TiDB 5.0.2 Release Notes
Release date: June 10, 2021
TiDB version: 5.0.2
Compatibility Changes
Tools
TiCDC
- Deprecate
--sort-dirin thecdc cli changefeedcommand. Instead, users can set--sort-dirin thecdc servercommand. #1795
- Deprecate
New Features
TiKV
- Enable the Hibernate Region feature by default #10266
Improvements
TiDB
- Avoid frequently reading the
mysql.stats_histogramstable if the cached statistics is up-to-date to avoid high CPU usage #24317
- Avoid frequently reading the
TiKV
- BR now supports the S3-compatible storage using the virtual-host addressing mode #10243
- Support the back pressure for TiCDC's scan speed #10151
- Reduce the memory usage of TiCDC's initial scan #10133
- Improve the cache hit ratio of the TiCDC's Old Value feature in the pessimistic transaction #10089
- Split Regions more evenly to mitigate the issue that the growth of Region size exceeds the splitting speed when there are hotspot writes #9785
TiFlash
- Optimize the table lock to prevent DDL jobs and data reads from blocking each other
- Support casting the
INTEGERorREALtype toREALtype
Tools
TiCDC
Backup & Restore (BR)
Dumpling
- Fix the issue that no error is output when a backup operation fails #280
Bug Fixes
TiDB
- Fix the panic issue caused by using the prefix index and index join in some cases #24547 #24716 #24717
- Fix the issue that the prepared plan cache of
point getis incorrectly used by thepoint getstatement in the transaction #24741 - Fix the issue of writing the wrong prefix index value when the collation is
ascii_binorlatin1_bin#24569 - Fix the issue that the ongoing transaction might be interrupted by the GC worker #24591
- Fix a bug that the point query might get wrong on the clustered index when
new-collationis enabled butnew-row-formatis disabled #24541 - Refactor the conversion of partition keys for shuffle hash join #24490
- Fix the panic issue that occurs when building the plan for queries that contain the
HAVINGclause #24045 - Fix the issue that the column pruning improvement causes the
ApplyandJoinoperators' results to go wrong #23887 - Fix a bug that the primary lock fallen back from async commit cannot be resolved #24384
- Fix a GC issue of statistics that might cause duplicated fm-sketch records #24357
- Avoid unnecessary pessimistic rollback when the pessimistic locking receives the
ErrKeyExistserror #23799 - Fix the issue that numeric literals cannot be recognized when the sql_mode contains
ANSI_QUOTES#24429 - Forbid statements such as
INSERT INTO table PARTITION (<partitions>) ... ON DUPLICATE KEY UPDATEto read data from non-listed partitions #24746 - Fix the potential
index out of rangeerror when a SQL statement contains bothGROUP BYandUNION#24281 - Fix the issue that the
CONCATfunction incorrectly handles the collation #24296 - Fix the issue that the
collation_serverglobal variable does not take effect in new sessions #24156
TiKV
- Fix a TiCDC OOM issue caused by reading old values #9996 #9981
- Fix the issue of empty value in the secondary index for the clustered primary key column when collation is
latin1_bin#24548 - Add the
abort-on-panicconfiguration, which allows TiKV to generate the core dump file when panic occurs. Users still need to correctly configure the environment to enable core dump #10216 - Fix the performance regression issue of
point getqueries that occurs when TiKV is not busy #10046
PD
TiFlash
- Fix the issue of incorrect results when cloning shared delta index concurrently
- Fix the potential issue that TiFlash fails to restart with incomplete data
- Fix the issue that old dm files are not removed automatically
- Fix the potential panic that occurs when the Compaction Filter feature is enabled
- Fix the potential issue that
ExchangeSendersends duplicated data - Fix the issue that TiFlash cannot resolve the lock fallen back from async commit
- Fix the issue of incorrect results returned when the casted result of the
TIMEZONEtype contains theTIMESTAMPtype - Fix the TiFlash panic issue that occurs during Segment Split
- Fix the issue that the execution information about the non-root MPP task is not accurate
Tools
TiCDC
- Fix the issue that the time zone information is lost in the Avro output #1712
- Support cleaning up stale temporary files in Unified Sorter and forbid sharing the
sort-dirdirectory #1742 - Fix a deadlock bug in the KV client that occurs when many stale Regions exist #1599
- Fix the wrong help information in the
--cert-allowed-cnflag #1697 - Revert the update for
explicit_defaults_for_timestampwhich requires theSUPERprivilege when replicating data to MySQL #1750 - Support the sink flow control to reduce the risk of memory overflow #1840
- Fix a bug that the replication task might stop when moving a table #1828
- Fix the issue that the TiKV GC safe point is blocked due to the stagnation of TiCDC changefeed checkpoint #1759
Backup & Restore (BR)
TiDB Lightning
- Fix the issue of TiDB Lightning panic that occurs when generating KV data #1127
- Fix the issue that TiDB Lightning in the TiDB-backend mode cannot load any data when the autocommit is disabled #1104
- Fix a bug that the batch split Region fails due to the total key size exceeding the raft entry limit during the data import #969