ClickHouse/src/Databases/DatabaseReplicated.cpp

1020 lines
46 KiB
C++
Raw Normal View History

2020-10-22 15:08:00 +00:00
#include <DataTypes/DataTypeString.h>
2020-04-05 12:18:51 +00:00
#include <Databases/DatabaseReplicated.h>
#include <IO/ReadBufferFromFile.h>
2020-05-11 12:55:17 +00:00
#include <IO/ReadBufferFromString.h>
2020-04-05 12:18:51 +00:00
#include <IO/ReadHelpers.h>
#include <IO/WriteHelpers.h>
#include <Interpreters/Context.h>
2020-05-11 12:55:17 +00:00
#include <Interpreters/executeQuery.h>
2020-04-05 12:18:51 +00:00
#include <Parsers/queryToString.h>
2020-05-12 13:35:05 +00:00
#include <Common/Exception.h>
2020-04-05 12:18:51 +00:00
#include <Common/ZooKeeper/KeeperException.h>
#include <Common/ZooKeeper/Types.h>
#include <Common/ZooKeeper/ZooKeeper.h>
2020-11-27 14:04:03 +00:00
#include <Databases/DatabaseReplicatedWorker.h>
#include <Interpreters/DDLTask.h>
#include <Interpreters/executeDDLQueryOnCluster.h>
2020-11-13 18:35:45 +00:00
#include <Interpreters/Cluster.h>
2021-10-02 07:13:14 +00:00
#include <base/getFQDNOrHostName.h>
#include <Parsers/ASTAlterQuery.h>
#include <Parsers/ASTDropQuery.h>
#include <Parsers/ASTFunction.h>
2020-11-27 14:04:03 +00:00
#include <Parsers/ParserCreateQuery.h>
#include <Parsers/parseQuery.h>
#include <Interpreters/InterpreterCreateQuery.h>
2021-06-01 10:24:06 +00:00
#include <Interpreters/evaluateConstantExpression.h>
2020-11-27 14:04:03 +00:00
#include <Parsers/formatAST.h>
#include <Backups/IRestoreCoordination.h>
#include <Backups/RestorerFromBackup.h>
2021-03-16 20:01:20 +00:00
#include <Common/Macros.h>
#include <base/chrono_io.h>
2020-04-05 12:18:51 +00:00
namespace DB
{
namespace ErrorCodes
{
extern const int NO_ZOOKEEPER;
extern const int LOGICAL_ERROR;
2020-06-24 12:45:42 +00:00
extern const int BAD_ARGUMENTS;
2020-11-13 18:35:45 +00:00
extern const int REPLICA_IS_ALREADY_EXIST;
2020-11-19 10:34:45 +00:00
extern const int DATABASE_REPLICATION_FAILED;
2020-11-20 16:06:27 +00:00
extern const int UNKNOWN_DATABASE;
2021-02-12 16:22:01 +00:00
extern const int UNKNOWN_TABLE;
2020-11-29 11:45:32 +00:00
extern const int NOT_IMPLEMENTED;
2021-02-04 19:41:44 +00:00
extern const int INCORRECT_QUERY;
2021-02-08 09:14:17 +00:00
extern const int ALL_CONNECTION_TRIES_FAILED;
2021-07-30 16:34:18 +00:00
extern const int NO_ACTIVE_REPLICAS;
extern const int INCONSISTENT_METADATA_FOR_BACKUP;
extern const int CANNOT_RESTORE_TABLE;
2020-04-05 12:18:51 +00:00
}
2021-02-08 09:14:17 +00:00
static constexpr const char * DROPPED_MARK = "DROPPED";
2021-02-11 22:23:40 +00:00
static constexpr const char * BROKEN_TABLES_SUFFIX = "_broken_tables";
static constexpr const char * BROKEN_REPLICATED_TABLES_SUFFIX = "_broken_replicated_tables";
2021-02-08 09:14:17 +00:00
2020-11-13 18:35:45 +00:00
zkutil::ZooKeeperPtr DatabaseReplicated::getZooKeeper() const
2020-04-05 12:18:51 +00:00
{
return getContext()->getZooKeeper();
2020-04-05 12:18:51 +00:00
}
static inline String getHostID(ContextPtr global_context, const UUID & db_uuid)
2020-04-05 12:18:51 +00:00
{
return Cluster::Address::toString(getFQDNOrHostName(), global_context->getTCPPort()) + ':' + toString(db_uuid);
2020-04-05 12:18:51 +00:00
}
2020-11-13 18:35:45 +00:00
DatabaseReplicated::~DatabaseReplicated() = default;
2020-04-05 12:18:51 +00:00
DatabaseReplicated::DatabaseReplicated(
const String & name_,
const String & metadata_path_,
UUID uuid,
2020-04-05 12:18:51 +00:00
const String & zookeeper_path_,
2020-10-27 09:19:45 +00:00
const String & shard_name_,
2020-04-05 12:18:51 +00:00
const String & replica_name_,
2021-02-19 23:41:58 +00:00
DatabaseReplicatedSettings db_settings_,
ContextPtr context_)
: DatabaseAtomic(name_, metadata_path_, uuid, "DatabaseReplicated (" + name_ + ")", context_)
2020-04-05 12:18:51 +00:00
, zookeeper_path(zookeeper_path_)
2020-10-27 09:19:45 +00:00
, shard_name(shard_name_)
2020-04-05 12:18:51 +00:00
, replica_name(replica_name_)
2021-02-19 23:41:58 +00:00
, db_settings(std::move(db_settings_))
2020-04-05 12:18:51 +00:00
{
2020-10-27 09:19:45 +00:00
if (zookeeper_path.empty() || shard_name.empty() || replica_name.empty())
2020-11-13 18:35:45 +00:00
throw Exception("ZooKeeper path, shard and replica names must be non-empty", ErrorCodes::BAD_ARGUMENTS);
if (shard_name.find('/') != std::string::npos || replica_name.find('/') != std::string::npos)
throw Exception("Shard and replica names should not contain '/'", ErrorCodes::BAD_ARGUMENTS);
2021-02-08 09:14:17 +00:00
if (shard_name.find('|') != std::string::npos || replica_name.find('|') != std::string::npos)
throw Exception("Shard and replica names should not contain '|'", ErrorCodes::BAD_ARGUMENTS);
2020-06-24 12:45:42 +00:00
if (zookeeper_path.back() == '/')
2020-04-05 12:18:51 +00:00
zookeeper_path.resize(zookeeper_path.size() - 1);
2020-11-13 18:35:45 +00:00
2020-10-22 15:08:00 +00:00
/// If zookeeper chroot prefix is used, path should start with '/', because chroot concatenates without it.
