Release of Vitess v21.0.0

Summary

Table of Contents

• Major Changes
  • Deprecations and Deletions
    • Deprecated VTTablet Flags
    • Deletion of deprecated metrics
    • Deprecated Metrics
  • Traffic Mirroring
  • New VTGate Shutdown Behavior
  • Tablet Throttler: Multi-Metric support
  • Allow Cross Cell Promotion in PRS
  • Support for recursive CTEs
  • VTGate Tablet Balancer
  • Query Timeout Override
  • New Backup Engine
  • Dynamic VReplication Configuration
  • Reference Table Materialization
  • New VEXPLAIN Modes: TRACE and KEYS
  • Errant GTID Detection on VTTablets
  • Automatically Replace MySQL auto_increment Clauses with Vitess Sequences
  • Experimental MySQL 8.4 support
  • Current Errant GTIDs Count Metric
  • vtctldclient ChangeTabletTags

Major Changes

Deprecations and Deletions

Deprecated VTTablet Flags

• queryserver-enable-settings-pool flag, added in v15, has been on by default since v17. It is now deprecated and will be removed in a future release.

Deletion of deprecated metrics

The following VTOrc metrics were deprecated in v20. They have now been deleted.

Deprecated Metrics

The following metrics are now deprecated and will be deleted in a future release, please use their replacements.

Traffic Mirroring

Traffic mirroring is intended to help reduce some of the uncertainty inherent to MoveTables SwitchTraffic. When traffic mirroring is enabled, VTGate will mirror a percentage of traffic from one keyspace to another.

Mirror rules may be enabled through vtctldclient with MoveTables MirrorTraffic. For example:

$ vtctldclient --server :15999 MoveTables --target-keyspace customer --workflow commerce2customer MirrorTraffic --percent 5.0

Mirror rules can be inspected with GetMirrorRules.

New VTGate Shutdown Behavior

We added a new option to VTGate to disallow new connections while VTGate is shutting down, while allowing existing connections to finish their work until they manually disconnect or until the --onterm_timeout is reached, without getting a Server shutdown in progress error.

This new behavior can be enabled by specifying the new --mysql-server-drain-onterm flag to VTGate.

You can find more information about this option in the RFC.

Tablet Throttler: Multi-Metric support

Up until v20, the tablet throttler would only monitor and use a single metric. That would be replication lag, by default, or could be the result of a custom query. In this release, we introduce a major redesign so that the throttler monitors and uses multiple metrics at the same time, including the above two.

The default behavior now is to monitor all metrics, but only use lag (if the custom query is undefined) or the custom metric (if the custom query is defined). This is backwards-compatible with v20. A v20 PRIMARY is compatible with a v21 REPLICA, and a v21 PRIMARY is compatible with a v20 REPLICA.

However, it is now possible to assign any combination of one or more metrics for a given app. The throttler would then accept or reject the app's requests based on the health of all assigned metrics. We have provided a pre-defined list of metrics:

• lag: replication lag based on heartbeat injection.
• threads_running: concurrent active threads on the MySQL server.
• loadavg: per core load average measured on the tablet instance/pod.
• custom: the result of a custom query executed on the MySQL server.

Each metric has a default threshold which can be overridden by the UpdateThrottlerConfig command.

The throttler also supports the catch-all "all" app name, and it is thus possible to assign metrics to all apps. Explicit app to metric assignments will override the catch-all configuration.

Metrics are assigned a default scope, which could be self (isolated to the tablet) or shard (max, aka worst value among shard tablets). It is further possible to require a different scope for each metric.

Allow Cross Cell Promotion in PRS

Up until now if the users wanted to promote a replica in a different cell from the current primary using PlannedReparentShard, they had to specify the new primary with the --new-primary flag.

We have now added a new flag --allow-cross-cell-promotion that lets PlannedReparentShard choose a primary in a different cell even if no new primary is provided explicitly.

Experimental support for recursive CTEs

We have added experimental support for recursive CTEs in Vitess. We are marking it as experimental because it is not yet fully tested and may have some limitations. We are looking for feedback from the community to improve this feature.

VTGate Tablet Balancer

When a VTGate routes a query and has multiple available tablets for a given shard / tablet type (e.g. REPLICA), the current default behavior routes the query with local cell affinity and round robin policy. The VTGate Tablet Balancer provides an alternate mechanism that routes queries to maintain an even distribution of query load to each tablet, while preferentially routing to tablets in the same cell as the VTGate.

The tablet balancer is enabled by a new flag --enable-balancer and configured by --balancer-vtgate-cells and --balancer-keyspaces.

See the RFC for more details on the design and configuration of this feature.

