TiDB 8.5.7 Release Notes

Release date: July 9, 2026

TiDB version: 8.5.7

Quick access: Quick start | Production deployment

Features

Performance

• Support CPU-aware hot Region scheduling to improve read load balancing #5718 #19373 @lhy1024

  In earlier versions, the hot Region scheduler balances read hotspots mainly by query rate and byte throughput. In some workloads, TiKV CPU usage remains uneven even when QPS and byte throughput appear balanced, such as when different queries have very different CPU costs or TiKV nodes have different performance profiles.

  Starting from v8.5.7, TiKV reports read CPU usage for hot Regions in store heartbeats, and PD can use CPU usage as a scheduling dimension for read hot Region scheduling. This mechanism helps PD identify CPU-based read hotspots and balance them across TiKV stores more accurately.

  In addition, PD adds CPU-related hotspot statistics and scheduler controls, including the cpu-read-rate field in hot store statistics and the min-hot-cpu-rate and cpu-rate-rank-step-ratio scheduler configurations.

  For more information, see documentation.

Reliability

• Support limiting the number of connections that a single user can establish on a TiDB instance #59203 @joccau

  Starting from v8.5.7, you can use the max_user_connections system variable to limit the maximum number of connections that a single user can establish to a single TiDB server instance. This helps prevent situations where excessive token consumption by one user delays responses to requests from other users.

  In addition, you can use WITH MAX_USER_CONNECTIONS N in CREATE USER and ALTER USER statements to limit the maximum number of connections that the corresponding user can establish.

  For more information, see documentation.

SQL

• Support partial indexes to reduce index storage and DML maintenance overhead #62664 #62761 #62758 #63447 #64344 @YangKeao @winoros @wjhuang2016

  Starting from v8.5.7, TiDB supports partial indexes, which index only rows that satisfy a predicate defined in the index WHERE clause. You can create a partial index using CREATE INDEX ... WHERE ..., ALTER TABLE ... ADD INDEX ... WHERE ..., or an index definition in CREATE TABLE.

  Partial indexes are useful when you frequently query a subset of rows based on specific conditions or need unique constraints that apply only under specific conditions. Because rows outside the predicate are not written to the index, partial indexes help reduce index storage and can lower index maintenance overhead during INSERT, UPDATE, and DELETE operations.

  To use partial indexes effectively, define a predicate that matches the filters in your common queries. TiDB selects a partial index only when the query predicates match or imply the partial index predicate. Currently, partial index predicates support basic comparison operators (=, !=, <, <=, >, >=), IS NULL, IS NOT NULL, and IN predicates with constant values.

  For more information, see documentation.

Observability

• The Top SQL page in TiDB Dashboard now supports collecting and displaying TiKV network traffic and logical I/O metrics #62916 @yibin87

  In earlier versions, TiDB Dashboard identified Top SQL queries based only on CPU-related metrics, making it difficult to identify performance bottlenecks related to network or storage access in complex scenarios.

  Starting from v8.5.7, you can enable TiKV Network IO collection (multi-dimensional) in the Top SQL settings to view metrics such as Network Bytes and Logical IO Bytes for TiKV nodes. You can also analyze these metrics across multiple dimensions, including By Query, By Table, By DB, and By Region, helping you identify resource hotspots more comprehensively.

  For more information, see documentation.

Data migration

• DM supports foreign key causality for static one-to-one schema/table routing #12350 @OliverS929

  Starting from v8.5.7, DM supports foreign key causality for static one-to-one schema/table routing when foreign_key_checks=1 and syncer.worker-count > 1.

  Before the replication task starts, target schemas and foreign key definitions must be created in the downstream database. This feature does not support many-to-one or shard-merge routing, dynamic foreign key DDL during replication, options that change DML statement boundaries such as syncer.compact or syncer.multiple-rows, or safe-mode UPDATE statements that modify primary key or unique key values during replication. When foreign key causality is enabled, DM does not support hot configuration updates that modify worker-count, case-sensitive, route rules, block-allow-list rules, binlog filter rules, or foreign_key_checks. To change these settings, stop the task, update the configuration, and then restart the task.

  For more information, see documentation.

• TiCDC supports table routing #4655 #4941 #4702 @3AceShowHand

  Starting from v8.5.7, the new TiCDC architecture supports table routing. You can use target-schema and target-table in the sink.dispatchers configuration of a changefeed to map upstream tables to specified downstream database or table names.

  This feature is useful when downstream database and table naming conventions differ from upstream ones, or when you need to replicate multiple source databases to the same target database while keeping target table names unique. With table routing, you can provide stable and expected target database and table names for downstream systems.

  This feature applies only to the new TiCDC architecture. For more information, see documentation.

Compatibility changes

For TiDB clusters newly deployed in v8.5.6 (that is, v8.5.6 clusters that are not upgraded from earlier versions), you can smoothly upgrade to v8.5.7. Most changes in v8.5.7 are safe for routine upgrades, but this release also includes several behavior changes, MySQL compatibility changes, system variable updates, configuration parameter updates, and deprecated features. Before upgrading, make sure to carefully review this section.

