that in the last step filter has UInt8 type
/// but contains values other than 0 and 1 (e.g. if it is (number%5) it contains 2,3,4)
/// 2. CAST the result to the exact type of the PREWHERE column from the original DAG
const auto & last_step_result_node_info = node_remap[steps.back().column_name];
auto & last_step_dag = steps.back().actions;
auto & last_step_result_node = steps.back().result_node;
/// Build AND(last_step_result_node, true)
const auto & and_node = addAndTrue(last_step_dag, *last_step_result_node_info.node, node_remap);
const auto & and_node = addAndTrue(last_step_dag, *last_step_result_node);
/// Build CAST(and_node, type of PREWHERE column)
const auto & cast_node = addCast(last_step_dag, and_node, output->result_type, node_remap);
const auto & cast_node = addCast(last_step_dag, and_node, output->result_type);
/// Add alias for the result with the name of the PREWHERE column
const auto & prewhere_result_node = last_step_dag->addAlias(cast_node, output->result_name);
last_step_dag->addOrReplaceInOutputs(prewhere_result_node);
last_step_dag->getOutputs().push_back(&prewhere_result_node);
steps.back().result_node = &prewhere_result_node;
}
else
{
const auto & node_in_new_dag = addClonedDAGToDAG(steps.size() - 1, output, steps.back().actions, node_remap, node_to_step);
steps.back().actions->addOrReplaceInOutputs(node_in_new_dag);
steps.back().actions->getOutputs().push_back(&node_in_new_dag);
}
}
@ -345,17 +342,18 @@ bool tryBuildPrewhereSteps(PrewhereInfoPtr prewhere_info, const ExpressionAction
{
.type = PrewhereExprStep::Filter,
.actions = std::make_shared<ExpressionActions>(std::move(*step.actions), actions_settings),
.filter_column_name = step.column_name,
.filter_column_name = step.result_node->result_name,
/// Don't remove if it's in the list of original outputs
.remove_filter_column =
!all_output_names.contains(step.column_name) && node_to_step[step.column_name] <= step_index,
.need_filter = false,
step.original_node && !all_outputs.contains(step.original_node) && node_to_step[step.original_node] <= step_index,
.need_filter = force_short_circuit_execution,
.perform_alter_conversions = true,
};
prewhere.steps.push_back(std::make_shared<PrewhereExprStep>(std::move(new_step)));
}
prewhere.steps.back()->remove_filter_column = prewhere_info->remove_prewhere_column;
prewhere.steps.back()->need_filter = prewhere_info->need_filter;
}

2
src/Storages/MergeTree/MergedBlockOutputStream.cpp
View File

@ -209,7 +209,7 @@ MergedBlockOutputStream::Finalizer MergedBlockOutputStream::finalizePartAsync(
new_part->index_granularity = writer->getIndexGranularity();
/// Just in case
new_part->index_granularity.shrinkToFitInMemory();
new_part->calculateColumnsAndSecondaryIndicesSizesOnDisk();
new_part->calculateColumnsAndSecondaryIndicesSizesOnDisk(writer->getColumnsSample());
/// In mutation, existing_rows_count is already calculated in PartMergerWriter
/// In merge situation, lightweight deleted rows was physically deleted, existing_rows_count equals rows_count

4
src/Storages/ObjectStorageQueue/ObjectStorageQueueSettings.cpp
View File

@ -30,8 +30,8 @@ namespace ErrorCodes
DECLARE(UInt64, tracked_files_limit, 1000, "For unordered mode. Max set size for tracking processed files in ZooKeeper", 0) \
DECLARE(UInt64, tracked_file_ttl_sec, 0, "Maximum number of seconds to store processed files in ZooKeeper node (store forever by default)", 0) \
DECLARE(UInt64, polling_min_timeout_ms, 1000, "Minimal timeout before next polling", 0) \
DECLARE(UInt64, polling_max_timeout_ms, 10000, "Maximum timeout before next polling", 0) \
DECLARE(UInt64, polling_backoff_ms, 1000, "Polling backoff", 0) \
DECLARE(UInt64, polling_max_timeout_ms, 10 * 60 * 1000, "Maximum timeout before next polling", 0) \
DECLARE(UInt64, polling_backoff_ms, 30 * 1000, "Polling backoff", 0) \
DECLARE(UInt32, cleanup_interval_min_ms, 60000, "For unordered mode. Polling backoff min for cleanup", 0) \
DECLARE(UInt32, cleanup_interval_max_ms, 60000, "For unordered mode. Polling backoff max for cleanup", 0) \
DECLARE(UInt32, buckets, 0, "Number of buckets for Ordered mode parallel processing", 0) \

