ClickHouse/programs/copier/ClusterCopierApp.h

94 lines
4.1 KiB
C++
Raw Normal View History

2020-02-11 18:34:48 +00:00
#pragma once
2020-02-20 10:01:02 +00:00
2020-02-11 18:34:48 +00:00
#include <Poco/Util/ServerApplication.h>
#include <daemon/BaseDaemon.h>
2020-02-20 10:01:02 +00:00
#include "ClusterCopier.h"
2020-02-11 18:34:48 +00:00
/* clickhouse cluster copier util
* Copies tables data from one cluster to new tables of other (possibly the same) cluster in distributed fault-tolerant manner.
*
* See overview in the docs: docs/en/utils/clickhouse-copier.md
*
* Implementation details:
*
* cluster-copier workers pull each partition of each shard of the source cluster and push it to the destination cluster through
2020-08-08 01:21:04 +00:00
* Distributed table (to perform data resharding). So, worker job is a partition of a source shard.
2020-02-11 18:34:48 +00:00
* A job has three states: Active, Finished and Abandoned. Abandoned means that worker died and did not finish the job.
*
* If an error occurred during the copying (a worker failed or a worker did not finish the INSERT), then the whole partition (on
* all destination servers) should be dropped and refilled. So, copying entity is a partition of all destination shards.
* If a failure is detected a special /is_dirty node is created in ZooKeeper signalling that other workers copying the same partition
* should stop, after a refilling procedure should start.
*
* ZooKeeper task node has the following structure:
* /task/path_root - path passed in --task-path parameter
* /description - contains user-defined XML config of the task
* /task_active_workers - contains ephemeral nodes of all currently active workers, used to implement max_workers limitation
* /server_fqdn#PID_timestamp - cluster-copier worker ID
* ...
* /tables - directory with table tasks
* /cluster.db.table1 - directory of table_hits task
* /partition1 - directory for partition1
* /shards - directory for source cluster shards
* /1 - worker job for the first shard of partition1 of table test.hits
* Contains info about current status (Active or Finished) and worker ID.
* /2
* ...
* /partition_active_workers
* /1 - for each job in /shards a corresponding ephemeral node created in /partition_active_workers
* It is used to detect Abandoned jobs (if there is Active node in /shards and there is no node in
* /partition_active_workers).
* Also, it is used to track active workers in the partition (when we need to refill the partition we do
* not DROP PARTITION while there are active workers)
* /2
* ...
* /is_dirty - the node is set if some worker detected that an error occurred (the INSERT is failed or an Abandoned node is
* detected). If the node appeared workers in this partition should stop and start cleaning and refilling
* partition procedure.
* During this procedure a single 'cleaner' worker is selected. The worker waits for stopping all partition
* workers, removes /shards node, executes DROP PARTITION on each destination node and removes /is_dirty node.
* /cleaner- An ephemeral node used to select 'cleaner' worker. Contains ID of the worker.
* /cluster.db.table2
* ...
*/
namespace DB
{
class ClusterCopierApp : public BaseDaemon
{
public:
void initialize(Poco::Util::Application & self) override;
void handleHelp(const std::string &, const std::string &);
void defineOptions(Poco::Util::OptionSet & options) override;
int main(const std::vector<std::string> &) override;
private:
using Base = BaseDaemon;
void mainImpl();
std::string config_xml_path;
std::string task_path;
std::string log_level = "trace";
bool is_safe_mode = false;
2020-03-16 21:05:38 +00:00
double copy_fault_probability = 0.0;
double move_fault_probability = 0.0;
2020-02-11 18:34:48 +00:00
bool is_help = false;
2020-04-21 17:37:40 +00:00
bool experimental_use_sample_offset{false};
2020-02-11 18:34:48 +00:00
std::string base_dir;
std::string process_path;
std::string process_id;
std::string host_id;
};
}