Behavior changes

• TiKV now rejects max_ts updates that are confirmed to be invalid by default, instead of only logging them. This change improves safety by preventing invalid timestamp updates without triggering a TiKV panic. To keep the previous log-only behavior, set storage.max-ts.action-on-invalid-update to log #19755 @ekexium
• Starting from v8.5.7, TiDB enables optimizer fix control 52869 by default. This change allows the optimizer to consider IndexMerge automatically when alternative indexes exist, which can change query plans in some cases. #26764 @time-and-fate

MySQL compatibility

• Support parsing the LATERAL syntax for derived tables to improve MySQL 8.0 compatibility, including comma joins, CROSS JOIN LATERAL, and INNER JOIN LATERAL

  Currently, TiDB only supports parsing the LATERAL derived table syntax and does not support executing queries that use this syntax. If you attempt to execute such a query, TiDB returns an error. You can track the progress of full execution capability for this feature in issue #40328.

• Support WITH MAX_USER_CONNECTIONS N in CREATE USER and ALTER USER to improve MySQL compatibility. TiDB also adds the max_user_connections column to mysql.user and lets you use the max_user_connections system variable to control the maximum number of connections that a user can establish to a TiDB server instance.

System variables

Configuration parameters

Compiler versions

• Upgrade the Go compiler version of TiDB from go1.25.8 to go1.25.10, which improves the TiDB performance. If you are a TiDB developer, upgrade your Go compiler version to ensure smooth compilation.
• Upgrade the Rust compiler version of TiKV from nightly-2023-12-28 to nightly-2025-02-28, which improves the TiKV performance. If you are a TiKV developer, upgrade your Rust compiler version to ensure smooth compilation.

Deprecated features

• Starting from v8.5.7, the telemetry feature in TiDB and TiDB Dashboard is deprecated.

Removed features

• Starting from TiDB v8.5.7, TiDB Lightning no longer supports the web interface. #67697 @D3Hunter

  To import data with TiDB Lightning, use the TiDB Lightning command-line tools: tidb-lightning for import tasks and tidb-lightning-ctl for checkpoint and troubleshooting operations.

  For new data import workloads, you can also use the IMPORT INTO statement.

  If this change affects your workflow, comment on #67697.