2020-06-24 12:45:42 +00:00
if (zookeeper_path.front() != '/')
2020-04-05 12:18:51 +00:00
zookeeper_path = "/" + zookeeper_path;
if (!db_settings.collection_name.value.empty())
fillClusterAuthInfo(db_settings.collection_name.value, context_->getConfigRef());
2020-11-13 18:35:45 +00:00
}
2020-05-24 17:13:53 +00:00
2021-02-08 09:14:17 +00:00
String DatabaseReplicated::getFullReplicaName() const
{
return shard_name + '|' + replica_name;
}
std::pair<String, String> DatabaseReplicated::parseFullReplicaName(const String & name)
{
String shard;
String replica;
auto pos = name.find('|');
if (pos == std::string::npos || name.find('|', pos + 1) != std::string::npos)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Incorrect replica identifier: {}", name);
shard = name.substr(0, pos);
replica = name.substr(pos + 1);
return {shard, replica};
}
ClusterPtr DatabaseReplicated::getCluster() const
{
std::lock_guard lock{mutex};
2021-03-29 13:44:41 +00:00
if (cluster)
return cluster;
cluster = getClusterImpl();
return cluster;
}
void DatabaseReplicated::setCluster(ClusterPtr && new_cluster)
{
std::lock_guard lock{mutex};
cluster = std::move(new_cluster);
}
ClusterPtr DatabaseReplicated::getClusterImpl() const
{
2021-02-08 09:14:17 +00:00
Strings hosts;
Strings host_ids;
auto zookeeper = getContext()->getZooKeeper();
2021-02-08 09:14:17 +00:00
constexpr int max_retries = 10;
int iteration = 0;
bool success = false;
while (++iteration <= max_retries)
{
host_ids.resize(0);
Coordination::Stat stat;
hosts = zookeeper->getChildren(zookeeper_path + "/replicas", &stat);
if (hosts.empty())
2021-07-30 16:34:18 +00:00
throw Exception(ErrorCodes::NO_ACTIVE_REPLICAS, "No replicas of database {} found. "
"It's possible if the first replica is not fully created yet "
2022-06-27 20:48:27 +00:00
"or if the last replica was just dropped or due to logical error", zookeeper_path);
2021-03-15 08:22:42 +00:00
Int32 cversion = stat.cversion;
2022-01-30 19:49:48 +00:00
::sort(hosts.begin(), hosts.end());
2021-02-08 09:14:17 +00:00
std::vector<zkutil::ZooKeeper::FutureGet> futures;
futures.reserve(hosts.size());
host_ids.reserve(hosts.size());
for (const auto & host : hosts)
futures.emplace_back(zookeeper->asyncTryGet(zookeeper_path + "/replicas/" + host));
success = true;
for (auto & future : futures)
{
auto res = future.get();
if (res.error != Coordination::Error::ZOK)
success = false;
host_ids.emplace_back(res.data);
}
zookeeper->get(zookeeper_path + "/replicas", &stat);
2021-03-15 08:22:42 +00:00
if (cversion != stat.cversion)
success = false;
if (success)
2021-02-08 09:14:17 +00:00
break;
}
if (!success)
2021-02-19 23:41:58 +00:00
throw Exception(ErrorCodes::ALL_CONNECTION_TRIES_FAILED, "Cannot get consistent cluster snapshot,"
"because replicas are created or removed concurrently");
2021-02-08 09:14:17 +00:00
assert(!hosts.empty());
assert(hosts.size() == host_ids.size());
String current_shard = parseFullReplicaName(hosts.front()).first;
std::vector<Strings> shards;
shards.emplace_back();
for (size_t i = 0; i < hosts.size(); ++i)
{
const auto & id = host_ids[i];
if (id == DROPPED_MARK)
continue;
auto [shard, replica] = parseFullReplicaName(hosts[i]);
2021-03-09 21:41:04 +00:00
auto pos = id.rfind(':');
String host_port = id.substr(0, pos);
2021-02-08 09:14:17 +00:00
if (shard != current_shard)
{
current_shard = shard;
if (!shards.back().empty())
shards.emplace_back();
}
2021-03-09 21:41:04 +00:00
shards.back().emplace_back(unescapeForFileName(host_port));
2021-02-08 09:14:17 +00:00
}
UInt16 default_port = getContext()->getTCPPort();
2021-02-08 09:14:17 +00:00
bool treat_local_as_remote = false;
bool treat_local_port_as_remote = getContext()->getApplicationType() == Context::ApplicationType::LOCAL;
return std::make_shared<Cluster>(
getContext()->getSettingsRef(),
shards,
cluster_auth_info.cluster_username,
cluster_auth_info.cluster_password,
default_port,
treat_local_as_remote,
treat_local_port_as_remote,
cluster_auth_info.cluster_secure_connection,
/*priority=*/1,
2022-06-28 14:25:29 +00:00
TSA_SUPPRESS_WARNING_FOR_READ(database_name), /// FIXME
cluster_auth_info.cluster_secret);
}
void DatabaseReplicated::fillClusterAuthInfo(String collection_name, const Poco::Util::AbstractConfiguration & config_ref)
{
const auto & config_prefix = fmt::format("named_collections.{}", collection_name);
if (!config_ref.has(config_prefix))
throw Exception(ErrorCodes::BAD_ARGUMENTS, "There is no collection named `{}` in config", collection_name);
cluster_auth_info.cluster_username = config_ref.getString(config_prefix + ".cluster_username", "");
cluster_auth_info.cluster_password = config_ref.getString(config_prefix + ".cluster_password", "");
cluster_auth_info.cluster_secret = config_ref.getString(config_prefix + ".cluster_secret", "");
cluster_auth_info.cluster_secure_connection = config_ref.getBool(config_prefix + ".cluster_secure_connection", false);
2021-02-08 09:14:17 +00:00
}
void DatabaseReplicated::tryConnectToZooKeeperAndInitDatabase(bool force_attach, bool is_create_query)
2021-02-15 00:04:46 +00:00
{
try
{
if (!getContext()->hasZooKeeper())
2021-02-15 00:04:46 +00:00
{
throw Exception("Can't create replicated database without ZooKeeper", ErrorCodes::NO_ZOOKEEPER);
}
auto current_zookeeper = getContext()->getZooKeeper();
2021-02-15 00:04:46 +00:00
if (!current_zookeeper->exists(zookeeper_path))
{
/// Create new database, multiple nodes can execute it concurrently
createDatabaseNodesInZooKeeper(current_zookeeper);
}
replica_path = fs::path(zookeeper_path) / "replicas" / getFullReplicaName();
2021-02-15 00:04:46 +00:00
String replica_host_id;
if (current_zookeeper->tryGet(replica_path, replica_host_id))
{
if (replica_host_id == DROPPED_MARK && !is_create_query)
{
LOG_WARNING(log, "Database {} exists locally, but marked dropped in ZooKeeper {}. "
"Will not try to start it up", getDatabaseName(), replica_path);
is_probably_dropped = true;
return;
}
String host_id = getHostID(getContext(), db_uuid);
if (is_create_query || replica_host_id != host_id)
{
throw Exception(
ErrorCodes::REPLICA_IS_ALREADY_EXIST,
"Replica {} of shard {} of replicated database at {} already exists. Replica host ID: '{}', current host ID: '{}'",
replica_name, shard_name, zookeeper_path, replica_host_id, host_id);
}
2021-02-15 00:04:46 +00:00
}
else if (is_create_query)
2021-02-15 00:04:46 +00:00
{
/// Create new replica. Throws if replica with the same name already exists
2021-02-15 00:04:46 +00:00
createReplicaNodesInZooKeeper(current_zookeeper);
}
else
{
/// It's not CREATE query, but replica does not exist. Probably it was dropped.
/// Do not create anything, continue as readonly.
