thevar1able/ClickHouse
1
0
Fork 0
mirror of https://github.com/ClickHouse/ClickHouse.git synced 2024-11-17 13:13:36 +00:00
Code Issues Packages Projects Releases Wiki Activity

Correct startup logic

Browse Source
This commit is contained in:
alesapin 2020-02-14 13:17:04 +03:00
parent d9ebec472b
commit 527325ef46
4 changed files with 99 additions and 99 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

111
dbms/src/Storages/MergeTree/ReplicatedMergeTreeTableMetadata.cpp
View File

@ -150,11 +150,9 @@ ReplicatedMergeTreeTableMetadata ReplicatedMergeTreeTableMetadata::parse(const S
return metadata;
}
ReplicatedMergeTreeTableMetadata::Diff
ReplicatedMergeTreeTableMetadata::checkAndFindDiff(const ReplicatedMergeTreeTableMetadata & from_zk, bool allow_alter) const
{
Diff diff;
void ReplicatedMergeTreeTableMetadata::checkImmutableFieldsEquals(const ReplicatedMergeTreeTableMetadata & from_zk) const
{
if (data_format_version < MERGE_TREE_DATA_MIN_FORMAT_VERSION_WITH_CUSTOM_PARTITIONING)
{
if (date_column != from_zk.date_column)
@ -163,10 +161,12 @@ ReplicatedMergeTreeTableMetadata::checkAndFindDiff(const ReplicatedMergeTreeTabl
ErrorCodes::METADATA_MISMATCH);
}
else if (!from_zk.date_column.empty())
{
throw Exception(
"Existing table metadata in ZooKeeper differs in date index column."
" Stored in ZooKeeper: " + from_zk.date_column + ", local is custom-partitioned.",
ErrorCodes::METADATA_MISMATCH);
}
if (sampling_expression != from_zk.sampling_expression)
throw Exception("Existing table metadata in ZooKeeper differs in sample expression."
@ -208,63 +208,46 @@ ReplicatedMergeTreeTableMetadata::checkAndFindDiff(const ReplicatedMergeTreeTabl
" Stored in ZooKeeper: " + from_zk.partition_key + ", local: " + partition_key,
ErrorCodes::METADATA_MISMATCH);
}
void ReplicatedMergeTreeTableMetadata::checkEquals(const ReplicatedMergeTreeTableMetadata & from_zk) const
{
checkImmutableFieldsEquals(from_zk);
if (sorting_key != from_zk.sorting_key)
{
if (allow_alter)
{
diff.sorting_key_changed = true;
diff.new_sorting_key = from_zk.sorting_key;
}
else
throw Exception(
"Existing table metadata in ZooKeeper differs in sorting key expression."
" Stored in ZooKeeper: " + from_zk.sorting_key + ", local: " + sorting_key,
ErrorCodes::METADATA_MISMATCH);
throw Exception(
"Existing table metadata in ZooKeeper differs in sorting key expression."
" Stored in ZooKeeper: " + from_zk.sorting_key + ", local: " + sorting_key,
ErrorCodes::METADATA_MISMATCH);
}
if (ttl_table != from_zk.ttl_table)
{
if (allow_alter)
{
diff.ttl_table_changed = true;
diff.new_ttl_table = from_zk.ttl_table;
}
else
throw Exception(
"Existing table metadata in ZooKeeper differs in TTL."
" Stored in ZooKeeper: " + from_zk.ttl_table +
", local: " + ttl_table,
ErrorCodes::METADATA_MISMATCH);
throw Exception(
"Existing table metadata in ZooKeeper differs in TTL."
" Stored in ZooKeeper: " + from_zk.ttl_table +
", local: " + ttl_table,
ErrorCodes::METADATA_MISMATCH);
}
if (skip_indices != from_zk.skip_indices)
{
if (allow_alter)
{
diff.skip_indices_changed = true;
diff.new_skip_indices = from_zk.skip_indices;
}
else
throw Exception(
"Existing table metadata in ZooKeeper differs in skip indexes."
" Stored in ZooKeeper: " + from_zk.skip_indices +
", local: " + skip_indices,
ErrorCodes::METADATA_MISMATCH);
throw Exception(
"Existing table metadata in ZooKeeper differs in skip indexes."
" Stored in ZooKeeper: " + from_zk.skip_indices +
", local: " + skip_indices,
ErrorCodes::METADATA_MISMATCH);
}
if (constraints != from_zk.constraints)
{
if (allow_alter)
{
diff.constraints_changed = true;
diff.new_constraints = from_zk.constraints;
}
else
throw Exception(
"Existing table metadata in ZooKeeper differs in constraints."
" Stored in ZooKeeper: " + from_zk.constraints +
", local: " + constraints,
ErrorCodes::METADATA_MISMATCH);
throw Exception(
"Existing table metadata in ZooKeeper differs in constraints."
" Stored in ZooKeeper: " + from_zk.constraints +
", local: " + constraints,
ErrorCodes::METADATA_MISMATCH);
}
if (from_zk.index_granularity_bytes_found_in_zk && index_granularity_bytes != from_zk.index_granularity_bytes)
@ -272,9 +255,41 @@ ReplicatedMergeTreeTableMetadata::checkAndFindDiff(const ReplicatedMergeTreeTabl
" Stored in ZooKeeper: " + DB::toString(from_zk.index_granularity_bytes) +
", local: " + DB::toString(index_granularity_bytes),
ErrorCodes::METADATA_MISMATCH);
}
ReplicatedMergeTreeTableMetadata::Diff
ReplicatedMergeTreeTableMetadata::checkAndFindDiff(const ReplicatedMergeTreeTableMetadata & from_zk) const
{
checkImmutableFieldsEquals(from_zk);
Diff diff;
if (sorting_key != from_zk.sorting_key)
{
diff.sorting_key_changed = true;
diff.new_sorting_key = from_zk.sorting_key;
}
if (ttl_table != from_zk.ttl_table)
{
diff.ttl_table_changed = true;
diff.new_ttl_table = from_zk.ttl_table;
}
if (skip_indices != from_zk.skip_indices)
{
diff.skip_indices_changed = true;
diff.new_skip_indices = from_zk.skip_indices;
}
if (constraints != from_zk.constraints)
{
diff.constraints_changed = true;
diff.new_constraints = from_zk.constraints;
}
return diff;
}
}

