Description PD max- �→ pending �→ -peer- �→ count Modified Controls the maximum number of pending peers in a single store. PD schedulers depend on this configu- ration to prevent too many Regions with outdated v2021.10.08.1 #4070 – Support that the evict leader scheduler can schedule Regions with unhealthy peers #4093 – Speed up the exit process of schedulers #4146 • TiFlash – Improve the execution efficiency incorrectly delete the peers with data and in pending status because the number of peers exceeds the number of configured peers #4045 – Fix the issue that PD does not fix down peers in time #4077 – Fix the

from a non-existent store #3660 – Fix the issue that the statistics are not updated after offline peers are merged #3611 • TiFlash – Fix the issue of incorrect results when casting the time type to the the complete archive because: • Data are replicated into multiple peers. When ingesting SSTs, these files have to be present on all peers. This is unlike back up where reading from a single node is enough pending-peer-region-count: The number of Regions with outdated Raft logs. It is normal that a few pending peers are generated in the scheduling process. However, it is not normal if this value is large for a period

the complete archive because: • Data are replicated into multiple peers. When ingesting SSTs, these files have to be present on all peers. This is unlike back up where reading from a single node is enough pending-peer-region-count: The number of Regions with outdated Raft logs. It is normal that a few pending peers are generated in the scheduling process. However, it is not normal if this value is large for a period max �→ - �→ pending �→ - �→ peer �→ - �→ count �→ Determines the maxi- mum num- ber of pend- ing peers in a sin- gle store schedule �→ . �→ max �→ - �→ store �→ - �→ down �→ - �→ time �→ The down- time

the complete archive because: • Data are replicated into multiple peers. When ingesting SSTs, these files have to be present on all peers. This is unlike back up where reading from a single node is enough pending-peer-region-count: The number of Regions with outdated Raft logs. It is normal that a few pending peers are generated in the scheduling process. However, it is not normal if this value is large for a period max �→ - �→ pending �→ - �→ peer �→ - �→ count �→ Determines the maxi- mum num- ber of pend- ing peers in a sin- gle store schedule �→ . �→ max �→ - �→ store �→ - �→ down �→ - �→ time �→ The down- time

increasing when a TiKV instance is taken offline #12161 • Fix the issue that TiKV panics and destroys peers unexpectedly because the target Region to be merged is invalid #12232 • Fix the issue that TiKV reports resource is insufficient #11379 • Fix the issue of TiKV panic caused by the race between destroying peers and batch splitting Regions #12368 • Fix the issue of TiKV memory usage spike in a short time after of Regions with the learner peer. The sources of learner peers can be various, for example, the learner peers in TiFlash, and the learner peers included in the configured Placement Rules. • miss-peer-region-count:

Auto Synchronous (DR Auto-Sync) mode #7221 @disksing • Fix the issue that PD might delete normal Peers when TiKV nodes are unavailable #7249 @lhy1024 • Fix the issue that adding multiple TiKV nodes to which can avoid triggering this problem. • Issue-43079: TiDB Lightning fails to refresh the Region Peers information during retry for NotLeader errors. • Issue-43291: TiDB Lightning does not retry in cases of Regions with the learner peer. The sources of learner peers can be various, for example, the learner peers in TiFlash, and the learner peers included in the configured Placement Rules. • miss-peer-region-count:

of Regions with the learner peer. The sources of learner peers can be various, for example, the learner peers in TiFlash, and the learner peers included in the configured Placement Rules. • miss-peer-region-count: pending-peer-region-count: The number of Regions with outdated Raft logs. It is normal that a few pending peers are generated in the scheduling process. However, it is not normal if this value is large for a period max �→ - �→ pending �→ - �→ peer �→ - �→ count �→ Determines the maxi- mum num- ber of pend- ing peers in a sin- gle store schedule �→ . �→ max �→ - �→ store �→ - �→ down �→ - �→ time �→ The down- time

which can avoid triggering this problem. • Issue-43079: TiDB Lightning fails to refresh the Region Peers information during retry for NotLeader errors. • Issue-43291: TiDB Lightning does not retry in cases pending-peer-region-count: The number of Regions with outdated Raft logs. It is normal that a few pending peers are generated in the scheduling process. However, it is not normal if this value is large for a period max �→ - �→ pending �→ - �→ peer �→ - �→ count �→ Determines the maxi- mum num- ber of pend- ing peers in a sin- gle store schedule �→ . �→ max �→ - �→ store �→ - �→ down �→ - �→ time �→ The down- time

which can avoid triggering this problem. • Issue-43079: TiDB Lightning fails to refresh the Region Peers information during retry for NotLeader errors. • Issue-43291: TiDB Lightning does not retry in cases of Regions with the learner peer. The sources of learner peers can be various, for example, the learner peers in TiFlash, and the learner peers included in the configured Placement Rules. • miss-peer-region-count: pending-peer-region-count: The number of Regions with outdated Raft logs. It is normal that a few pending peers are generated in the scheduling process. However, it is not normal if this value is large for a period

apply feature by default. With this feature enabled, the Raft leader can apply logs after quorum peers have persisted the logs, without waiting for the leader itself to persist the log, reducing the impact write hotspots might break placement policy constraints #7848 @lhy1024 • Fix the issue that down peers might not recover when using Placement Rules #7808 @rleungx • Fix the issue that a large number which can avoid triggering this problem. • Issue-43079: TiDB Lightning fails to refresh the Region Peers information during retry for NotLeader errors. • Issue-43291: TiDB Lightning does not retry in cases