LOG_WARNING(log, "Database {} exists locally, but its replica does not exist in ZooKeeper {}. "
"Assuming it was dropped, will not try to start it up", getDatabaseName(), replica_path);
is_probably_dropped = true;
return;
}
2021-02-15 00:04:46 +00:00
is_readonly = false;
}
2021-02-15 20:00:59 +00:00
catch (...)
2021-02-15 00:04:46 +00:00
{
if (!force_attach)
throw;
2021-02-15 20:00:59 +00:00
/// It's server startup, ignore error.
/// Worker thread will try to setup ZooKeeper connection
2021-02-15 00:04:46 +00:00
tryLogCurrentException(log);
}
}
2020-11-13 18:35:45 +00:00
bool DatabaseReplicated::createDatabaseNodesInZooKeeper(const zkutil::ZooKeeperPtr & current_zookeeper)
{
current_zookeeper->createAncestors(zookeeper_path);
2020-06-20 15:39:58 +00:00
2020-11-13 18:35:45 +00:00
Coordination::Requests ops;
ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path, "", zkutil::CreateMode::Persistent));
ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/log", "", zkutil::CreateMode::Persistent));
2020-11-19 10:34:45 +00:00
ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/replicas", "", zkutil::CreateMode::Persistent));
2020-11-20 16:06:27 +00:00
ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/counter", "", zkutil::CreateMode::Persistent));
2021-02-19 23:41:58 +00:00
/// We create and remove counter/cnt- node to increment sequential number of counter/ node and make log entry numbers start from 1.
/// New replicas are created with log pointer equal to 0 and log pointer is a number of the last executed entry.
/// It means that we cannot have log entry with number 0.
2020-11-27 14:04:03 +00:00
ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/counter/cnt-", "", zkutil::CreateMode::Persistent));
ops.emplace_back(zkutil::makeRemoveRequest(zookeeper_path + "/counter/cnt-", -1));
2020-11-19 10:34:45 +00:00
ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/metadata", "", zkutil::CreateMode::Persistent));
2020-11-27 14:04:03 +00:00
ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/max_log_ptr", "1", zkutil::CreateMode::Persistent));
2021-02-04 19:41:44 +00:00
ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/logs_to_keep", "1000", zkutil::CreateMode::Persistent));
2020-11-13 18:35:45 +00:00
Coordination::Responses responses;
auto res = current_zookeeper->tryMulti(ops, responses);
if (res == Coordination::Error::ZOK)
2021-02-15 20:00:59 +00:00
return true; /// Created new database (it's the first replica)
2020-11-13 18:35:45 +00:00
if (res == Coordination::Error::ZNODEEXISTS)
2021-02-15 20:00:59 +00:00
return false; /// Database exists, we will add new replica
2020-11-13 18:35:45 +00:00
2021-02-15 20:00:59 +00:00
/// Other codes are unexpected, will throw
2020-11-13 18:35:45 +00:00
zkutil::KeeperMultiException::check(res, ops, responses);
2022-06-29 14:27:21 +00:00
chassert(false);
2020-11-27 14:04:03 +00:00
__builtin_unreachable();
2020-05-11 12:55:17 +00:00
}
2021-05-31 13:31:03 +00:00
void DatabaseReplicated::createEmptyLogEntry(const ZooKeeperPtr & current_zookeeper)
2020-10-22 15:08:00 +00:00
{
2020-11-19 10:34:45 +00:00
/// On replica creation add empty entry to log. Can be used to trigger some actions on other replicas (e.g. update cluster info).
2021-02-19 23:41:58 +00:00
DDLLogEntry entry{};
2021-05-31 13:31:03 +00:00
DatabaseReplicatedDDLWorker::enqueueQueryImpl(current_zookeeper, entry, this, true);
}
bool DatabaseReplicated::waitForReplicaToProcessAllEntries(UInt64 timeout_ms)
{
return ddl_worker->waitForReplicaToProcessAllEntries(timeout_ms);
}
void DatabaseReplicated::createReplicaNodesInZooKeeper(const zkutil::ZooKeeperPtr & current_zookeeper)
{
/// Write host name to replica_path, it will protect from multiple replicas with the same name
auto host_id = getHostID(getContext(), db_uuid);
for (int attempts = 10; attempts > 0; --attempts)
{
Coordination::Stat stat;
String max_log_ptr_str = current_zookeeper->get(zookeeper_path + "/max_log_ptr", &stat);
Coordination::Requests ops;
ops.emplace_back(zkutil::makeCreateRequest(replica_path, host_id, zkutil::CreateMode::Persistent));
ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/log_ptr", "0", zkutil::CreateMode::Persistent));
/// In addition to creating the replica nodes, we record the max_log_ptr at the instant where
/// we declared ourself as an existing replica. We'll need this during recoverLostReplica to
/// notify other nodes that issued new queries while this node was recovering.
ops.emplace_back(zkutil::makeCheckRequest(zookeeper_path + "/max_log_ptr", stat.version));
Coordination::Responses responses;
const auto code = current_zookeeper->tryMulti(ops, responses);
if (code == Coordination::Error::ZOK)
{
max_log_ptr_at_creation = parse<UInt32>(max_log_ptr_str);
break;
}
else if (code == Coordination::Error::ZNODEEXISTS || attempts == 1)
{
/// If its our last attempt, or if the replica already exists, fail immediately.
zkutil::KeeperMultiException::check(code, ops, responses);
}
}
2021-05-31 13:31:03 +00:00
createEmptyLogEntry(current_zookeeper);
2020-06-20 15:39:58 +00:00
}
2021-09-13 19:11:16 +00:00
void DatabaseReplicated::beforeLoadingMetadata(ContextMutablePtr /*context*/, bool /*force_restore*/, bool force_attach)
2020-10-27 09:19:45 +00:00
{
tryConnectToZooKeeperAndInitDatabase(force_attach, /* is_create_query */ !force_attach);
2021-08-31 08:53:48 +00:00
}
2021-02-15 00:04:46 +00:00
2021-08-31 08:53:48 +00:00
void DatabaseReplicated::loadStoredObjects(
2021-09-13 19:11:16 +00:00
ContextMutablePtr local_context, bool force_restore, bool force_attach, bool skip_startup_tables)
2021-08-31 08:53:48 +00:00
{
2021-09-13 19:11:16 +00:00
beforeLoadingMetadata(local_context, force_restore, force_attach);
DatabaseAtomic::loadStoredObjects(local_context, force_restore, force_attach, skip_startup_tables);
2021-08-31 08:53:48 +00:00
}
2020-11-13 18:35:45 +00:00
2021-08-31 08:53:48 +00:00
void DatabaseReplicated::startupTables(ThreadPool & thread_pool, bool force_restore, bool force_attach)
{
DatabaseAtomic::startupTables(thread_pool, force_restore, force_attach);
ddl_worker = std::make_unique<DatabaseReplicatedDDLWorker>(this, getContext());
if (is_probably_dropped)
return;
2020-12-01 17:20:42 +00:00
ddl_worker->startup();
2020-10-27 09:19:45 +00:00
}
void DatabaseReplicated::checkQueryValid(const ASTPtr & query, ContextPtr query_context) const
2020-10-22 15:08:00 +00:00
{
2021-02-15 20:00:59 +00:00
/// Replicas will set correct name of current database in query context (database name can be different on replicas)
2021-03-16 20:01:20 +00:00
if (auto * ddl_query = dynamic_cast<ASTQueryWithTableAndOutput *>(query.get()))
{
if (ddl_query->getDatabase() != getDatabaseName())
throw Exception(ErrorCodes::UNKNOWN_DATABASE, "Database was renamed");
ddl_query->database.reset();
2021-03-16 20:01:20 +00:00
if (auto * create = query->as<ASTCreateQuery>())
{
bool replicated_table = create->storage && create->storage->engine && startsWith(create->storage->engine->name, "Replicated");
if (!replicated_table || !create->storage->engine->arguments)
return;
2021-06-01 10:24:06 +00:00
ASTs & args_ref = create->storage->engine->arguments->children;
ASTs args = args_ref;
2021-03-16 20:01:20 +00:00
if (args.size() < 2)
return;
2021-06-01 10:24:06 +00:00
/// It can be a constant expression. Try to evaluate it, ignore exception if we cannot.