7
dbms/src/Storages/MergeTree/ReplicatedMergeTreeTableMetadata.h
View File

@ -60,9 +60,14 @@ struct ReplicatedMergeTreeTableMetadata
}
};
Diff checkAndFindDiff(const ReplicatedMergeTreeTableMetadata & from_zk, bool allow_alter) const;
void checkEquals(const ReplicatedMergeTreeTableMetadata & from_zk) const;
Diff checkAndFindDiff(const ReplicatedMergeTreeTableMetadata & from_zk) const;
private:
void checkImmutableFieldsEquals(const ReplicatedMergeTreeTableMetadata & from_zk) const;
bool index_granularity_bytes_found_in_zk = false;
};

71
dbms/src/Storages/StorageReplicatedMergeTree.cpp
View File

@ -286,14 +286,29 @@ StorageReplicatedMergeTree::StorageReplicatedMergeTree(
createTableIfNotExists();
checkTableStructure(false, false);
checkTableStructure(zookeeper_path);
createReplica();
Coordination::Stat metadata_stat;
current_zookeeper->get(zookeeper_path + "/metadata", &metadata_stat);
metadata_version = metadata_stat.version;
}
else
{
checkTableStructure(skip_sanity_checks, true);
checkTableStructure(replica_path);
checkParts(skip_sanity_checks);
if (current_zookeeper->exists(replica_path + "/metadata_version"))
{
metadata_version = parse<int>(current_zookeeper->get(replica_path + "/metadata_version"));
}
else /// This replica was created on old version, so we have to take version of global node
{
Coordination::Stat metadata_stat;
current_zookeeper->get(zookeeper_path + "/metadata", &metadata_stat);
metadata_version = metadata_stat.version;
}
/// Temporary directories contain untinalized results of Merges or Fetches (after forced restart)
/// and don't allow to reinitialize them, so delete each of them immediately
clearOldTemporaryDirectories(0);
@ -301,9 +316,6 @@ StorageReplicatedMergeTree::StorageReplicatedMergeTree(
createNewZooKeeperNodes();
Coordination::Stat metadata_stat;
current_zookeeper->get(zookeeper_path + "/metadata", &metadata_stat);
metadata_version = metadata_stat.version;
other_replicas_fixed_granularity = checkFixedGranualrityInZookeeper();
}
@ -450,10 +462,6 @@ void StorageReplicatedMergeTree::createTableIfNotExists()
ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/replicas", "",
zkutil::CreateMode::Persistent));
////std::cerr << "Creating alters node\n";
ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/alters", "",
zkutil::CreateMode::Persistent));
Coordination::Responses responses;
auto code = zookeeper->tryMulti(ops, responses);
if (code && code != Coordination::ZNODEEXISTS)
@ -464,45 +472,23 @@ void StorageReplicatedMergeTree::createTableIfNotExists()
/** Verify that list of columns and table storage_settings_ptr match those specified in ZK (/ metadata).
* If not, throw an exception.
*/
void StorageReplicatedMergeTree::checkTableStructure(bool skip_sanity_checks, bool allow_alter)
void StorageReplicatedMergeTree::checkTableStructure(const String & zookeeper_prefix)
{
auto zookeeper = getZooKeeper();
ReplicatedMergeTreeTableMetadata old_metadata(*this);
Coordination::Stat metadata_stat;
String metadata_str = zookeeper->get(zookeeper_path + "/metadata", &metadata_stat);
String metadata_str = zookeeper->get(zookeeper_prefix + "/metadata", &metadata_stat);
auto metadata_from_zk = ReplicatedMergeTreeTableMetadata::parse(metadata_str);
auto metadata_diff = old_metadata.checkAndFindDiff(metadata_from_zk, allow_alter);
old_metadata.checkEquals(metadata_from_zk);
Coordination::Stat columns_stat;
auto columns_from_zk = ColumnsDescription::parse(zookeeper->get(zookeeper_path + "/columns", &columns_stat));
/// TODO(alesap) remove this trash
const ColumnsDescription & old_columns = getColumns();
if (columns_from_zk != old_columns || !metadata_diff.empty())
{
if (allow_alter &&
(skip_sanity_checks ||
old_columns.getOrdinary().sizeOfDifference(columns_from_zk.getOrdinary()) +
old_columns.getMaterialized().sizeOfDifference(columns_from_zk.getMaterialized()) <= 2))
{
LOG_WARNING(log, "Table structure in ZooKeeper is a little different from local table structure. Assuming ALTER.");
