mirror of
https://github.com/ClickHouse/ClickHouse.git
synced 2024-12-01 03:52:15 +00:00
Merge pull request #25663 from azat/dist-startup
Improve startup time of Distributed engine.
This commit is contained in:
commit
1b644b9a31
@ -752,7 +752,7 @@ void DistributedBlockOutputStream::writeToShard(const Block & block, const std::
|
|||||||
auto sleep_ms = context->getSettingsRef().distributed_directory_monitor_sleep_time_ms;
|
auto sleep_ms = context->getSettingsRef().distributed_directory_monitor_sleep_time_ms;
|
||||||
for (const auto & dir_name : dir_names)
|
for (const auto & dir_name : dir_names)
|
||||||
{
|
{
|
||||||
auto & directory_monitor = storage.requireDirectoryMonitor(disk, dir_name);
|
auto & directory_monitor = storage.requireDirectoryMonitor(disk, dir_name, /* startup= */ false);
|
||||||
directory_monitor.addAndSchedule(file_size, sleep_ms.totalMilliseconds());
|
directory_monitor.addAndSchedule(file_size, sleep_ms.totalMilliseconds());
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -800,12 +800,33 @@ void StorageDistributed::startup()
|
|||||||
if (!storage_policy)
|
if (!storage_policy)
|
||||||
return;
|
return;
|
||||||
|
|
||||||
for (const DiskPtr & disk : data_volume->getDisks())
|
const auto & disks = data_volume->getDisks();
|
||||||
createDirectoryMonitors(disk);
|
|
||||||
|
|
||||||
for (const String & path : getDataPaths())
|
/// Make initialization for large number of disks parallel.
|
||||||
|
ThreadPool pool(disks.size());
|
||||||
|
|
||||||
|
for (const DiskPtr & disk : disks)
|
||||||
|
{
|
||||||
|
pool.scheduleOrThrowOnError([&]()
|
||||||
|
{
|
||||||
|
createDirectoryMonitors(disk);
|
||||||
|
});
|
||||||
|
}
|
||||||
|
pool.wait();
|
||||||
|
|
||||||
|
const auto & paths = getDataPaths();
|
||||||
|
std::vector<UInt64> last_increment(paths.size());
|
||||||
|
for (size_t i = 0; i < paths.size(); ++i)
|
||||||
|
{
|
||||||
|
pool.scheduleOrThrowOnError([&, i]()
|
||||||
|
{
|
||||||
|
last_increment[i] = getMaximumFileNumber(paths[i]);
|
||||||
|
});
|
||||||
|
}
|
||||||
|
pool.wait();
|
||||||
|
|
||||||
|
for (const auto inc : last_increment)
|
||||||
{
|
{
|
||||||
UInt64 inc = getMaximumFileNumber(path);
|
|
||||||
if (inc > file_names_increment.value)
|
if (inc > file_names_increment.value)
|
||||||
file_names_increment.value.store(inc);
|
file_names_increment.value.store(inc);
|
||||||
}
|
}
|
||||||
@ -907,30 +928,50 @@ void StorageDistributed::createDirectoryMonitors(const DiskPtr & disk)
|
|||||||
}
|
}
|
||||||
else
|
else
|
||||||
{
|
{
|
||||||
requireDirectoryMonitor(disk, dir_path.filename().string());
|
requireDirectoryMonitor(disk, dir_path.filename().string(), /* startup= */ true);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
StorageDistributedDirectoryMonitor& StorageDistributed::requireDirectoryMonitor(const DiskPtr & disk, const std::string & name)
|
StorageDistributedDirectoryMonitor& StorageDistributed::requireDirectoryMonitor(const DiskPtr & disk, const std::string & name, bool startup)
|
||||||
{
|
{
|
||||||
const std::string & disk_path = disk->getPath();
|
const std::string & disk_path = disk->getPath();
|
||||||
const std::string key(disk_path + name);
|
const std::string key(disk_path + name);
|
||||||
|
|
||||||
|
auto create_node_data = [&]()
|
||||||
|
{
|
||||||
|
ClusterNodeData data;
|
||||||
|
data.connection_pool = StorageDistributedDirectoryMonitor::createPool(name, *this);
|
||||||
|
data.directory_monitor = std::make_unique<StorageDistributedDirectoryMonitor>(
|
||||||
|
*this, disk, relative_data_path + name,
|
||||||
|
data.connection_pool,
|
||||||
|
monitors_blocker,
|
||||||
|
getContext()->getDistributedSchedulePool());
|
||||||
|
return data;
|
||||||
|
};
|
||||||
|
|
||||||
|
/// In case of startup the lock can be acquired later.
|
||||||
|
if (startup)
|
||||||
|
{
|
||||||
|
auto tmp_node_data = create_node_data();
|
||||||
|
std::lock_guard lock(cluster_nodes_mutex);
|
||||||
|
auto & node_data = cluster_nodes_data[key];
|
||||||
|
assert(!node_data.directory_monitor);
|
||||||
|
node_data = std::move(tmp_node_data);
|
||||||
|
return *node_data.directory_monitor;
|
||||||
|
}
|
||||||
|
else
|
||||||
|
{
|
||||||
std::lock_guard lock(cluster_nodes_mutex);
|
std::lock_guard lock(cluster_nodes_mutex);
|
||||||
auto & node_data = cluster_nodes_data[key];
|
auto & node_data = cluster_nodes_data[key];
|
||||||
if (!node_data.directory_monitor)
|
if (!node_data.directory_monitor)
|
||||||
{
|
{
|
||||||
node_data.connection_pool = StorageDistributedDirectoryMonitor::createPool(name, *this);
|
node_data = create_node_data();
|
||||||
node_data.directory_monitor = std::make_unique<StorageDistributedDirectoryMonitor>(
|
|
||||||
*this, disk, relative_data_path + name,
|
|
||||||
node_data.connection_pool,
|
|
||||||
monitors_blocker,
|
|
||||||
getContext()->getDistributedSchedulePool());
|
|
||||||
}
|
}
|
||||||
return *node_data.directory_monitor;
|
return *node_data.directory_monitor;
|
||||||
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
std::vector<StorageDistributedDirectoryMonitor::Status> StorageDistributed::getDirectoryMonitorsStatuses() const
|
std::vector<StorageDistributedDirectoryMonitor::Status> StorageDistributed::getDirectoryMonitorsStatuses() const
|
||||||
|
@ -160,7 +160,7 @@ private:
|
|||||||
/// create directory monitors for each existing subdirectory
|
/// create directory monitors for each existing subdirectory
|
||||||
void createDirectoryMonitors(const DiskPtr & disk);
|
void createDirectoryMonitors(const DiskPtr & disk);
|
||||||
/// ensure directory monitor thread and connectoin pool creation by disk and subdirectory name
|
/// ensure directory monitor thread and connectoin pool creation by disk and subdirectory name
|
||||||
StorageDistributedDirectoryMonitor & requireDirectoryMonitor(const DiskPtr & disk, const std::string & name);
|
StorageDistributedDirectoryMonitor & requireDirectoryMonitor(const DiskPtr & disk, const std::string & name, bool startup);
|
||||||
|
|
||||||
/// Return list of metrics for all created monitors
|
/// Return list of metrics for all created monitors
|
||||||
/// (note that monitors are created lazily, i.e. until at least one INSERT executed)
|
/// (note that monitors are created lazily, i.e. until at least one INSERT executed)
|
||||||
|
Loading…
Reference in New Issue
Block a user