2021-06-01 12:15:13 +00:00
bool has_expression_argument = args_ref[0]->as<ASTFunction>() || args_ref[1]->as<ASTFunction>();
2021-06-01 10:24:06 +00:00
if (has_expression_argument)
{
try
{
args[0] = evaluateConstantExpressionAsLiteral(args_ref[0]->clone(), query_context);
args[1] = evaluateConstantExpressionAsLiteral(args_ref[1]->clone(), query_context);
}
catch (...)
{
}
}
2021-03-16 20:01:20 +00:00
ASTLiteral * arg1 = args[0]->as<ASTLiteral>();
ASTLiteral * arg2 = args[1]->as<ASTLiteral>();
if (!arg1 || !arg2 || arg1->value.getType() != Field::Types::String || arg2->value.getType() != Field::Types::String)
return;
String maybe_path = arg1->value.get<String>();
String maybe_replica = arg2->value.get<String>();
/// Looks like it's ReplicatedMergeTree with explicit zookeeper_path and replica_name arguments.
/// Let's ensure that some macros are used.
/// NOTE: we cannot check here that substituted values will be actually different on shards and replicas.
Macros::MacroExpansionInfo info;
info.table_id = {getDatabaseName(), create->getTable(), create->uuid};
2022-04-13 14:51:59 +00:00
info.shard = getShardName();
info.replica = getReplicaName();
query_context->getMacros()->expand(maybe_path, info);
2021-03-16 20:01:20 +00:00
bool maybe_shard_macros = info.expanded_other;
info.expanded_other = false;
query_context->getMacros()->expand(maybe_replica, info);
2021-03-16 20:01:20 +00:00
bool maybe_replica_macros = info.expanded_other;
bool enable_functional_tests_helper = getContext()->getConfigRef().has("_functional_tests_helper_database_replicated_replace_args_macros");
2021-03-16 20:01:20 +00:00
2021-03-18 12:49:31 +00:00
if (!enable_functional_tests_helper)
LOG_WARNING(log, "It's not recommended to explicitly specify zookeeper_path and replica_name in ReplicatedMergeTree arguments");
if (maybe_shard_macros && maybe_replica_macros)
return;
2021-06-01 10:24:06 +00:00
if (enable_functional_tests_helper && !has_expression_argument)
2021-03-16 20:01:20 +00:00
{
if (maybe_path.empty() || maybe_path.back() != '/')
maybe_path += '/';
2021-06-01 10:24:06 +00:00
args_ref[0]->as<ASTLiteral>()->value = maybe_path + "auto_{shard}";
args_ref[1]->as<ASTLiteral>()->value = maybe_replica + "auto_{replica}";
2021-03-16 20:01:20 +00:00
return;
}
throw Exception(ErrorCodes::INCORRECT_QUERY,
"Explicit zookeeper_path and replica_name are specified in ReplicatedMergeTree arguments. "
"If you really want to specify it explicitly, then you should use some macros "
"to distinguish different shards and replicas");
}
}
2021-02-08 09:46:30 +00:00
if (const auto * query_alter = query->as<ASTAlterQuery>())
2020-06-27 13:39:41 +00:00
{
for (const auto & command : query_alter->command_list->children)
{
if (!isSupportedAlterType(command->as<ASTAlterCommand&>().type))
throw Exception("Unsupported type of ALTER query", ErrorCodes::NOT_IMPLEMENTED);
}
2020-06-27 13:39:41 +00:00
}
if (auto * query_drop = query->as<ASTDropQuery>())
{
if (query_drop->kind == ASTDropQuery::Kind::Detach && query_context->getSettingsRef().database_replicated_always_detach_permanently)
query_drop->permanently = true;
if (query_drop->kind == ASTDropQuery::Kind::Detach && !query_drop->permanently)
throw Exception(ErrorCodes::INCORRECT_QUERY, "DETACH TABLE is not allowed for Replicated databases. "
"Use DETACH TABLE PERMANENTLY or SYSTEM RESTART REPLICA or set "
"database_replicated_always_detach_permanently to 1");
}
2021-03-16 20:01:20 +00:00
}
BlockIO DatabaseReplicated::tryEnqueueReplicatedDDL(const ASTPtr & query, ContextPtr query_context, bool internal)
2021-03-16 20:01:20 +00:00
{
if (query_context->getCurrentTransaction() && query_context->getSettingsRef().throw_on_unsupported_query_inside_transaction)
throw Exception(ErrorCodes::NOT_IMPLEMENTED, "Distributed DDL queries inside transactions are not supported");
2021-03-16 20:01:20 +00:00
if (is_readonly)
throw Exception(ErrorCodes::NO_ZOOKEEPER, "Database is in readonly mode, because it cannot connect to ZooKeeper");
if (!internal && (query_context->getClientInfo().query_kind != ClientInfo::QueryKind::INITIAL_QUERY))
2021-03-16 20:01:20 +00:00
throw Exception(ErrorCodes::INCORRECT_QUERY, "It's not initial query. ON CLUSTER is not allowed for Replicated database.");
2021-03-16 20:01:20 +00:00
checkQueryValid(query, query_context);
LOG_DEBUG(log, "Proposing query: {}", queryToString(query));
DDLLogEntry entry;
entry.query = queryToString(query);
entry.initiator = ddl_worker->getCommonHostID();
entry.setSettingsIfRequired(query_context);
2021-02-04 19:41:44 +00:00
String node_path = ddl_worker->tryEnqueueAndExecuteEntry(entry, query_context);
2021-02-08 19:36:17 +00:00
Strings hosts_to_wait = getZooKeeper()->getChildren(zookeeper_path + "/replicas");
return getDistributedDDLStatus(node_path, entry, query_context, hosts_to_wait);
}
static UUID getTableUUIDIfReplicated(const String & metadata, ContextPtr context)
2021-02-15 00:04:46 +00:00
{
bool looks_like_replicated = metadata.find("ReplicatedMergeTree") != std::string::npos;
if (!looks_like_replicated)
return UUIDHelpers::Nil;
ParserCreateQuery parser;
auto size = context->getSettingsRef().max_query_size;
auto depth = context->getSettingsRef().max_parser_depth;
2021-02-15 00:04:46 +00:00
ASTPtr query = parseQuery(parser, metadata, size, depth);
const ASTCreateQuery & create = query->as<const ASTCreateQuery &>();
if (!create.storage || !create.storage->engine)
return UUIDHelpers::Nil;
if (!startsWith(create.storage->engine->name, "Replicated") || !endsWith(create.storage->engine->name, "MergeTree"))
return UUIDHelpers::Nil;
2022-06-29 14:27:21 +00:00
chassert(create.uuid != UUIDHelpers::Nil);
2021-02-15 00:04:46 +00:00
return create.uuid;
}
2021-02-09 15:14:20 +00:00
void DatabaseReplicated::recoverLostReplica(const ZooKeeperPtr & current_zookeeper, UInt32 our_log_ptr, UInt32 max_log_ptr)
2020-10-22 15:08:00 +00:00
{
is_recovering = true;
SCOPE_EXIT({ is_recovering = false; });
2021-02-15 20:00:59 +00:00
/// Let's compare local (possibly outdated) metadata with (most actual) metadata stored in ZooKeeper
/// and try to update the set of local tables.