Improvements

• TiDB

  • Improve the performance of ORDER BY ... LIMIT queries that contain OR and IN conditions. The optimizer can now choose IndexMerge more effectively and supports merge sort for IN condition paths in IndexMerge, enabling Limit pushdown to partial paths and reducing unnecessary row reads and I/O overhead. #65712 @time-and-fate
  • Improve slow query observability by logging client connection attributes in the slow query log and making them queryable in INFORMATION_SCHEMA.SLOW_QUERY and INFORMATION_SCHEMA.CLUSTER_SLOW_QUERY; performance_schema_session_connect_attrs_size now controls attribute truncation, and truncated bytes are recorded in _truncated #66616 @jiong-nba
  • Add the tidb_enable_strict_not_null_check system variable to control whether TiDB enforces strict NOT NULL checks for single-row INSERT statements, helping reduce upgrade risk for workloads that depend on the previous non-strict behavior #68108 @xhebox
  • Improve the performance and stability of runaway query watch handling, including more reliable watch synchronization across TiDB instances and more efficient background flushing and syncing #65746 @JmPotato
  • Add the global system variable tidb_enable_batch_query_region to control whether TiDB uses batched Region queries to PD, improving the efficiency of fetching Region information; this variable is disabled by default #58439 #8690 @JmPotato
  • Improve the optimizer performance for queries on tables with many indexes by pruning irrelevant indexes before cost estimation, reducing query planning time and avoiding unnecessary full-range out-of-range estimation #63856 @terry1purcell @qw4990
  • Enhance the Ping Latency panel in the Blackbox exporter dashboard by adding a Max Ping Latency metric using the max_over_time alert rule. This change aligns the dashboard visualization with the TiDB alerting logic, helping you identify peak latency spikes and verify alert triggers more easily #1071 @yibin87
  • Support partial ordered index optimization for ORDER BY ... LIMIT/OFFSET queries on matching prefix indexes. When tidb_opt_partial_ordered_index_for_topn is set to COST, TiDB can use the partial ordering of indexes to reduce full table scans and improve TOPN query performance #63280 #65813 #66338 @elsa0520 @xzhangxian1008 @winoros
  • Optimize the performance of high-cardinality GROUP BY queries that use Stream Aggregate by reducing CPU overhead in memory tracking #68475 @guo-shaoge
  • Mitigate coprocessor request bursts for IndexLookUp queries on highly partitioned tables with local indexes to improve query stability and reduce performance spikes #67545 @gengliqi
  • Optimize CPU and memory usage for INSERT ... ON DUPLICATE KEY UPDATE statements by reducing unnecessary expression buffer allocations during execution #65003 @windtalker
  • Optimize query planning performance for statements with large IN lists by reducing CPU and memory overhead during range building #67756 @winoros
  • Improve the default performance and consistency of auto ANALYZE by aligning the default values of tidb_auto_build_stats_concurrency and tidb_sysproc_scan_concurrency with manual ANALYZE #67195 @0xTars
  • Add the tidb_opt_enable_alternative_logical_plans system variable to enable alternative logical plan optimization for subquery decorrelation #66676 @AilinKid
  • Add the tidb_enable_cache_prepare_stmt system variable to cache repeated prepared statements in the same session and reduce CPU overhead for prepare-per-request workloads #67815 @guo-shaoge
  • Improve join reordering so that TiDB can handle projections between join groups, reducing unnecessary Cartesian joins and allowing the LEADING hint to take effect in more queries #50229 @Reminiscent
  • Support nested parentheses in the LEADING optimizer hint to specify more complex join orders, such as LEADING((a, b), (c, d)) #63253 @guo-shaoge
  • Improve join plan selection to avoid choosing inefficient index joins when the estimated probe rows are close to a full scan, improving query performance in some HASHAGG + join scenarios #67610 @qw4990
  • Improve query performance for nested OR conditions by enabling more efficient IndexMerge plans and allowing redundant global filters to be removed so LIMIT can be pushed down #65822 @time-and-fate
  • Support the FLUSH STATS_DELTA statement to persist pending optimizer statistics deltas for all, database, or table scopes #65668 @0xPoe
  • Improve query optimization by enabling the optimizer fix control for considering IndexMerge when alternative indexes exist by default, so TiDB can choose IndexMerge plans in more applicable queries #26764 @time-and-fate
  • Support caching prepared and non-prepared queries that use set_var and resource_group hints to improve plan cache hit rates for hinted queries #60920 @qw4990
  • Optimize ORDER BY ... LIMIT and ORDER BY ... TOPN queries that use IndexMerge by pushing Limit or TopN down to individual partial paths when possible, reducing unnecessary scans and sorting in some query plans #68773 @time-and-fate
  • Improve ANALYZE performance on newly bootstrapped clusters by using clustered primary keys for the mysql.stats_* system tables #66751 @0xPoe
  • Improve graceful shutdown handling by returning an error for COM_PING requests so proxies and load balancers can detect that a TiDB server is shutting down and stop sending new connections #58007 @dveeden
  • Accelerate INSERT statements on tables with many generated columns, significantly improving performance for wide-table workloads #67916 @bb7133
  • Support viewing session connection attributes on the Slow Query page in TiDB Dashboard, including an optional list column and a dedicated detail tab, while remaining compatible with earlier TiDB versions that do not provide this field #1899 @yibin87
  • Improve transactional timestamp validation by enforcing exact max-ts checks for non-TiDB requests, reducing the risk that external components use future timestamps incorrectly #68799 @ekexium
  • Optimize auto-commit optimistic transactions by skipping lock resolution on the first execution, reducing latency in high-contention scenarios #58675 @ekexium

• TiKV

  • Add dynamic resource group isolation to protect sustained workloads from traffic spikes by deprioritizing requests from groups that exceed their historical RU baselines and optionally delaying or rejecting over-baseline requests under high CPU load, with new resource-control configuration options that are disabled by default #19607 @mittalrishabh
  • Use a global rate limiter for all TiKV background resource groups and aggregate related background resource-control metrics without the resource_group label, requiring affected dashboards and alerts to be updated #19497 @mittalrishabh
  • Improve TiKV unified read pool scheduling fairness by allowing higher-priority requests to evict lower-priority queued tasks when the queue is full, helping ensure fairer resource allocation across tenants and background traffic #19386 @mittalrishabh
  • Mark PITR-restored Put and Delete write CF records with the physical import transaction source during apply log processing so TiCDC can ignore PITR-imported data #19669 @YuJuncen
  • Add automatic fail-fast exit for TiKV nodes that detect disk I/O hangs to speed up recovery from unresponsive stores #19626 @hbisheng
  • Improve TiKV stability and security by upgrading vulnerable third-party dependencies for TiKV 8.5 and aligning the required compatibility fixes with upstream #19713 @LykxSassinator
  • Support rank-based limit processing in TiKV to return more accurate results for queries that use truncate key expressions #19388 @xzhangxian1008
  • Improve load-based Region splitting in TiKV by relaxing the default split thresholds and adding observability for split-key selection and CPU fallback decisions, helping operators handle hot Regions more effectively #18932 @lhy1024

• PD

  • Add PD maintenance endpoints and pd-ctl commands to serialize TiKV maintenance tasks and prevent Raft quorum loss by ensuring that only one maintenance task is active at a time #9477 @SerjKol80 @HaoW30
  • Disable split scatter by default in PD to avoid unexpected scheduling after Region splits; you can still enable it by setting schedule.split-scatter-schedule-limit to a positive value #10592 @lhy1024
  • Optimize unsafe recovery empty-region plan generation to improve performance and reduce timeout risk in large clusters with many Regions and gaps #10638 @Connor1996
  • Improve PD transaction duration metrics to better reflect production latency distributions and improve observability in dashboards and alerts #10705 @bufferflies

