What vitessio/vitess shipped

Weekly in Vitess: tablet selection no longer takes process-wide locks per query, semi-sync durability holds through demotion and promotion, UNION limits apply to the statement, and TryLock fails fast. Plus a session insert-id race fix and a new TABLET_THROTTLER strategy.

This week in Vitess: per-query tablet selection and stats lookup no longer take process-wide locks, and TryLock now fails fast instead of check-then-lock. Semi-sync durability is preserved through old-primary demotion, draining and promotion. sql_select_limit on a UNION now applies to the statement rather than each branch, and a session last insert id race is fixed for parallel and streamed scatter routes. MoveTables SwitchTraffic with --initialize-target-sequences matches sequence VSchema names the way VTGate does. VTTablet's query throttler gains a TABLET_THROTTLER strategy based on lag or loadavg, and TabletManager stops its background goroutines when Start fails.

This week: BackupShard can pick the tablet type to back up from, vtbackup catches errant GTIDs and stops at the committed position, vtctld acts once per shard on repeated lists, and vtgate reports REPLACE as REPLACE.

This week in Vitess: BackupShard can now be asked for a specific tablet type instead of always choosing by replication lag. vtbackup adds errant GTID checks and stops catch-up at the primary's committed position, so it no longer applies transactions that are not yet in gtid_executed. vtctld RPCs that take a shard list now act once per shard even when a shard repeats, which applies to vtctldclient, VTAdmin and direct gRPC callers, and vtctldclient Workflow update honors --shards. JSON equality is fixed in compiled and interpreted evalengine paths, vtgate reports planned REPLACE statements as REPLACE, and VReplication runs ANALYZE TABLE on targets created with stats_auto_recalc disabled.

This week: optional --include-binlogs for full builtin backups, vtgate row comparisons with subqueries it can't merge, trailing comments kept before statement terminators, concurrent target shard refresh on Reshard read switches, and fixes for on-demand heartbeats and vtctldclien

This week's changes: full builtin backups can optionally include binary logs via --include-binlogs on vtbackup and vtctldclient Backup and BackupShard, so a restored server can serve binlog streams. vtgate now supports row comparisons with multi-column subqueries it cannot merge, and trailing comments before a statement terminator are kept out of the query. Reshard refreshes target shards concurrently when switching reads. Fixes cover on-demand heartbeats on tablets the primary does not probe, vtctldclient --server=internal dialing, ERS stop-replication, and a single multi-stage Dockerfile for the bootstrap image.

This week: VTAdmin requires execute permission for VExplain modes that run the statement, VReplication cancels the switch when reading source positions fails, VTOrc stops repairing replicas with unreadable MySQL state, and v23.0.8 and v24.0.5 are out.

This week in Vitess: VTAdmin now requires an execute permission for VExplain modes that run the statement. VReplication cancels the traffic switch when reading source positions fails, so writes are no longer left stopped on the source. VTOrc no longer repairs a replica whose MySQL state it can no longer read. The VStream after_data_columns bitmap is now projected onto target expressions, and the SQL parser quotes charset, collation and engine names when writing them back out. Releases v23.0.8 and v24.0.5 are available.

This week in Vitess: bind variables that are not single SQL literals are rejected, only ASCII-safe connection character sets are accepted, // comments are rewritten before reaching MySQL, VTAdmin state-changing routes require POST with RBAC, and buffer fixes avoid lost wakeups.

This week in Vitess: bind variables that are not single SQL literals are now rejected, and only ASCII-safe MySQL connection character sets are accepted, reducing risk from malformed statements and unsafe connections. VTGate rewrites // comments to #/ before statement text reaches MySQL and counts their use. VTAdmin state-changing API routes now require POST and check RBAC, including ReloadSchemaShard. In the buffer, each shard buffer is created only once and the timeout thread's queue-not-empty wakeup is no longer lost. Plan caches also share the all-roles ACL of a table group.

Vitess v24.0.4 and v23.0.7 released with lz4 backup engine fixes, emergency reparenting safeguards, and ORCA load balancing improvements.

Vitess v24.0.4 and v23.0.7 are now available. These releases include a critical upgrade to the lz4 backup engine library fixing amd64 decoding issues, enhanced emergency reparenting with required-position validation to prevent stale replica promotion, ORCA QPS metric reporting for gRPC load balancing, and numerous reliability improvements across replication, connection management, and locking scenarios.

Vitess ships fixes for reparenting validation, system variables in sessions, stored procedure state isolation, table ACL coverage, connection pooling safety, and gRPC proxy compatibility.

This week's Vitess releases address critical operational and security improvements: EmergencyReparentShard now correctly validates promotion candidates, session variables return accurate values, stored procedures can no longer leak state, table ACLs cover more statement types, and connection pooling handles concurrent capacity changes safely. VTAdmin also now works reliably behind proxies.

Shipped: table ACL and TLS security fixes, resolved e2e_race test failures, updated CI infrastructure for current ZooKeeper and S3 services.

This week's releases include important security fixes: VTTablet now properly enforces table ACL for all statement types, CTE name resolution matches MySQL behavior, and TLS certificate revocation checks work across all SSL modes and resumed sessions. We also fixed a data race in transaction timeouts and updated CI infrastructure as upstream projects archive releases.