/// We could drop all local tables and create the new ones just like it's new replica.
/// But it will cause all ReplicatedMergeTree tables to fetch all data parts again and data in other tables will be lost.
2021-02-09 15:14:20 +00:00
bool new_replica = our_log_ptr == 0;
if (new_replica)
LOG_INFO(log, "Will create new replica from log pointer {}", max_log_ptr);
else
LOG_WARNING(log, "Will recover replica with staled log pointer {} from log pointer {}", our_log_ptr, max_log_ptr);
auto table_name_to_metadata = tryGetConsistentMetadataSnapshot(current_zookeeper, max_log_ptr);
2020-06-24 12:45:42 +00:00
2021-02-15 00:04:46 +00:00
/// For ReplicatedMergeTree tables we can compare only UUIDs to ensure that it's the same table.
/// Metadata can be different, it's handled on table replication level.
2021-02-15 20:00:59 +00:00
/// We need to handle renamed tables only.
2021-02-15 00:04:46 +00:00
/// TODO maybe we should also update MergeTree SETTINGS if required?
std::unordered_map<UUID, String> zk_replicated_id_to_name;
for (const auto & zk_table : table_name_to_metadata)
{
UUID zk_replicated_id = getTableUUIDIfReplicated(zk_table.second, getContext());
2021-02-15 00:04:46 +00:00
if (zk_replicated_id != UUIDHelpers::Nil)
zk_replicated_id_to_name.emplace(zk_replicated_id, zk_table.first);
}
2021-02-15 20:00:59 +00:00
/// We will drop or move tables which exist only in local metadata
Strings tables_to_detach;
2021-02-15 00:04:46 +00:00
std::vector<std::pair<String, String>> replicated_tables_to_rename;
size_t total_tables = 0;
2021-02-15 00:04:46 +00:00
std::vector<UUID> replicated_ids;
for (auto existing_tables_it = getTablesIterator(getContext(), {}); existing_tables_it->isValid();
existing_tables_it->next(), ++total_tables)
{
String name = existing_tables_it->name();
2021-02-15 00:04:46 +00:00
UUID local_replicated_id = UUIDHelpers::Nil;
if (existing_tables_it->table()->supportsReplication())
{
2021-02-15 20:00:59 +00:00
/// Check if replicated tables have the same UUID
2021-02-15 00:04:46 +00:00
local_replicated_id = existing_tables_it->table()->getStorageID().uuid;
auto it = zk_replicated_id_to_name.find(local_replicated_id);
if (it != zk_replicated_id_to_name.end())
{
2021-02-15 00:04:46 +00:00
if (name != it->second)
2021-02-15 20:00:59 +00:00
{
/// Need just update table name
2021-02-15 00:04:46 +00:00
replicated_tables_to_rename.emplace_back(name, it->second);
2021-02-15 20:00:59 +00:00
}
2021-02-15 00:04:46 +00:00
continue;
}
2021-02-15 00:04:46 +00:00
}
2021-02-15 00:04:46 +00:00
auto in_zk = table_name_to_metadata.find(name);
if (in_zk == table_name_to_metadata.end() || in_zk->second != readMetadataFile(name))
{
2021-02-15 20:00:59 +00:00
/// Local table does not exits in ZooKeeper or has different metadata
tables_to_detach.emplace_back(std::move(name));
}
}
2021-07-30 16:34:18 +00:00
auto make_query_context = [this, current_zookeeper]()
{
auto query_context = Context::createCopy(getContext());
query_context->makeQueryContext();
query_context->getClientInfo().query_kind = ClientInfo::QueryKind::SECONDARY_QUERY;
query_context->getClientInfo().is_replicated_database_internal = true;
2022-06-27 20:48:27 +00:00
query_context->setCurrentDatabase(getDatabaseName());
2021-07-30 16:34:18 +00:00
query_context->setCurrentQueryId("");
2021-08-03 13:28:03 +00:00
auto txn = std::make_shared<ZooKeeperMetadataTransaction>(current_zookeeper, zookeeper_path, false, "");
2021-07-30 16:34:18 +00:00
query_context->initZooKeeperMetadataTransaction(txn);
return query_context;
};
2021-02-11 22:23:40 +00:00
String db_name = getDatabaseName();
String to_db_name = getDatabaseName() + BROKEN_TABLES_SUFFIX;
String to_db_name_replicated = getDatabaseName() + BROKEN_REPLICATED_TABLES_SUFFIX;
2021-02-19 23:41:58 +00:00
if (total_tables * db_settings.max_broken_tables_ratio < tables_to_detach.size())
2021-02-11 22:23:40 +00:00
throw Exception(ErrorCodes::DATABASE_REPLICATION_FAILED, "Too many tables to recreate: {} of {}", tables_to_detach.size(), total_tables);
else if (!tables_to_detach.empty())
2021-02-11 22:23:40 +00:00
{
LOG_WARNING(log, "Will recreate {} broken tables to recover replica", tables_to_detach.size());
/// It's too dangerous to automatically drop tables, so we will move them to special database.
/// We use Ordinary engine for destination database, because it's the only way to discard table UUID
/// and make possible creation of new table with the same UUID.
String query = fmt::format("CREATE DATABASE IF NOT EXISTS {} ENGINE=Ordinary", backQuoteIfNeed(to_db_name));
auto query_context = Context::createCopy(getContext());
2022-06-27 14:10:00 +00:00
query_context->setSetting("allow_deprecated_database_ordinary", 1);
2021-02-11 22:23:40 +00:00
executeQuery(query, query_context, true);
/// But we want to avoid discarding UUID of ReplicatedMergeTree tables, because it will not work
/// if zookeeper_path contains {uuid} macro. Replicated database do not recreate replicated tables on recovery,
/// so it's ok to save UUID of replicated table.
query = fmt::format("CREATE DATABASE IF NOT EXISTS {} ENGINE=Atomic", backQuoteIfNeed(to_db_name_replicated));
query_context = Context::createCopy(getContext());
executeQuery(query, query_context, true);
2021-02-11 22:23:40 +00:00
}
2021-02-11 22:23:40 +00:00
size_t moved_tables = 0;
std::vector<UUID> dropped_tables;
2021-04-26 10:20:57 +00:00
size_t dropped_dictionaries = 0;
for (const auto & table_name : tables_to_detach)
{
2021-02-11 22:23:40 +00:00
DDLGuardPtr table_guard = DatabaseCatalog::instance().getDDLGuard(db_name, table_name);
if (getDatabaseName() != db_name)
throw Exception(ErrorCodes::UNKNOWN_DATABASE, "Database was renamed, will retry");
auto table = tryGetTable(table_name, getContext());
2021-04-21 13:45:13 +00:00
auto move_table_to_database = [&](const String & broken_table_name, const String & to_database_name)
{
/// Table probably stores some data. Let's move it to another database.
