What pingcap/tidb shipped

This week: column-level privileges are enforced at query time, snapshot backup gains a repository layout, a new tidb_max_keys_read limit, log redaction on by default, and planner fixes for joins, multi-valued indexes and LIKE filters.

This week the project shipped a set of correctness-focused changes. Column-level privileges such as SELECT on a single column are now enforced at query time rather than only stored and displayed. Snapshot backup supports a repository layout so one storage prefix can hold multiple snapshots. A new tidb_max_keys_read setting caps storage-engine keys read by SELECT statements, and log redaction is now on by default for TiDB and Lightning. On the materialized view side, auto analyze defers views still being built, nested views have a create-time depth limit, and column type changes propagate to dependent views. Planner and DDL fixes cover join reordering, LIKE filter retention, TRUNCATE PARTITION cleanup for global multi-valued indexes, and retryable backfill errors that previously produced pa

This week: automatic materialized view maintenance wired into the running service, a scheduler for refresh and purge tasks, and MV comparison support. Also: DROP DATABASE streams table metadata, EXCHANGE PARTITION raises ID allocators, and reverse sort keeps NULL ordering.

This week in TiDB: materialized view maintenance moved closer to production, with a scheduling engine that dispatches refresh and purge tasks and wiring into the running service. Comparison support lets users check an MV against the source-query snapshot of its last successful refresh. Elsewhere, DROP DATABASE now streams table metadata instead of materializing every table, EXCHANGE PARTITION raises ID allocators to avoid reuse, and the planner preserves NULL ordering for reverse sort expressions. Diagnostic logs now redact keys, values and SQL in more places.

This week: DROP DATABASE streams table metadata to cut memory spikes, NextGen add-index retries after TiKV disk full, REORGANIZE PARTITION no longer leaves broken global index copies, and materialized view refresh gets executor builders and SQL entry points.

This week in TiDB: DROP DATABASE now streams table metadata instead of materializing every table, reducing memory spikes on schemas with very many tables. NextGen add-index jobs retry automatically after TiKV disk-full errors rather than waiting for a manual resume. REORGANIZE PARTITION no longer leaves global index copies behind on cancel or broken after completion, and ADMIN CHECK TABLE routing for multi-valued indexes is fixed. DISTRIBUTE TABLE works on keyspace clusters, session migration preserves collation, isolation mode and read timestamps, and the graceful shutdown drain timeout is configurable. Materialized view refresh also gained executor builders, mlog writes and SQL entry points.

This week: a read-only RECOMMEND INDEX RETURN RESULT FOR statement, basic geospatial design with ST_* functions, keyspace-scoped affinity groups, and tidb_dml_max_execution_time for transactional DML. Plus sampling-based NDV analysis and partition fixes for multi-table UPDATE and

What shipped in TiDB this week: a new read-only RECOMMEND INDEX RETURN RESULT FOR statement that returns index advisor recommendations without saving them, and a basic geospatial design that adds value representation and ST_* functions. Affinity group IDs are now scoped by keyspace so tables in different keyspaces no longer collide, and tidb_dml_max_execution_time caps total elapsed time for transactional DML. ANALYZE gains sampling-based NDV estimation through NDVRATE to reduce cost on large tables. Correctness fixes cover global index backfill handle equality, multi-table DELETE and UPDATE on nonclustered partitioned tables, and a nil pointer dereference in system variable handling for internal sessions.

This week: WITH CHECK OPTION parses without CASCADED or LOCAL, LINES STARTING BY works in SELECT INTO OUTFILE, credentials are redacted in logs, and LOAD STATS keeps FM sketches for correct global NDV.

This week in the project: CREATE VIEW... WITH CHECK OPTION now parses without CASCADED or LOCAL, matching MySQL, and SELECT INTO OUTFILE honors LINES STARTING BY. JSON double formatting was fixed inside containers and for non-ASCII path keys, and materialized view cleanup now rolls back the transaction on failure. Statistics improved too: ANALYZE derives NDV from schema-guaranteed uniqueness and LOAD STATS keeps FM sketches, so a later ANALYZE of one partition no longer merges a wrong global NDV. Storage logs and errors now redact credentials, IMPORT INTO retries transient OSS credential fetch failures, and on the nextgen kernel ADD INDEX requires global sort when security enhanced mode is enabled.

This week: fixed SELECT privilege bypass, enabled plan caching for COALESCE updates, improved index selection logic, added diagnostic mode guards for RPC clients, and fixed client-go cancellation race.

New fixes and improvements in TiDB this week: resolved a SELECT privilege bypass vulnerability, enabled plan caching for non-prepared UPDATE statements with COALESCE, refined index selection in query planning to prevent unnecessary seeks, added diagnostic-mode guards for RPC clients, and fixed a client-go cancellation race that could block BatchCommands operations. These changes enhance security, query performance, and cluster stability.

TiDB v26.3.18 and recent releases ship fixes for statement latency attribution, distributed add-index error propagation, and statement summary data races. Materialized view purge scheduling and adaptive LIMIT scans improve resource efficiency.

This week's TiDB updates focus on reliability and efficiency improvements. We fixed latency attribution in multi-statement queries to ensure accurate performance metrics, resolved data races in the statement summary system, and enhanced DDL resource accounting for data reorganization jobs. New features include materialized view log purge execution and adaptive admission control for LIMIT scans, which reduces unnecessary data reads and improves overall query efficiency.

This week: Lightning adds OSS support, prepared statements fix CTE accumulation bugs, and global paging budgets apply to existing sessions. Multiple test flakiness issues resolved.

This week's updates focus on reliability and feature completeness. Lightning now accepts OSS URLs natively, resolving data source configuration limitations. Session handling improvements fix issues where failed authentication could leave connections in invalid states. Query optimization fixes prevent CTE consumer counts from accumulating across prepared statement reprepares. Global paging byte budgets now apply uniformly to existing sessions. Multiple test flakiness issues stabilized.