tikv: add doc for new config max-apply-unpersisted-log-limit (pingc…

…ap#17857)

dynamic-config.md

@@ -135,6 +135,7 @@ The following TiKV configuration items can be modified dynamically:
 | `raftstore.raft-max-size-per-msg` | The soft limit on the size of a single message packet that is allowed to be generated |
 | `raftstore.raft-entry-max-size` | The hard limit on the maximum size of a single Raft log |
 | `raftstore.raft-entry-cache-life-time` | The maximum remaining time allowed for the log cache in memory |
+| `raftstore.max-apply-unpersisted-log-limit` | The maximum number of committed but not persisted Raft logs that can be applied |
 | `raftstore.split-region-check-tick-interval` | The time interval at which to check whether the Region split is needed |
 | `raftstore.region-split-check-diff` | The maximum value by which the Region data is allowed to exceed before Region split |
 | `raftstore.region-compact-check-interval` | The time interval at which to check whether it is necessary to manually trigger RocksDB compaction |

tikv-configuration-file.md

@@ -738,6 +738,16 @@ Configuration items related to Raftstore.
 + Default value: `"30s"`
 + Minimum value: `0`
 
+### `max-apply-unpersisted-log-limit` <span class="version-mark">New in v8.1.0</span>
+
++ The maximum number of committed but not persisted Raft logs that can be applied.
+
+    + Setting this configuration item to a value greater than `0` enables the TiKV node to apply committed but not persisted Raft logs in advance, effectively reducing long-tail latency caused by IO jitter on that node. However, it might also increase the memory usage of TiKV and the disk space occupied by Raft logs.
+    + Setting this configuration item to `0` disables this feature, meaning that TiKV must wait until Raft logs are both committed and persisted before applying them. This behavior is consistent with the behavior before v8.2.0.
+
++ Default value: `1024`
++ Minimum value: `0`
+
 ### `hibernate-regions`
 
 + Enables or disables Hibernate Region. When this option is enabled, a Region idle for a long time is automatically set as hibernated. This reduces the extra overhead caused by heartbeat messages between the Raft leader and the followers for idle Regions. You can use `peer-stale-state-check-interval` to modify the heartbeat interval between the leader and the followers of hibernated Regions.