• TiFlash

  • Remove unnecessary gRPC connections in TiFlash to reduce repeated invalid connection logs after TiKV nodes are scaled in #9806 @gengliqi

• Tools

  • Backup & Restore (BR)

    • Support using Workload Identity Federation to access Google Cloud Storage backup buckets in BR #19442 @Leavrth
    • Mitigate the impact of BR backup and restore on online DDL by loading only the necessary database schema information, reducing schema reload overhead and DDL blocking time #64833 @YuJuncen
    • Improve PITR compatibility with TiCDC by marking PITR-restored write CF entries as physical import transactions so TiCDC can ignore PITR-imported data #68660 @YuJuncen
    • Add the --region-scan-concurrency parameter to limit the number of Region scan requests that BR sends to PD at the same time, improving restore stability when many Region scans are issued #66821 @Leavrth

  • TiCDC

    • Optimize TiCDC event store write and iterator paths to reduce memory allocations and improve event processing performance #4928 @lidezhu
    • Remove TiCDC's dependency on tidb_ddl_history for capturing DDLs and rely on tidb_ddl_job to support accelerated table creation #2272 @wlwilliamx
    • Optimize TiCDC hot paths to reduce CPU overhead and improve replication performance in high-throughput scenarios #5107 @lidezhu
    • Add redo checkpoint and resolved timestamp metrics to improve the observability of TiCDC redo log progress #5264 @wk989898
    • Mitigate high etcd pressure when many TiCDC changefeeds enter the warning state by avoiding unnecessary persistence of unchanged runtime states #5268 @wk989898
    • Enhance TiCDC monitoring by showing the error occurrence time in the Changefeed Error Details Grafana panel to help operators diagnose changefeed failures more efficiently #5085 @wlwilliamx
    • Mitigate TiCDC incremental scan CPU spikes by limiting split table merge scheduling to at most 8 merge operators per group in each checker run #5047 @hongyunyan
    • Improve TiCDC scheduling responsiveness and callback correctness for the cloud storage sink by acknowledging DML events when they enter the write pipeline instead of waiting for flush completion, reducing wake latency while keeping checkpoint semantics unchanged #4269 @3AceShowHand
    • Reduce TiCDC bootstrap head-of-line blocking across changefeeds to improve changefeed creation and recovery throughput when sink initialization is slow #5139 @hongyunyan
    • Improve TiCDC Kafka sink stability by retrying transient producer send failures by default #1920 @3AceShowHand
    • Add the max-retry Kafka sink URI parameter and enable bounded retries for transient Kafka producer send failures by default to improve TiCDC Kafka sink stability #12655 @3AceShowHand
    • Add a local spool to the TiCDC cloud storage sink to stage accepted encoded DML locally before flushing to external storage, reducing memory pressure and improving stability when object storage is slow #3745 @3AceShowHand
    • Support replicating DDL statements for partial indexes #3698 @YangKeao
    • Add metrics for the TiCDC blackhole sink to improve observability of changefeed replication and DDL processing #5362 @wk989898
    • Optimize TiCDC log puller performance to reduce CPU usage and memory allocations when processing resolved-ts-heavy workloads, especially for small batches #4697 @asddongmen
    • Optimize TiCDC MySQL sink conflict detection to improve replication performance #4582 @wk989898
    • Improve TiCDC resolve lock observability and avoid duplicate resolve attempts for the same Region to reduce unnecessary lock resolution work and make troubleshooting easier through new metrics and dashboards #5016 @lidezhu
    • Add a Changefeed Operation History panel in TiCDC Grafana dashboards to help investigate recent user-initiated changefeed operations such as create, update, pause, resume, and delete #5087 @wlwilliamx
    • Support parsing and replicating DDL statements that use partial indexes, including CREATE TABLE, CREATE INDEX, and ALTER TABLE ... ADD INDEX with WHERE clauses #12503 @YangKeao
    • Support table routing for the MySQL sink so routed changefeeds apply DDL and DML to the target schema and table names #4818 @3AceShowHand
    • Add a TiCDC metric to display the error information of changefeeds in the warning or failed state, making it easier to diagnose problems through monitoring instead of logs #4498 @wk989898

  • TiDB Lightning

    • Support automatically detecting the file format from the file name in TiDB IMPORT INTO when FORMAT is not specified, simplifying data import for CSV, SQL, and Parquet files #59540 @JQWong7

