TiDB 7.5.8 Release Notes

Release date: September 17, 2026

TiDB version: 7.5.8

Quick access: Quick start | Production deployment

Improvements

• PD

  • Add the store label to the pd_cluster_status metric so that PD reports per-store status metrics, making it easier to identify affected TiKV stores and aggregate cluster-wide values when needed #9855 @SerjKol80

Bug fixes

• TiDB

  • Fix the issue that TiDB might crash with a SIGSEGV due to a race between UNION query cleanup and worker goroutines #66391 @bb7133
  • Fix the issue that TiDB consumes excessive CPU during query planning in some scenarios because of expensive full-range checks #63235 @terry1purcell @qw4990
  • Fix a potential panic during asynchronous statistics loading for newly added indexes when tidb_opt_objective is set to determinate #64274 @0xPoe

• TiKV

  • Fix the issue that TiKV might panic with txn record found but not expected due to a race between external SST ingestion and foreground writes #19891 @gengliqi
  • Fix the issue that stale connections between TiKV and TiCDC might not be fully cleaned up after TiCDC restarts or network failures, exhausting the TiKV CDC memory quota and causing changefeeds to stall #18169 #19610 @asddongmen @wk989898
  • Fix the issue that TiKV throughput might continue to drop during a TiKV I/O hang because PD-related work is blocked by I/O operations #17939 @LykxSassinator
  • Fix the issue that TiKV might crash with a null pointer dereference during TiDB Lightning data import #18671 #18756 @Dog-Du

• PD

  • Fix the issue that PD might experience a goroutine surge and become unstable when TiDB Lightning concurrently calls SetRegionLabelRule during import #9854 @lhy1024
  • Fix the issue that stale pd_cluster_status metrics of removed stores might remain after TiKV scale-in #9942 @okJiang

• TiFlash

  • Fix a potential data inconsistency issue between TiFlash and TiKV after executing a DDL statement to remove the NOT NULL constraint of a column #10680 @JaySon-Huang

• Tools

  • BR

    • Fix the issue that BR leaves stale GC service safepoints after a log backup task is stopped, which might block GC from progressing as expected #19832 @Leavrth

  • TiCDC

    • Fix the issue that retrying failed DDL or checkpoint event delivery in Kafka changefeeds leaks Kafka clients, leading to continuous memory growth #12666 @3AceShowHand
    • Fix the issue that Kafka sinks might leak underlying Sarama clients when admin or producer initialization fails or when admin and synchronous-producer wrappers are closed #12572 @wlwilliamx