2
src/Storages/ObjectStorageQueue/ObjectStorageQueueSource.cpp
View File

@ -659,7 +659,7 @@ void ObjectStorageQueueSource::applyActionAfterProcessing(const String & path)
{
if (files_metadata->getTableMetadata().after_processing == ObjectStorageQueueAction::DELETE)
{
object_storage->removeObject(StoredObject(path));
object_storage->removeObjectIfExists(StoredObject(path));
}
}

16
src/Storages/Utils.cpp
View File

@ -1,10 +1,13 @@
#include <vector>
#include <Storages/Utils.h>
#include <Storages/IStorage.h>
#include <Storages/StorageReplicatedMergeTree.h>
namespace CurrentMetrics
{
extern const Metric AttachedTable;
extern const Metric AttachedReplicatedTable;
extern const Metric AttachedView;
extern const Metric AttachedDictionary;
}
@ -12,17 +15,20 @@ namespace CurrentMetrics
namespace DB
{
CurrentMetrics::Metric getAttachedCounterForStorage(const StoragePtr & storage)
std::vector<CurrentMetrics::Metric> getAttachedCountersForStorage(const StoragePtr & storage)
{
if (storage->isView())
{
return CurrentMetrics::AttachedView;
return {CurrentMetrics::AttachedView};
}
if (storage->isDictionary())
{
return CurrentMetrics::AttachedDictionary;
return {CurrentMetrics::AttachedDictionary};
}
return CurrentMetrics::AttachedTable;
if (auto * replicated_storage = typeid_cast<StorageReplicatedMergeTree *>(storage.get()))
{
return {CurrentMetrics::AttachedTable, CurrentMetrics::AttachedReplicatedTable};
}
return {CurrentMetrics::AttachedTable};
}
}

2
src/Storages/Utils.h
View File

@ -6,5 +6,5 @@
namespace DB
{
CurrentMetrics::Metric getAttachedCounterForStorage(const StoragePtr & storage);
std::vector<CurrentMetrics::Metric> getAttachedCountersForStorage(const StoragePtr & storage);
}

4
tests/ci/ci.py
View File

@ -798,10 +798,6 @@ def _upload_build_profile_data(
logging.info("Unknown CI logs host, skip uploading build profile data")
return
if not pr_info.number == 0:
logging.info("Skipping uploading build profile data for PRs")
return
instance_type = get_instance_type()
instance_id = get_instance_id()
auth = {

8
tests/integration/compose/docker_compose_rabbitmq.yml
View File

@ -16,4 +16,10 @@ services:
- /misc/rabbitmq/ca-cert.pem:/etc/rabbitmq/ca-cert.pem
- /misc/rabbitmq/server-cert.pem:/etc/rabbitmq/server-cert.pem
- /misc/rabbitmq/server-key.pem:/etc/rabbitmq/server-key.pem
- /misc/rabbitmq/enabled_plugins:/etc/rabbitmq/enabled_plugins
- /misc/rabbitmq/enabled_plugins:/etc/rabbitmq/enabled_plugins
# https://www.rabbitmq.com/docs/monitoring#health-checks
healthcheck:
test: rabbitmq-diagnostics -q ping
interval: 10s
retries: 10
timeout: 2s

29
tests/integration/test_backup_restore_on_cluster/test_cancel_backup.py
View File