String to_name = fmt::format("{}_{}_{}", broken_table_name, max_log_ptr, thread_local_rng() % 1000);
LOG_DEBUG(log, "Will RENAME TABLE {} TO {}.{}", backQuoteIfNeed(broken_table_name), backQuoteIfNeed(to_database_name), backQuoteIfNeed(to_name));
assert(db_name < to_database_name);
DDLGuardPtr to_table_guard = DatabaseCatalog::instance().getDDLGuard(to_database_name, to_name);
auto to_db_ptr = DatabaseCatalog::instance().getDatabase(to_database_name);
DatabaseAtomic::renameTable(make_query_context(), broken_table_name, *to_db_ptr, to_name, false, false);
++moved_tables;
};
2021-04-21 13:45:13 +00:00
if (!table->storesDataOnDisk())
2021-02-11 22:23:40 +00:00
{
LOG_DEBUG(log, "Will DROP TABLE {}, because it does not store data on disk and can be safely dropped", backQuoteIfNeed(table_name));
dropped_tables.push_back(tryGetTableUUID(table_name));
2021-04-26 10:20:57 +00:00
dropped_dictionaries += table->isDictionary();
2021-05-13 07:03:00 +00:00
table->flushAndShutdown();
2021-07-30 16:34:18 +00:00
DatabaseAtomic::dropTable(make_query_context(), table_name, true);
}
else if (!table->supportsReplication())
{
move_table_to_database(table_name, to_db_name);
}
else
{
move_table_to_database(table_name, to_db_name_replicated);
}
}
2021-02-11 22:23:40 +00:00
if (!tables_to_detach.empty())
2021-04-26 10:20:57 +00:00
LOG_WARNING(log, "Cleaned {} outdated objects: dropped {} dictionaries and {} tables, moved {} tables",
tables_to_detach.size(), dropped_dictionaries, dropped_tables.size() - dropped_dictionaries, moved_tables);
2021-02-11 22:23:40 +00:00
2021-02-15 00:04:46 +00:00
/// Now database is cleared from outdated tables, let's rename ReplicatedMergeTree tables to actual names
for (const auto & old_to_new : replicated_tables_to_rename)
{
const String & from = old_to_new.first;
const String & to = old_to_new.second;
LOG_DEBUG(log, "Will RENAME TABLE {} TO {}", backQuoteIfNeed(from), backQuoteIfNeed(to));
/// TODO Maybe we should do it in two steps: rename all tables to temporary names and then rename them to actual names?
DDLGuardPtr table_guard = DatabaseCatalog::instance().getDDLGuard(db_name, std::min(from, to));
DDLGuardPtr to_table_guard = DatabaseCatalog::instance().getDDLGuard(db_name, std::max(from, to));
2021-07-30 16:34:18 +00:00
DatabaseAtomic::renameTable(make_query_context(), from, *this, to, false, false);
2021-02-15 00:04:46 +00:00
}
2021-02-11 22:23:40 +00:00
for (const auto & id : dropped_tables)
DatabaseCatalog::instance().waitTableFinallyDropped(id);
2021-02-09 15:14:20 +00:00
for (const auto & name_and_meta : table_name_to_metadata)
2020-10-22 15:08:00 +00:00
{
if (isTableExist(name_and_meta.first, getContext()))
{
assert(name_and_meta.second == readMetadataFile(name_and_meta.first));
continue;
}
2021-02-09 15:14:20 +00:00
auto query_ast = parseQueryFromMetadataInZooKeeper(name_and_meta.first, name_and_meta.second);
2020-11-27 14:04:03 +00:00
LOG_INFO(log, "Executing {}", serializeAST(*query_ast));
2021-07-30 16:34:18 +00:00
auto create_query_context = make_query_context();
InterpreterCreateQuery(query_ast, create_query_context).execute();
2020-11-19 10:34:45 +00:00
}
if (max_log_ptr_at_creation != 0)
{
/// If the replica is new and some of the queries applied during recovery
/// where issued after the replica was created, then other nodes might be
/// waiting for this node to notify them that the query was applied.
for (UInt32 ptr = max_log_ptr_at_creation; ptr <= max_log_ptr; ++ptr)
{
auto entry_name = DDLTaskBase::getLogEntryName(ptr);
auto path = fs::path(zookeeper_path) / "log" / entry_name / "finished" / getFullReplicaName();
auto status = ExecutionStatus(0).serializeText();
auto res = current_zookeeper->tryCreate(path, status, zkutil::CreateMode::Persistent);
if (res == Coordination::Error::ZOK)
LOG_INFO(log, "Marked recovered {} as finished", entry_name);
}
}
2021-02-09 15:14:20 +00:00
}
std::map<String, String> DatabaseReplicated::tryGetConsistentMetadataSnapshot(const ZooKeeperPtr & zookeeper, UInt32 & max_log_ptr)
{
std::map<String, String> table_name_to_metadata;
constexpr int max_retries = 10;
int iteration = 0;
while (++iteration <= max_retries)
{
table_name_to_metadata.clear();
LOG_DEBUG(log, "Trying to get consistent metadata snapshot for log pointer {}", max_log_ptr);
Strings table_names = zookeeper->getChildren(zookeeper_path + "/metadata");
std::vector<zkutil::ZooKeeper::FutureGet> futures;
futures.reserve(table_names.size());
for (const auto & table : table_names)
futures.emplace_back(zookeeper->asyncTryGet(zookeeper_path + "/metadata/" + table));
for (size_t i = 0; i < table_names.size(); ++i)
{
auto res = futures[i].get();
if (res.error != Coordination::Error::ZOK)
break;
table_name_to_metadata.emplace(unescapeForFileName(table_names[i]), res.data);
2021-02-09 15:14:20 +00:00
}
UInt32 new_max_log_ptr = parse<UInt32>(zookeeper->get(zookeeper_path + "/max_log_ptr"));
if (new_max_log_ptr == max_log_ptr && table_names.size() == table_name_to_metadata.size())
break;
if (max_log_ptr < new_max_log_ptr)
{
LOG_DEBUG(log, "Log pointer moved from {} to {}, will retry", max_log_ptr, new_max_log_ptr);
max_log_ptr = new_max_log_ptr;
}
else
{
2022-06-29 14:27:21 +00:00
chassert(max_log_ptr == new_max_log_ptr);
chassert(table_names.size() != table_name_to_metadata.size());
2021-02-09 15:14:20 +00:00
LOG_DEBUG(log, "Cannot get metadata of some tables due to ZooKeeper error, will retry");
}
}
if (max_retries < iteration)
throw Exception(ErrorCodes::DATABASE_REPLICATION_FAILED, "Cannot get consistent metadata snapshot");
LOG_DEBUG(log, "Got consistent metadata snapshot for log pointer {}", max_log_ptr);
return table_name_to_metadata;
2020-06-20 15:39:58 +00:00
}
2020-11-27 14:04:03 +00:00
ASTPtr DatabaseReplicated::parseQueryFromMetadataInZooKeeper(const String & node_name, const String & query)
{
ParserCreateQuery parser;
String description = "in ZooKeeper " + zookeeper_path + "/metadata/" + node_name;
auto ast = parseQuery(parser, query, description, 0, getContext()->getSettingsRef().max_parser_depth);
2020-11-27 14:04:03 +00:00
auto & create = ast->as<ASTCreateQuery &>();
2021-11-11 13:28:18 +00:00
if (create.uuid == UUIDHelpers::Nil || create.getTable() != TABLE_WITH_UUID_NAME_PLACEHOLDER || create.database)
2020-11-27 14:04:03 +00:00
throw Exception(ErrorCodes::LOGICAL_ERROR, "Got unexpected query from {}: {}", node_name, query);