Bug fixes

• TiDB

  • Fix the issue that TiDB does not close disconnected client connections promptly when a query is still running, causing the connection to remain in SHOW PROCESSLIST until the statement finishes #57531 @Defined2014
  • Fix the issue that stale Region cache entries might continue to be used after RegionNotFound errors, causing repeated retries, delayed Region reloads, and extra cross-AZ traffic #1892 #69197 @ekexium
  • Fix the potential crash that occurs when evaluating vectorized ILIKE with constant string arguments #67001 @zanmato1984
  • Fix the issue that point UPDATE statements might use assignment conversion semantics that differ from normal UPDATE statements when assigning negative values to unsigned numeric columns or integer values to SET columns, causing inconsistent results, out-of-range errors, or MySQL compatibility issues #63455 #67534 @fzzf678
  • Fix the issue that TiDB might crash with SIGSEGV during concurrent query execution in rare cases #66391 @bb7133
  • Fix the issue that queries using user variables with different casing might generate suboptimal execution plans and fail to use index range scans #66339 @qw4990
  • Fix the issue that TiDB might run out of memory when multiple sessions concurrently hit global bindings and corrupt the global binding cache #68015 @qw4990
  • Fix the issue that TiDB might return incorrect results for queries that use outer joins with conditions that are sensitive to NULL values. The issue occurs because the optimizer incorrectly determines whether an outer join can be simplified to an inner join. Affected scenarios include WHERE clauses that contain predicates or expressions such as OR, IS NULL, COALESCE(), NULLIF(), CAST(), or IN (NULL, ...), as well as outer join queries involving derived tables or UNION ALL. This issue might cause incorrect results, missing rows, empty results, or unexpected rows that contain NULL values. #58793 #59162 #60080 #60081 #60370 #61327 #66824 #66825 #67330 #67373 @winoros
  • Fix the issue that prepared statements on outer joins might skip the prepared plan cache when the null-reject check is independent of parameter values #67048 @winoros
  • Fix the issue that TiDB might log unnecessary warnings about asynchronous index histogram loading during query planning for full-range index scans #64791 @terry1purcell
  • Fix the issue that plans generated when synchronous statistics loading times out and falls back to pseudo or partial statistics might be cached and reused after the required statistics are loaded #66585 @winoros
  • Fix the issue that TiDB might generate an incorrect join order for outer joins and return wrong query results #63887 @guo-shaoge
  • Fix the issue that TiDB might choose a suboptimal join order for some queries involving outer joins, which can lead to less efficient execution plans #67774 @AilinKid
  • Fix the issue that tidb_ignore_inlist_plan_digest might be unexpectedly reset during upgrade when its row is missing from mysql.global_variables #68136 @qw4990
  • Fix the issue that TiDB might choose an inefficient MergeJoin plan with an underestimated IndexFullScan cost when queries with ORDER BY and LIMIT are optimized, which can cause poor query performance #67595 @qw4990
  • Fix the issue that queries using expression indexes with duplicate expressions might return the Unexpected missing column error because TiDB resolves the wrong hidden generated column #67552 @AilinKid
  • Fix the issue that ANALYZE might not stop promptly after KILL QUERY or cancellation, causing the analyze job to hang #65818 @hawkingrei
  • Fix the issue that manual ANALYZE might become slower after upgrading a cluster from versions earlier than v7.6 because tidb_analyze_distsql_scan_concurrency is not initialized from the existing tidb_distsql_scan_concurrency setting #65423 @winoros
  • Fix the issue that TiDB cannot convert some non-correlated IN subqueries into correlated execution plans, which can prevent index lookups from being used and lead to less efficient query plans #66320 @terry1purcell
  • Fix the issue that queries using CAST(... AS BINARY) on indexed string columns might use an index full scan instead of an index range scan #67899 @terry1purcell
  • Fix the potential panic that occurs when executing ANALYZE TABLE on indexed stored generated columns that depend on skipped column types such as JSON #66359 #66918 @xhebox
  • Fix the issue that TiDB might generate incorrect join orders for queries with outer joins, which can lead to incorrect query results #67290 @guo-shaoge
  • Fix the issue that SHOW ANALYZE STATUS displays a negative Remaining_seconds value when the session time_zone is ahead of UTC #67230 @0xPoe
  • Fix the potential TiDB server crash that occurs during ANALYZE on partitioned tables, including auto-analyze, and reports fatal error: concurrent map read and map write #68457 @mjonss
  • Fix the issue that partial indexes might be incorrectly used to enforce foreign keys, which can cause deletes or updates on parent rows to miss matching child rows and leave orphan rows #68587 @YangKeao
  • Fix the issue that INSERT statements on tables without generated columns suffer a performance regression, reducing throughput for common workloads #68129 @bb7133
  • Fix the issue that CONCAT_WS() returns the ERROR 3854 error instead of a binary result when it combines utf8mb4_0900_bin columns with BLOB columns #68845 @tiancaiamao
  • Fix the issue that autocommit INSERT, UPDATE, and DELETE statements experience a performance regression because of connection monitoring overhead #68633 @King-Dylan
  • Fix the issue that GRANT and REVOKE on mixed-case schema names might fail, create duplicate privilege rows, or fail to match existing privilege rows when new_collation_enabled is disabled; after the fix, such mixed-case identifiers map to the same privilege record #66867 #68406 @expxiaoli
  • Fix the issue that autocommit write statements might not be interrupted promptly after the client disconnects #68236 @King-Dylan
  • Fix the issue that non-TiDB transaction-related requests might incorrectly pass future timestamp validation and cause inconsistent behavior in ScanLock, backup range, CheckTxnStatus, Cleanup, and CheckSecondaryLocks #19656 @ekexium
  • Fix the issue that foreign key cascade updates in pessimistic transactions might fail with use Op::SharedLock to prewrite on a shared lock when tidb_foreign_key_check_in_shared_lock is enabled #68133 @wfxr