@ -251,23 +251,16 @@ def kill_query(
if is_initial_query is not None
else ""
)
old_time = time.monotonic()
node.query(
f"KILL QUERY WHERE (query_kind='{query_kind}') AND (query LIKE '%{id}%'){filter_for_is_initial_query} SYNC"
)
node.query("SYSTEM FLUSH LOGS")
duration = (
int(
node.query(
f"SELECT query_duration_ms FROM system.query_log WHERE query_kind='KillQuery' AND query LIKE '%{id}%' AND type='QueryFinish'"
)
)
/ 1000
)
waited = time.monotonic() - old_time
print(
f"{get_node_name(node)}: Cancelled {operation_name} {id} after {duration} seconds"
f"{get_node_name(node)}: Cancelled {operation_name} {id} after {waited} seconds"
)
if timeout is not None:
assert duration < timeout
assert waited < timeout
# Stops all ZooKeeper servers.
@ -305,7 +298,7 @@ def sleep(seconds):
class NoTrashChecker:
def __init__(self):
self.expect_backups = []
self.expect_unfinished_backups = []
self.allow_unfinished_backups = []
self.expect_errors = []
self.allow_errors = []
self.check_zookeeper = True
@ -373,7 +366,7 @@ class NoTrashChecker:
if unfinished_backups:
print(f"Found unfinished backups: {unfinished_backups}")
assert new_backups == set(self.expect_backups)
assert unfinished_backups == set(self.expect_unfinished_backups)
assert unfinished_backups.difference(self.allow_unfinished_backups) == set()
all_errors = set()
start_time = time.strftime(
@ -641,7 +634,7 @@ def test_long_disconnection_stops_backup():
assert get_status(initiator, backup_id=backup_id) == "CREATING_BACKUP"
assert get_num_system_processes(initiator, backup_id=backup_id) >= 1
no_trash_checker.expect_unfinished_backups = [backup_id]
no_trash_checker.allow_unfinished_backups = [backup_id]
no_trash_checker.allow_errors = [
"FAILED_TO_SYNC_BACKUP_OR_RESTORE",
"KEEPER_EXCEPTION",
@ -674,7 +667,7 @@ def test_long_disconnection_stops_backup():
# A backup is expected to fail, but it isn't expected to fail too soon.
print(f"Backup failed after {time_to_fail} seconds disconnection")
assert time_to_fail > 3
assert time_to_fail < 30
assert time_to_fail < 35
# A backup must NOT be stopped if Zookeeper is disconnected shorter than `failure_after_host_disconnected_for_seconds`.
@ -695,7 +688,7 @@ def test_short_disconnection_doesnt_stop_backup():
backup_id = random_id()
initiator.query(
f"BACKUP TABLE tbl ON CLUSTER 'cluster' TO {get_backup_name(backup_id)} SETTINGS id='{backup_id}' ASYNC",
settings={"backup_restore_failure_after_host_disconnected_for_seconds": 6},
settings={"backup_restore_failure_after_host_disconnected_for_seconds": 10},
)
assert get_status(initiator, backup_id=backup_id) == "CREATING_BACKUP"
@ -703,13 +696,13 @@ def test_short_disconnection_doesnt_stop_backup():
# Dropping connection for less than `failure_after_host_disconnected_for_seconds`
with PartitionManager() as pm:
random_sleep(3)
random_sleep(4)
node_to_drop_zk_connection = random_node()
print(
f"Dropping connection between {get_node_name(node_to_drop_zk_connection)} and ZooKeeper"
)
pm.drop_instance_zk_connections(node_to_drop_zk_connection)
random_sleep(3)
random_sleep(4)
print(
f"Restoring connection between {get_node_name(node_to_drop_zk_connection)} and ZooKeeper"
)

