Improvements
- Make the prewrite requests as idempotent as possible to reduce the chance of undetermined errors #10586
- Prevent the risk of stack overflow when handling many expired commands #10502
- Avoid excessive commit request retrying by not using the Stale Read request's
start_ts
to updatemax_ts
#10451 - Handle read ready and write ready separately to reduce read latency #10592
- Reduce the impact on data import speed when the I/O rate limiting is enabled #10390
- Improve the load balance between Raft gRPC connections #10495
Bug fixes
- Fix the issue that the duration calculation might panic on certain platforms #10569
- Fix the issue that Load Base Split mistakenly uses the unencoded keys of
batch_get_command
#10542 - Fix the issue that changing the
resolved-ts.advance-ts-interval
configuration online cannot take effect immediately #10426 - Fix the issue of follower metadata corruption in rare cases with more than 4 replicas #10225
- Fix the panic issue that occurs when building a snapshot twice if encryption is enabled #9786 #10407
- Fix the wrong
tikv_raftstore_hibernated_peer_state
metric #10330 - Fix the wrong arguments type of the
json_unquote()
function in the coprocessor #10176 - Fix a bug that the index keys in a pessimistic transaction might be repeatedly committed #10468
- Fix the issue that the
ReadIndex
request returns stale result right after the leader is transferred #9351