/// We delay setting table structure till startup() because otherwise new table metadata file can
/// be overwritten in DatabaseOrdinary::createTable.
set_table_structure_at_startup = [columns_from_zk, metadata_diff, this]()
{
/// Without any locks, because table has not been created yet.
setTableStructure(std::move(columns_from_zk), metadata_diff);
};
}
else
{
throw Exception("Table structure in ZooKeeper is too different from local table structure",
ErrorCodes::INCOMPATIBLE_COLUMNS);
}
}
if (columns_from_zk != old_columns)
throw Exception("Table columns structure in ZooKeeper is different from local table structure", ErrorCodes::INCOMPATIBLE_COLUMNS);
}
@ -2934,9 +2920,6 @@ void StorageReplicatedMergeTree::startup()
if (is_readonly)
return;
if (set_table_structure_at_startup)
set_table_structure_at_startup();
queue.initialize(
zookeeper_path, replica_path,
getStorageID().getFullTableName() + " (ReplicatedMergeTreeQueue)",
@ -3223,7 +3206,7 @@ bool StorageReplicatedMergeTree::executeMetadataAlter(const StorageReplicatedMer
auto columns_from_entry = ColumnsDescription::parse(entry.columns_str);
auto metadata_from_entry = ReplicatedMergeTreeTableMetadata::parse(entry.metadata_str);
auto metadata_diff = ReplicatedMergeTreeTableMetadata(*this).checkAndFindDiff(metadata_from_entry, /* allow_alter = */ true);
auto metadata_diff = ReplicatedMergeTreeTableMetadata(*this).checkAndFindDiff(metadata_from_entry);
MergeTreeData::DataParts parts;
@ -3232,6 +3215,7 @@ bool StorageReplicatedMergeTree::executeMetadataAlter(const StorageReplicatedMer
requests.emplace_back(zkutil::makeSetRequest(replica_path + "/columns", entry.columns_str, -1));
requests.emplace_back(zkutil::makeSetRequest(replica_path + "/metadata", entry.metadata_str, -1));
zookeeper->multi(requests);
{
@ -3245,6 +3229,9 @@ bool StorageReplicatedMergeTree::executeMetadataAlter(const StorageReplicatedMer
LOG_INFO(log, "Applied changes to the metadata of the table. Current metadata version: " << metadata_version);
}
/// This transaction may not happen, but it's ok, because on the next retry we will eventually create this node
zookeeper->createOrUpdate(replica_path + "/metadata_version", std::to_string(metadata_version), zkutil::CreateMode::Persistent);
recalculateColumnSizes();
return true;
@ -3287,7 +3274,7 @@ void StorageReplicatedMergeTree::alter(
/// Clear nodes from previous iteration
alter_entry.emplace();
mutation_znode.emplace();
mutation_znode.reset();
/// We can safely read structure, because we guarded with alter_intention_lock
if (is_readonly)
@ -3432,7 +3419,7 @@ void StorageReplicatedMergeTree::alter(
if (!unwaited.empty())
throw Exception("Some replicas doesn't finish metadata alter", ErrorCodes::UNFINISHED);
if (mutation_znode.has_value() && !mutation_znode->empty())
if (mutation_znode)
{
LOG_DEBUG(log, "Metadata changes applied. Will wait for data changes.");
waitMutation(*mutation_znode, query_context.getSettingsRef().replication_alter_partitions_sync);

9
dbms/src/Storages/StorageReplicatedMergeTree.h
View File

@ -221,9 +221,6 @@ private:
*/
zkutil::EphemeralNodeHolderPtr replica_is_active_node;
/// Used to delay setting table structure till startup() in case of an offline ALTER.
std::function<void()> set_table_structure_at_startup;
/** Is this replica "leading". The leader replica selects the parts to merge.
*/
std::atomic<bool> is_leader {false};
@ -308,11 +305,7 @@ private:
*/
void createNewZooKeeperNodes();
/** Verify that the list of columns and table settings match those specified in ZK (/metadata).
* If not, throw an exception.
* Must be called before startup().
*/
void checkTableStructure(bool skip_sanity_checks, bool allow_alter);
void checkTableStructure(const String & zookeeper_prefix);
/// A part of ALTER: apply metadata changes only (data parts are altered separately).
/// Must be called under IStorage::lockStructureForAlter() lock.