Query Timeout Override

VTGate sends an authoritative query timeout to VTTablet when the QUERY_TIMEOUT_MS comment directive, query_timeout session system variable, or query-timeout flag is set. The order of precedence is: comment directive > session variable > VTGate flag. VTTablet overrides its default query timeout with the value received from VTGate. All timeouts are specified in milliseconds.

When a query is executed inside a transaction, there is an additional nuance. The actual timeout used will be the smaller of the transaction timeout and the query timeout.

A query can also be set to have no timeout by using the QUERY_TIMEOUT_MS comment directive with a value of 0.

Example usage: select /*vt+ QUERY_TIMEOUT_MS=30 */ col from tbl

New Backup Engine (EXPERIMENTAL)

We are introducing a new backup engine for logical backups in order to support use cases that require something other than physical backups. This feature is experimental and is based on MySQL Shell.

The new engine is enabled by using --backup_engine_implementation=mysqlshell. There are other options that are required, so please read the documentation to learn which options are required and how to configure them.

Dynamic VReplication Configuration

Previously, many of the configuration options for VReplication Workflows had to be provided using VTTablet flags. This meant that any change to VReplication configuration required restarting VTTablets. We now allow these to be overridden while creating a workflow or dynamically after the workflow is already in progress.

Reference Table Materialization

There is a new option in Materialize workflows to keep a synced copy of reference or lookup tables (countries, states, zip codes, etc) from an unsharded keyspace, which holds the source of truth for the reference table, to all shards in a sharded keyspace.

New VEXPLAIN Modes: TRACE and KEYS

VEXPLAIN TRACE

The new TRACE mode for VEXPLAIN provides a detailed execution trace of queries, showing how they're processed through various operators and interactions with tablets. This mode is particularly useful for:

• Identifying performance bottlenecks
• Understanding query execution patterns
• Optimizing complex queries
• Debugging unexpected query behavior

TRACE mode runs the query and logs all interactions, returning a JSON representation of the query execution plan with additional statistics like number of calls, average rows processed, and number of shards queried.

VEXPLAIN KEYS

The KEYS mode for VEXPLAIN offers a concise summary of query structure, highlighting columns used in joins, filters, and grouping operations. This information is crucial for:

• Identifying potential sharding key candidates
• Optimizing query performance
• Analyzing query patterns to inform database design decisions

KEYS mode analyzes the query structure without executing it, providing JSON output that includes grouping columns, join columns, filter columns (potential candidates for indexes, primary keys, or sharding keys), and the statement type.

These new VEXPLAIN modes enhance Vitess's query analysis capabilities, allowing for more informed decisions about sharding strategies and query optimization.

Errant GTID Detection on VTTablets

VTTablets now run an errant GTID detection logic before they join the replication stream. So, if a replica has an errant GTID, it will not start replicating from the primary. This protects us from running into situations which are very difficult to recover from.

For users running with the vitess-operator on Kubernetes, this change means that replica tablets with errant GTIDs will have broken replication and will report as unready. Users will need to manually replace and clean up these errant replica tablets.

Automatically Replace MySQL auto_increment Clauses with Vitess Sequences

In https://github.com/vitessio/vitess/pull/16860 we added support for replacing MySQL auto_increment clauses with Vitess Sequences, performing all of the setup and initialization work automatically during the MoveTables workflow. As part of that work we have deprecated the --remove-sharded-auto-increment boolean flag and you should begin using the new --sharded-auto-increment-handling flag instead. Please see the new MoveTables Auto Increment Handling documentation for additional details.

Experimental MySQL 8.4 support

We have added experimental support for MySQL 8.4. It passes the Vitess test suite, but it is otherwise not yet tested. We are looking for feedback from the community to improve this to move support out of the experimental phase in a future release.

Current Errant GTIDs Count Metric

A new metric called CurrentErrantGTIDCount has been added to the VTOrc component. This metric shows the current count of the errant GTIDs in the tablets.

vtctldclient ChangeTabletTags command

The vtctldclient command ChangeTabletTags was added to allow the tags of a tablet to be changed dynamically.

The entire changelog for this release can be found here.

The release includes 338 merged Pull Requests.

Thanks to all our contributors: @GrahamCampbell, @GuptaManan100, @Utkar5hM, @anshikavashistha, @app/dependabot, @app/vitess-bot, @arthurschreiber, @beingnoble03, @brendar, @cameronmccord2, @chrism1001, @cuishuang, @dbussink, @deepthi, @demmer, @frouioui, @harshit-gangal, @harshitasao, @icyflame, @kirtanchandak, @mattlord, @mattrobenolt, @maxenglander, @mcrauwel, @notfelineit, @perminov, @rafer, @rohit-nayak-ps, @runewake2, @rvrangel, @shanth96, @shlomi-noach, @systay, @timvaillancourt, @vitess-bot