30
tests/integration/test_failed_mutations/test.py
View File

@ -27,6 +27,7 @@ REPLICATED_POSTPONE_MUTATION_LOG = (
POSTPONE_MUTATION_LOG = (
"According to exponential backoff policy, do not perform mutations for the part"
)
FAILING_MUTATION_QUERY = "ALTER TABLE test_mutations DELETE WHERE x IN (SELECT throwIf(1)) SETTINGS allow_nondeterministic_mutations = 1"
all_nodes = [node_with_backoff, node_no_backoff]
@ -83,17 +84,13 @@ def test_exponential_backoff_with_merge_tree(started_cluster, node, found_in_log
assert not node.contains_in_log(POSTPONE_MUTATION_LOG)
# Executing incorrect mutation.
node.query(
"ALTER TABLE test_mutations DELETE WHERE x IN (SELECT x FROM notexist_table) SETTINGS allow_nondeterministic_mutations=1"
)
node.query(FAILING_MUTATION_QUERY)
check_logs()
node.query("KILL MUTATION WHERE table='test_mutations'")
# Check that after kill new parts mutations are postponing.
node.query(
"ALTER TABLE test_mutations DELETE WHERE x IN (SELECT x FROM notexist_table) SETTINGS allow_nondeterministic_mutations=1"
)
node.query(FAILING_MUTATION_QUERY)
check_logs()
@ -101,9 +98,7 @@ def test_exponential_backoff_with_merge_tree(started_cluster, node, found_in_log
def test_exponential_backoff_with_replicated_tree(started_cluster):
prepare_cluster(True)
node_with_backoff.query(
"ALTER TABLE test_mutations DELETE WHERE x IN (SELECT x FROM notexist_table) SETTINGS allow_nondeterministic_mutations=1"
)
node_with_backoff.query(FAILING_MUTATION_QUERY)
assert node_with_backoff.wait_for_log_line(REPLICATED_POSTPONE_MUTATION_LOG)
assert not node_no_backoff.contains_in_log(REPLICATED_POSTPONE_MUTATION_LOG)
@ -114,7 +109,7 @@ def test_exponential_backoff_create_dependent_table(started_cluster):
# Executing incorrect mutation.
node_with_backoff.query(
"ALTER TABLE test_mutations DELETE WHERE x IN (SELECT x FROM dep_table) SETTINGS allow_nondeterministic_mutations=1"
"ALTER TABLE test_mutations DELETE WHERE x IN (SELECT x FROM dep_table) SETTINGS allow_nondeterministic_mutations = 1, validate_mutation_query = 0"
)
# Creating dependent table for mutation.
@ -148,9 +143,7 @@ def test_exponential_backoff_setting_override(started_cluster):
node.query("INSERT INTO test_mutations SELECT * FROM system.numbers LIMIT 10")
# Executing incorrect mutation.
node.query(
"ALTER TABLE test_mutations DELETE WHERE x IN (SELECT x FROM dep_table) SETTINGS allow_nondeterministic_mutations=1"
)
node.query(FAILING_MUTATION_QUERY)
assert not node.contains_in_log(POSTPONE_MUTATION_LOG)
@ -166,9 +159,7 @@ def test_backoff_clickhouse_restart(started_cluster, replicated_table):
node = node_with_backoff
# Executing incorrect mutation.
node.query(
"ALTER TABLE test_mutations DELETE WHERE x IN (SELECT x FROM dep_table) SETTINGS allow_nondeterministic_mutations=1"
)
node.query(FAILING_MUTATION_QUERY)
assert node.wait_for_log_line(
REPLICATED_POSTPONE_MUTATION_LOG if replicated_table else POSTPONE_MUTATION_LOG
)
@ -193,11 +184,10 @@ def test_no_backoff_after_killing_mutation(started_cluster, replicated_table):
node = node_with_backoff
# Executing incorrect mutation.
node.query(
"ALTER TABLE test_mutations DELETE WHERE x IN (SELECT x FROM dep_table) SETTINGS allow_nondeterministic_mutations=1"
)
node.query(FAILING_MUTATION_QUERY)
# Executing correct mutation.
node.query("ALTER TABLE test_mutations DELETE WHERE x=1")
node.query("ALTER TABLE test_mutations DELETE WHERE x=1")
assert node.wait_for_log_line(
REPLICATED_POSTPONE_MUTATION_LOG if replicated_table else POSTPONE_MUTATION_LOG
)

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