| title | summary | category |
|---|---|---|
Raftstore Configuration |
Learn about Raftstore configuration in TiKV. |
reference |
Raftstore is TiKV's implementation of Multi-raft to manage multiple Raft peers on one node. Raftstore is comprised of two major components:
- Raftstore component writes Raft logs into RaftDB.
- Apply component resolves Raft logs and flush the data in the log into the underlying storage engine.
This document introduces the following features of Raftstore and their configurations:
Multi-thread support for the Raftstore and the Apply components means higher throughput and lower latency per each single node. In the multi-thread mode, each thread obtains peers from the queue in batch, so that small writes of multiple peers can be consolidated into a big write for better throughput.
Note:
In the multi-thread mode, peers are obtained in batch, so pressure from hot write Regions cannot be scattered evenly to each CPU. For better load balancing, it is recommended you use smaller batch sizes.
You can specify the following items in the TiKV configuration file to configure multi-thread Raftstore:
raftstore.store-max-batch-size
Determines the maximum number of peers that a single thread can obtain in a batch. The value must be a positive integer. A smaller value provides better load balancing for CPU, but may cause more frequent writes.
raftstore.store-pool-size
Determines the number of threads to process peers in batch. The value must be a positive integer. For better performance, it is recommended that you set a value less than or equal to the number of CPU cores on your machine.
raftstore.apply-max-batch-size
Determines the maximum number of ApplyDelegates requests that a single thread can resolve in a batch. The value must be a positive integer. A smaller value provides better load balancing for CPU, but may cause more frequent writes.
raftstore.apply-pool-size
Determines the number of threads. The value must be a positive integer. For better performance, it is recommended that you set a value less than or equal to the number of CPU cores on your machine.
Hibernate Region is a Raftstore feature to reduce the extra overhead caused by heartbeat messages between the Raft leader and the followers for idle Regions. With this feature enabled, a Region idle for a long time is automatically set as hibernated. The heartbeat interval for the leader to maintain its lease becomes much longer, and the followers do not initiate elections simply because they cannot receive heartbeats from the leader.
Note:
- Hibernate Region is still an Experimental feature and is disabled by default.
- Any requests from the client or disconnections will activate the Region from the hibernated state.
You can specify the following items in the TiKV configuration file to configure Hibernate Region:
raftstore.hibernate-regions
Enables or disables Hibernate Region. Possible values are true and false. The default value is false.
raftstore.peer-stale-state-check-interval
Modifies the state check interval for hibernated Regions. The default value is 5 minutes. This value also determines the heartbeat interval between the leader and followers of the hibernated Regions.