• TiKV

  • Fix the issue that the resolved_ts module consumes excessive memory when the number of Regions is large #19535 @glorv
  • Fix the issue that TiKV might start the next compaction round immediately after a long-running round when MVCC read-aware compaction is enabled, resulting in insufficient statistics collection for load-based compaction #19362 @mittalrishabh
  • Fix the issue that TiKV memory usage increases over time in stable workloads when using raft-engine #19544 @LykxSassinator
  • Fix the issue that manually evicting Regions from the TiKV in-memory engine leaves Regions stuck in the Evicting state and prevents them from being automatically loaded again #19584 @overvenus
  • Fix the issue that TiKV consumes excessive memory in raft message queues when a store uses multiple gRPC raft connections #19542 @glorv
  • Fix the issue that aborted CPU profiling requests in TiKV leave profiling stuck in the active state, causing subsequent profiling requests to fail with Already in CPU Profiling #19703 @hujiatao0
  • Fix the issue that TiKV might retain excessive memory in the Raft log after entries are persisted, causing abnormally high memory usage on replicas #19593 @LykxSassinator
  • Fix the issue that TiKV might enter an infinite retry loop and repeatedly log invalid store ID ..., not found errors when resolving the address of a removed tombstone store from PD #17875 @LykxSassinator
  • Fix the issue that TiKV GCS backups might fail when the access token is close to expiry during long-running uploads #19659 @RidRisR
  • Fix the issue that the TiKV apply_msg_len metric has no data points during normal traffic, which affects monitoring of apply message length distribution #18800 @squalfof
  • Fix the potential panic that occurs when PD merges a Region immediately after a split because TiKV reports incorrect pending peer information in Region heartbeats #17992 @hbisheng
  • Fix the issue that the TiKV In-Memory Engine debug API /debug/ime/cached_regions cannot be accessed #19546 @glorv
  • Fix the issue that the server.grpc_memory_pool_quota configuration in TiKV cannot be changed dynamically #19104 @glorv
  • Fix the issue that TiKV does not throttle background traffic when online traffic uses most CPU resources, which can increase online request latency under high load #19401 @mittalrishabh

• PD

  • Fix the issue that PD resource control rate limiting might become weaker under high concurrency because the token bucket accrues excess tokens #10744 @YuhaoZhang00
  • Fix a goroutine leak in the resource group client controller when token bucket notification timers are reset #9745 @lhy1024
  • Fix the issue that SQL requests to resource groups might experience about 1 second latency spikes when the configured RU fill rate is much higher than the actual RU consumption #10251 @JmPotato
  • Fix the issue that PD affinity scheduling might treat a Region as properly replicated even when the highest isolation level required by placement rules is not satisfied, which can lead to incorrect replica placement decisions #10149 @HunDunDM
  • Fix the potential PD panic that occurs when Region heartbeat breakdown metrics encounter a brief backward jump in the host's monotonic clock, causing the counter cannot decrease in value error #10901 @JmPotato
  • Fix the issue that PD incorrectly reports affinity checker operator limit metrics when no affinity groups are configured #10687 @lhy1024
  • Fix the issue that PD followers cannot serve read-only Region HTTP APIs from their local synced Region cache unless the PD-Allow-Follower-Handle: true header is explicitly set #10681 @okJiang
  • Fix the issue that a PD follower might stop receiving Region updates when the PD leader fails to send Region sync responses or closes the sync stream during shutdown #10684 @okJiang
  • Fix a Region syncer issue that might leave a PD follower waiting on a stale stream after the PD leader fails to send updates and removes the stream from memory #10666 @okJiang
  • Fix the issue that PD might keep scheduling redundant subtree update tasks after flow-only Region heartbeat updates, increasing memory pressure in large clusters #10722 @JmPotato

• Tools

  • Backup & Restore (BR)

