[pre-production] clickhouse-keeper 

ClickHouse server use ZooKeeper coordination system for data replication and distributed DDL queries execution. ClickHouse Keeper is an alternative coordination system compatible with ZooKeeper.

Implementation details 

ZooKeeper is one of the first well-known open-source coordination systems. It's implemented in Java, has quite a simple and powerful data model. ZooKeeper's coordination algorithm called ZAB (ZooKeeper Atomic Broadcast) doesn't provide linearizability guarantees for reads, because each ZooKeeper node serves reads locally. Unlike ZooKeeper clickhouse-keeper written in C++ and use RAFT algorithm implementation. This algorithm allows to have linearizability for reads and writes, has several open-source implementations in different languages.

By default, clickhouse-keeper provides the same guarantees as ZooKeeper (linearizable writes, non-linearizable reads). It has a compatible client-server protocol, so any standard ZooKeeper client can be used to interact with clickhouse-keeper. Snapshots and logs have an incompatible format with ZooKeeper, but clickhouse-keeper-converter tool allows to convert ZooKeeper data to clickhouse-keeper snapshot. Interserver protocol in clickhouse-keeper also incompatible with ZooKeeper so mixed ZooKeeper/clickhouse-keeper cluster is impossible.


clickhouse-keeper can be used as a standalone replacement for ZooKeeper or as an internal part of the clickhouse-server, but in both cases configuration is almost the same .xml file. The main clickhouse-keeper configuration tag is <keeper_server>. Keeper configuration has the following parameters:

  • tcp_port — the port for a client to connect (default for ZooKeeper is 2181)
  • tcp_port_secure — the secure port for a client to connect
  • server_id — unique server id, each participant of the clickhouse-keeper cluster must have a unique number (1, 2, 3, and so on)
  • log_storage_path — path to coordination logs, better to store logs on the non-busy device (same for ZooKeeper)
  • snapshot_storage_path — path to coordination snapshots

Other common parameters are inherited from clickhouse-server config (listen_host, logger and so on).

Internal coordination settings are located in <keeper_server>.<coordination_settings> section:

  • operation_timeout_ms — timeout for a single client operation (default: 10000)
  • session_timeout_ms — timeout for client session (default: 30000)
  • dead_session_check_period_ms — how often clickhouse-keeper check dead sessions and remove them (default: 500)
  • heart_beat_interval_ms — how often a clickhouse-keeper leader will send heartbeats to followers (default: 500)
  • election_timeout_lower_bound_ms — if follower didn't receive heartbeats from the leader in this interval, then it can initiate leader election (default: 1000)
  • election_timeout_upper_bound_ms — if follower didn't receive heartbeats from the leader in this interval, then it must initiate leader election (default: 2000)
  • rotate_log_storage_interval — how many log records to store in a single file (default: 100000)
  • reserved_log_items — how many coordination log records to store before compaction (default: 100000)
  • snapshot_distance — how often clickhouse-keeper will create new snapshots (in the number of records in logs) (default: 100000)
  • snapshots_to_keep — how many snapshots to keep (default: 3)
  • stale_log_gap — the threshold when leader consider follower as stale and send snapshot to it instead of logs (default: 10000)
  • fresh_log_gap - when node became fresh (default: 200)
  • max_requests_batch_size - max size of batch in requests count before it will be sent to RAFT (default: 100)
  • force_sync — call fsync on each write to coordination log (default: true)
  • quorum_reads - execute read requests as writes through whole RAFT consesus with similar speed (default: false)
  • raft_logs_level — text logging level about coordination (trace, debug, and so on) (default: system default)
  • auto_forwarding - allow to forward write requests from followers to leader (default: true)
  • shutdown_timeout — wait to finish internal connections and shutdown (ms) (default: 5000)
  • startup_timeout — if the server doesn't connect to other quorum participants in the specified timeout it will terminate (ms) (default: 30000)

Quorum configuration is located in <keeper_server>.<raft_configuration> section and contain servers description. The only parameter for the whole quorum is secure, which enables encrypted connection for communication between quorum participants. The main parameters for each <server> are:

  • id — server_id in quorum
  • hostname — hostname where this server placed
  • port — port where this server listen for connections

Examples of configuration for quorum with three nodes can be found in integration tests with test_keeper_ prefix. Example configuration for server #1:




How to run 

clickhouse-keeper is bundled into clickhouse-server package, just add configuration of <keeper_server> and start clickhouse-server as always. If you want to run standalone clickhouse-keeper you can start it in a similar way with:

clickhouse-keeper --config /etc/your_path_to_config/config.xml --daemon

[experimental] Migration from ZooKeeper 

Seamlessly migration from ZooKeeper to clickhouse-keeper is impossible you have to stop your ZooKeeper cluster, convert data and start clickhouse-keeper. clickhouse-keeper-converter tool allows to convert ZooKeeper logs and snapshots to clickhouse-keeper snapshot. It works only with ZooKeeper > 3.4. Steps for migration:

  1. Stop all ZooKeeper nodes.

  2. [optional, but recommended] Found ZooKeeper leader node, start and stop it again. It will force ZooKeeper to create consistent snapshot.

  3. Run clickhouse-keeper-converter on leader, example

clickhouse-keeper-converter --zookeeper-logs-dir /var/lib/zookeeper/version-2 --zookeeper-snapshots-dir /var/lib/zookeeper/version-2 --output-dir /path/to/clickhouse/keeper/snapshots
  1. Copy snapshot to clickhouse-server nodes with configured keeper or start clickhouse-keeper instead of ZooKeeper. Snapshot must persist only on leader node, leader will sync it automatically to other nodes.

Rating: 5 - 1 votes

Was this content helpful?