2021-03-08 17:26:38 +00:00
bool is_materialized_view_with_inner_table = create.is_materialized_view && create.to_table_id.empty();
create.setDatabase(getDatabaseName());
create.setTable(unescapeForFileName(node_name));
2021-03-08 17:26:38 +00:00
create.attach = is_materialized_view_with_inner_table;
2020-11-27 14:04:03 +00:00
return ast;
}
void DatabaseReplicated::drop(ContextPtr context_)
2020-06-20 15:39:58 +00:00
{
2020-11-13 18:35:45 +00:00
auto current_zookeeper = getZooKeeper();
2021-05-31 13:31:03 +00:00
current_zookeeper->set(replica_path, DROPPED_MARK, -1);
createEmptyLogEntry(current_zookeeper);
2020-06-20 15:39:58 +00:00
DatabaseAtomic::drop(context_);
2021-02-04 19:41:44 +00:00
current_zookeeper->tryRemoveRecursive(replica_path);
2021-02-08 09:14:17 +00:00
/// TODO it may leave garbage in ZooKeeper if the last node lost connection here
if (current_zookeeper->tryRemove(zookeeper_path + "/replicas") == Coordination::Error::ZOK)
{
/// It was the last replica, remove all metadata
current_zookeeper->tryRemoveRecursive(zookeeper_path);
}
2020-05-24 17:13:53 +00:00
}
2021-02-04 19:41:44 +00:00
void DatabaseReplicated::stopReplication()
2020-11-13 18:35:45 +00:00
{
if (ddl_worker)
ddl_worker->shutdown();
2021-02-04 19:41:44 +00:00
}
void DatabaseReplicated::shutdown()
{
stopReplication();
ddl_worker = nullptr;
2020-11-13 18:35:45 +00:00
DatabaseAtomic::shutdown();
}
2020-11-29 11:45:32 +00:00
2022-06-23 07:59:13 +00:00
void DatabaseReplicated::dropTable(ContextPtr local_context, const String & table_name, bool sync)
2020-11-29 11:45:32 +00:00
{
auto txn = local_context->getZooKeeperMetadataTransaction();
2021-03-08 17:26:38 +00:00
assert(!ddl_worker->isCurrentlyActive() || txn || startsWith(table_name, ".inner_id."));
2021-02-19 23:41:58 +00:00
if (txn && txn->isInitialQuery())
2020-11-29 11:45:32 +00:00
{
String metadata_zk_path = zookeeper_path + "/metadata/" + escapeForFileName(table_name);
2021-02-19 23:41:58 +00:00
txn->addOp(zkutil::makeRemoveRequest(metadata_zk_path, -1));
2020-11-29 11:45:32 +00:00
}
2022-06-23 07:59:13 +00:00
DatabaseAtomic::dropTable(local_context, table_name, sync);
2020-11-29 11:45:32 +00:00
}
void DatabaseReplicated::renameTable(ContextPtr local_context, const String & table_name, IDatabase & to_database,
2020-11-29 11:45:32 +00:00
const String & to_table_name, bool exchange, bool dictionary)
{
auto txn = local_context->getZooKeeperMetadataTransaction();
2020-11-29 11:45:32 +00:00
assert(txn);
2021-02-19 23:41:58 +00:00
if (txn->isInitialQuery())
2020-11-29 11:45:32 +00:00
{
2021-02-08 09:14:17 +00:00
if (this != &to_database)
throw Exception(ErrorCodes::NOT_IMPLEMENTED, "Moving tables between databases is not supported for Replicated engine");
2021-02-08 19:36:17 +00:00
if (table_name == to_table_name)
throw Exception(ErrorCodes::INCORRECT_QUERY, "Cannot rename table to itself");
if (!isTableExist(table_name, local_context))
2021-02-04 19:41:44 +00:00
throw Exception(ErrorCodes::UNKNOWN_TABLE, "Table {} does not exist", table_name);
if (exchange && !to_database.isTableExist(to_table_name, local_context))
2021-02-04 19:41:44 +00:00
throw Exception(ErrorCodes::UNKNOWN_TABLE, "Table {} does not exist", to_table_name);
String statement = readMetadataFile(table_name);
2021-02-19 23:41:58 +00:00
String metadata_zk_path = zookeeper_path + "/metadata/" + escapeForFileName(table_name);
String metadata_zk_path_to = zookeeper_path + "/metadata/" + escapeForFileName(to_table_name);
txn->addOp(zkutil::makeRemoveRequest(metadata_zk_path, -1));
2020-11-29 11:45:32 +00:00
if (exchange)
{
2021-02-11 15:14:38 +00:00
String statement_to = readMetadataFile(to_table_name);
2021-02-19 23:41:58 +00:00
txn->addOp(zkutil::makeRemoveRequest(metadata_zk_path_to, -1));
txn->addOp(zkutil::makeCreateRequest(metadata_zk_path, statement_to, zkutil::CreateMode::Persistent));
2020-11-29 11:45:32 +00:00
}
2021-02-19 23:41:58 +00:00
txn->addOp(zkutil::makeCreateRequest(metadata_zk_path_to, statement, zkutil::CreateMode::Persistent));
2020-11-29 11:45:32 +00:00
}
DatabaseAtomic::renameTable(local_context, table_name, to_database, to_table_name, exchange, dictionary);
2020-11-29 11:45:32 +00:00
}
void DatabaseReplicated::commitCreateTable(const ASTCreateQuery & query, const StoragePtr & table,
const String & table_metadata_tmp_path, const String & table_metadata_path,
ContextPtr query_context)
2020-11-29 11:45:32 +00:00
{
auto txn = query_context->getZooKeeperMetadataTransaction();
2020-11-29 11:45:32 +00:00
assert(!ddl_worker->isCurrentlyActive() || txn);
2021-02-19 23:41:58 +00:00
if (txn && txn->isInitialQuery())
2020-11-29 11:45:32 +00:00
{
String metadata_zk_path = zookeeper_path + "/metadata/" + escapeForFileName(query.getTable());
2020-11-29 11:45:32 +00:00
String statement = getObjectDefinitionFromCreateQuery(query.clone());
/// zk::multi(...) will throw if `metadata_zk_path` exists
2021-02-19 23:41:58 +00:00
txn->addOp(zkutil::makeCreateRequest(metadata_zk_path, statement, zkutil::CreateMode::Persistent));
2020-11-29 11:45:32 +00:00
}
DatabaseAtomic::commitCreateTable(query, table, table_metadata_tmp_path, table_metadata_path, query_context);
}
void DatabaseReplicated::commitAlterTable(const StorageID & table_id,
const String & table_metadata_tmp_path, const String & table_metadata_path,
const String & statement, ContextPtr query_context)
2020-11-29 11:45:32 +00:00
{
auto txn = query_context->getZooKeeperMetadataTransaction();
2021-02-19 23:41:58 +00:00
if (txn && txn->isInitialQuery())
2020-11-29 11:45:32 +00:00
{
2021-02-19 23:41:58 +00:00
String metadata_zk_path = zookeeper_path + "/metadata/" + escapeForFileName(table_id.table_name);
txn->addOp(zkutil::makeSetRequest(metadata_zk_path, statement, -1));
2020-11-29 11:45:32 +00:00
}
DatabaseAtomic::commitAlterTable(table_id, table_metadata_tmp_path, table_metadata_path, statement, query_context);
}
void DatabaseReplicated::detachTablePermanently(ContextPtr local_context, const String & table_name)
{
auto txn = local_context->getZooKeeperMetadataTransaction();
assert(!ddl_worker->isCurrentlyActive() || txn);
2021-02-19 23:41:58 +00:00
if (txn && txn->isInitialQuery())
{
2021-03-09 00:48:37 +00:00
/// We have to remove metadata from zookeeper, because we do not distinguish permanently detached tables
/// from attached tables when recovering replica.