    • Fix the issue that BR restore still fails version compatibility checks when --check-requirements=false is specified, preventing restore from proceeding even though version checking is explicitly disabled #67402 @RidRisR
    • Fix the issue that BR log truncate might fail in S3-compatible storage when the lock file is at the storage root #65897 @YuJuncen
    • Fix the issue that BR point-in-time restore might run out of memory during metadata restoration when the backup contains many mDB:* meta keys across multiple versions #67196 @vldmit
    • Fix the issue that BR might fail to access a custom AWS S3 endpoint when AWS FIPS endpoint mode is enabled #68966 @v01dstar
    • Fix the issue that BR physically restores the mysql.user table during snapshot restore when the target cluster has a different column count, which can overwrite the newer table schema instead of falling back to logical restore #68861 @Leavrth
    • Fix the issue that br operator base64ify does not preserve the S3 Object Lock status in the generated storage backend #68551 @YuJuncen
    • Fix the issue that BR log backup checkpoint advancement might get stuck when TiKV does not respond to a flush subscription request #68411 @Leavrth
    • Fix the issue that BR point-in-time restore with table filters might create duplicate databases or modify the target database unexpectedly when restoring tables from the same database multiple times #68908 @Leavrth
    • Fix the issue that BR log backup metrics for the current last Region ID and leader store ID are missing from the /metrics endpoint #62839 @YuJuncen
    • Fix the issue that BR log backup to S3 might get stuck under heavy write pressure because multipart uploads exceed the 10000-part limit #19162 @vldmit
    • Fix the issue that BR silently loses unrecognized backup metadata such as merge_option attributes when restoring a backup created by a newer BR version with an older restore tool. Now BR reports the compatibility issue during requirement checks #67016 @JoyC-dev
    • Fix the issue that BR log backup might leak external backup directories or leave stale read locks during truncate or migration loading failures #67819 @RidRisR
    • Fix the issue that BR S3 permission checks might fail when credentials are scoped to the configured prefix, causing backup or restore tasks to fail before starting #68583 @YuJuncen
    • Fix the issue that BR might not write the PiTR blocklist when resuming a restore from a checkpoint after a failed restore with log backup enabled #68171 @Leavrth
    • Fix the issue that a BR log backup task startup hangs when initializing more than 32768 Regions #19615 @YuJuncen
    • Fix the issue that BR restore might fail with the Transaction is too large error when logically restoring system tables by adding the --txn-total-size-limit parameter to adjust the transaction memory quota #66806 @Leavrth