String metadata_zk_path = zookeeper_path + "/metadata/" + escapeForFileName(table_name);
2021-02-19 23:41:58 +00:00
txn->addOp(zkutil::makeRemoveRequest(metadata_zk_path, -1));
}
DatabaseAtomic::detachTablePermanently(local_context, table_name);
}
void DatabaseReplicated::removeDetachedPermanentlyFlag(ContextPtr local_context, const String & table_name, const String & table_metadata_path, bool attach) const
2021-03-09 00:48:37 +00:00
{
auto txn = local_context->getZooKeeperMetadataTransaction();
2021-03-09 00:48:37 +00:00
assert(!ddl_worker->isCurrentlyActive() || txn);
2021-03-09 10:24:08 +00:00
if (txn && txn->isInitialQuery() && attach)
2021-03-09 00:48:37 +00:00
{
String metadata_zk_path = zookeeper_path + "/metadata/" + escapeForFileName(table_name);
String statement = readMetadataFile(table_name);
txn->addOp(zkutil::makeCreateRequest(metadata_zk_path, statement, zkutil::CreateMode::Persistent));
}
DatabaseAtomic::removeDetachedPermanentlyFlag(local_context, table_name, table_metadata_path, attach);
2021-03-09 00:48:37 +00:00
}
String DatabaseReplicated::readMetadataFile(const String & table_name) const
{
String statement;
ReadBufferFromFile in(getObjectMetadataPath(table_name), 4096);
readStringUntilEOF(statement, in);
return statement;
}
std::vector<std::pair<ASTPtr, StoragePtr>>
DatabaseReplicated::getTablesForBackup(const FilterByNameFunction & filter, const ContextPtr &) const
{
/// Here we read metadata from ZooKeeper. We could do that by simple call of DatabaseAtomic::getTablesForBackup() however
2022-06-30 08:10:12 +00:00
/// reading from ZooKeeper is better because thus we won't be dependent on how fast the replication queue of this database is.
std::vector<std::pair<ASTPtr, StoragePtr>> res;
auto zookeeper = getContext()->getZooKeeper();
auto escaped_table_names = zookeeper->getChildren(zookeeper_path + "/metadata");
for (const auto & escaped_table_name : escaped_table_names)
{
String table_name = unescapeForFileName(escaped_table_name);
if (!filter(table_name))
continue;
String zk_metadata;
if (!zookeeper->tryGet(zookeeper_path + "/metadata/" + escaped_table_name, zk_metadata))
throw Exception(ErrorCodes::INCONSISTENT_METADATA_FOR_BACKUP, "Metadata for table {} was not found in ZooKeeper", table_name);
2022-06-30 08:10:12 +00:00
ParserCreateQuery parser;
auto create_table_query = parseQuery(parser, zk_metadata, 0, getContext()->getSettingsRef().max_parser_depth);
auto & create = create_table_query->as<ASTCreateQuery &>();
create.attach = false;
create.setTable(table_name);
create.setDatabase(getDatabaseName());
StoragePtr storage;
if (create.uuid != UUIDHelpers::Nil)
{
storage = DatabaseCatalog::instance().tryGetByUUID(create.uuid).second;
if (storage)
storage->adjustCreateQueryForBackup(create_table_query);
}
res.emplace_back(create_table_query, storage);
}
return res;
}
2022-06-22 22:56:41 +00:00
void DatabaseReplicated::createTableRestoredFromBackup(
const ASTPtr & create_table_query,
ContextMutablePtr local_context,
std::shared_ptr<IRestoreCoordination> restore_coordination,
UInt64 timeout_ms)
{
/// Because of the replication multiple nodes can try to restore the same tables again and failed with "Table already exists"
/// because of some table could be restored already on other node and then replicated to this node.
/// To solve this problem we use the restore coordination: the first node calls
/// IRestoreCoordination::acquireCreatingTableInReplicatedDatabase() and then for other nodes this function returns false which means
/// this table is already being created by some other node.
String table_name = create_table_query->as<const ASTCreateQuery &>().getTable();
2022-06-22 22:56:41 +00:00
if (restore_coordination->acquireCreatingTableInReplicatedDatabase(getZooKeeperPath(), table_name))
{
2022-06-22 22:56:41 +00:00
DatabaseAtomic::createTableRestoredFromBackup(create_table_query, local_context, restore_coordination, timeout_ms);
}
/// Wait until the table is actually created no matter if it's created by the current or another node and replicated to the
/// current node afterwards. We have to wait because `RestorerFromBackup` is going to restore data of the table then.
/// TODO: The following code doesn't look very reliable, probably we need to rewrite it somehow.
2022-06-22 22:56:41 +00:00
auto timeout = std::chrono::milliseconds{timeout_ms};
auto start_time = std::chrono::steady_clock::now();
2022-06-22 22:56:41 +00:00
while (!isTableExist(table_name, local_context))
{
waitForReplicaToProcessAllEntries(50);
2022-06-19 13:48:52 +00:00
2022-06-22 22:56:41 +00:00
auto elapsed = std::chrono::steady_clock::now() - start_time;
if (elapsed > timeout)
throw Exception(ErrorCodes::CANNOT_RESTORE_TABLE,
"Couldn't restore table {}.{} on other node or sync it (elapsed {})",
backQuoteIfNeed(getDatabaseName()), backQuoteIfNeed(table_name), to_string(elapsed));
}
}
2020-04-05 12:18:51 +00:00
}