  • TiCDC

    • Fix the issue that TiCDC Grafana dashboards display overlapping panels and incorrect section ordering #4508 @lidezhu
    • Fix the issue that TiCDC does not report the actual error when event iterator creation or the first read fails, which can hide the root cause of event scan failures #5005 @lidezhu
    • Fix the issue that TiCDC Kafka changefeeds might leak Kafka client instances and cause memory usage to keep increasing when sending DDL events or checkpoints fails and retries are triggered #12666 @3AceShowHand
    • Fix the issue that TiCDC incremental scans use excessive CPU and advance checkpoints slowly when scanning insert-like prewrite locks with missing old values #19565 @zier-one
    • Fix the issue that TiCDC might scan data below the checkpoint after a dispatcher reset, which might cause a panic in the event store #4492 @lidezhu
    • Fix the issue that TiCDC might leak connections, storage handles, or background goroutines when component initialization fails #4516 @wk989898
    • Fix the issue that TiCDC changefeeds might stall when the available memory quota is 0 due to a deadlock in the event processing pipeline #4899 @lidezhu
    • Fix the issue that TiCDC might skip downstream remove-only cleanup when a later changefeed removal request arrives while an earlier close request is still being processed #4825 @hongyunyan
    • Fix the issue that TiCDC MySQL-compatible sinks get stuck when tidb_cdc.ddl_ts_v1 is missing and the downstream returns nonstandard missing-table error codes #5003 @hongyunyan
    • Fix the issue that TiCDC might leak Kafka client connections and background resources when Kafka sink initialization fails or the sink is closed #12572 @wlwilliamx
    • Fix the issue that TiCDC resolved-ts lag and CPU usage periodically increase when replicating a large number of active tables #4887 @lidezhu
    • Fix the issue that TiCDC replication might stall after a local EventService removes a dispatcher and the collector re-registers it locally #5088 @lidezhu
    • Fix the potential panic that occurs when TiCDC initializes a Pulsar sink and the producer initialization fails #4937 @wk989898
    • Fix the issue that TiCDC might keep inaccurate memory accounting after a path is removed, causing incorrect pause or release feedback in dynstream #4644 @asddongmen
    • Fix the issue that TiCDC exits with code 1 during graceful SIGTERM shutdown #4563 @pingyu
    • Fix the issue that TiCDC might fail to start the redo log sink because redo configuration is not initialized correctly #4512 @3AceShowHand
    • Fix the potential panic that occurs when TiCDC applies redo DDL events for columns with time-related default values such as CURRENT_TIMESTAMP #4699 @wk989898
    • Fix the issue that TiCDC might load changefeeds from another cluster after restart when multiple TiCDC clusters share the same PD/etcd and their cluster IDs have a prefix relationship #4756 @wk989898
    • Fix a data race in TiCDC redo readiness during bootstrap and failover that might cause failover tasks to report DATA RACE warnings #4402 @3AceShowHand
    • Fix the issue that TiCDC fails to replicate cross-database RENAME TABLE operations when the source database name is not specified #4424 @lidezhu
    • Fix the issue that TiCDC changefeeds might fail or get stuck when the log puller encounters a transient gRPC EOF error #4880 @lidezhu
    • Fix the issue that TiCDC changefeeds initialize slowly and experience high maintainer latency when replicating a very large number of tables #4951 @hongyunyan
    • Fix the issue that TiCDC CLI does not return an error when querying a non-existent changefeed #4648 @wk989898
    • Fix the issue that calling the TiCDC unsafe/service_gc_safepoint API closes the PD client unexpectedly #4638 @wk989898
    • Fix the potential panic that occurs when TiCDC retries maintainer bootstrap after a bootstrap failure, so the changefeed reports the original error instead of crashing #4509 @wk989898
    • Fix the issue that TiCDC might leave shared locks unresolved during stale lock resolution, which can affect shared-lock compatibility in nextgen deployments #5206 @wfxr
    • Fix the issue that TiCDC might unexpectedly reschedule a dispatcher after it has been removed #4874 @wlwilliamx
    • Fix the issue that TiCDC might leave a data replication gap after a restart when a recreated dispatcher starts from a stale checkpoint #3846 @hongyunyan
    • Fix the issue that TiCDC might hang and become unresponsive when processing coordinator messages #4440 @wk989898
    • Fix the issue that TiCDC create, pause, and remove changefeed operations might hang for a long time or remain stuck after the request context is canceled #4417 @wk989898
    • Fix the potential panic that occurs when TiCDC handles fast Region errors or replication worker shutdown during log pulling #4472 @lidezhu
    • Fix the issue that TiCDC might spend excessive time handling duplicate block status requests under syncpoint-enabled workloads with frequent DDLs, causing maintainer slow logs and barrier processing delays #4957 @hongyunyan
    • Fix the issue that TiCDC cli changefeed list might show changefeeds from different clusters after the etcd cluster membership changes #5137 @wk989898
    • Fix the issue that TiCDC thread pool shutdown might hang when tasks are submitted or rescheduled during shutdown #4640 @wk989898
    • Fix the issue that TiCDC waits about 5 seconds before advancing a barrier for some DDLs such as CREATE TABLE ... LIKE ... when a dispatcher's WAITING status is temporarily ignored by the maintainer #4810 @zier-one
    • Fix the issue that TiCDC changefeed lag increases and CPU usage becomes high after pausing and resuming many changefeeds #4653 @lidezhu
    • Fix the issue that TiCDC fails to replicate cross-database CREATE TABLE ... LIKE statements when the source table schema is not explicitly specified #5025 @lidezhu
    • Fix the issue that TiCDC might replicate cross-schema CREATE VIEW statements incorrectly when the view definition uses unqualified source table names, causing downstream replication to fail or the view to reference the wrong table #5026 @lidezhu
    • Fix the issue that TiCDC uses incorrect etcd endpoints after PD nodes are scaled in or out, causing removed PD members to be added back repeatedly #12368 @wk989898
    • Fix the issue that TiCDC redo dispatchers might get stuck in the Initializing state when an invalid global checkpoint is committed, causing redo metadata to stop advancing #4703 @hongyunyan
    • Fix the issue that TiCDC changefeed initialization is slow when handling a very large number of tables #5014 @lidezhu
    • Fix the issue that TiCDC might leave a stale cluster-level service GC safepoint in PD after the last changefeed is removed, which might block upstream GC progress #4610 @hongyunyan
    • Fix the issue that TiCDC might fail to decode or incorrectly handle multi-table DDL events when MultipleTableInfos is non-empty #4415 @asddongmen
    • Fix the issue that TiCDC corrupts partition names containing escaped backticks when replicating EXCHANGE PARTITION DDL statements #4450 @lidezhu
    • Fix the issue that TiCDC might generate out-of-order checkpoint updates in the event store during concurrent subscription checkpoint updates, causing inconsistent replication progress tracking #4992 @lidezhu
    • Fix the issue that TiCDC might deadlock during shutdown and block cleanup when the EventCollector is still receiving messages #4434 @wk989898
    • Fix the issue that TiCDC might generate inconsistent index names in the downstream when replicating anonymous ADD INDEX DDL statements, especially during retries or CREATE TABLE LIKE replication #2327 @wk989898
    • Fix the issue that TiCDC might keep stale CDC connections after a connection watchdog abort, delaying connection cleanup and causing changefeed replication to get stuck when sink memory is full #19610 @wk989898
    • Fix the issue that a re-registered capture in TiCDC might be removed again after it comes back online because of a stale delayed-removal tombstone #4695 @hongyunyan

  • Dumpling

    • Fix the issue that Dumpling might overwrite chunk files and produce incomplete dump results when --rows/-r or --filesize/-F is used with an --output-filename-template that does not include a standalone {{.Index}} outside conditional blocks #68611 @D3Hunter