145435: sql: add sqlcommenter support r=kyle-a-wong a=kyle-a-wong commit 1/2: sql: refactor parser.go to take parser options Refactored parser functions to use a new ParserOptions type configure various options when parsing sql. There are now two types of Parse functions, Parse*() and Parse*WithOptions(). Parse* will use default parse options and the WithOptions alternative allows specfic options to be provided. Epic: None Release note: None ---- commit 2/2: sql: retain SQL commenter tags SQL commenter library is used by ORM libraries and customer applications to add application context to the queries using inline SQL comments. This PR includes the following changes: - A new cluster setting, `sql.sqlcommenter.enabled`, which retains comments during parsing phase. - If the comments follow [SQL commenter specification](https://google.github.io/sqlcommenter/spec/#introduction), they are converted to SQL tags. - Changes to insights stats to also include SQL commenter tags. - Changes to the crdb_internal.*_execution_insights virtual tables to have a new `query_tags` column. - Addition of log tags on the conn_ex context. Each sql commenter tag will be added as a log tag, prefixed with 'querytag-'. Epic: None Release note (sql change): Added support for query tagging, which allows users to add query tags to their sql statements via comments. These query tags will be logged as log tags and will be included in all log entries logged during the executing of a sql statement. These tags will also be included in traces. Finally, these tags will be included in the `cluster_execution_insights` and `node_exeuction_insights` virtual tables in a new `query_tags` JSONB column. This feature can be enabled with the `sql.sqlcommenter.enabled` cluster setting and is disabled by default. Comments must adhere to the sqlcommenter spec, as defined here: https://google.github.io/sqlcommenter/spec/ 146120: kvcoord: fix the write buffer's adjustError r=yuzefovich,stevendanna a=miraradeva Previously, `adjustError` handled the cases of stripped requests (not sent to KV at all) and non-transformed requests (sent directly to KV), but didn't always handle the cases of transformed requests correctly. This commit simplifies the logic in `adjustError` and adds additional testing. Fixes: [#144694](#144694) Release note: None 146285: roachtest: init providers in main(), not init() r=tbg a=tbg Initializing them in an init function means we're calling this method in tests, where it is likely not useful and may even be implicated in opaque test timeouts such as [1]. [1]: https://cockroachlabs.slack.com/archives/CJ0H8Q97C/p1746629472748029?thread_ts=1746626113.617579&cid=CJ0H8Q97C The roachtest job still works[^1]: <img width="1352" alt="image" src="https://github.com/user-attachments/assets/9233459b-5fd6-4e91-a8a5-deba1177eaab" /> [^1]: https://teamcity.cockroachdb.com/buildConfiguration/Cockroach_Nightlies_RoachtestNightlyGceBazel/19646638?buildTab=overview&hideProblemsFromDependencies=false&hideTestsFromDependencies=false&expandBuildChangesSection=true Epic: none 146298: revert "sql: remove old hyperloglog version" r=yuzefovich a=yuzefovich This reverts commits 088306f and 70e9742. We haven't yet advanced necessary versions on master to make an incompatible change with 24.3 without any fallout, so let's revert the change for now (a couple of weeks). Fixes: #146268. 146318: roachtest: update backup fixtures collection URIs to millisecond precision r=msbutler a=kev-cao The current backup fixture driver creates backups to collection URIs that are formatted based on the current time, accurate to the minute. This causes test flakes when two roachtest fixture runs run at the same minute and create backup schedules that backup to the same collection URI. To avoid this, we update the fixtures to be accurate to the millisecond. Fixes: #146031, #146030 Release note: None 146328: sql: fix crdb_internal.fingerprint in edge cases r=yuzefovich a=yuzefovich This commit fixes a couple of bugs that were introduced in 5e42a93 in `crdb_internal.fingerprint` builtin. Namely, after that change it was possible that the new coordinator goroutine never exits (because we forgot to check for the context cancellation) which could block the server shutdown. It could also happen if we had a `statement_timeout` set. Relatedly, if the coordinator encounters an error itself (rather than receives it from one of the workers), the coordinator would exit without notifying the workers. This was the case since we only close the work channel after having pushed out all sub-spans. I don't think this was problematic in practice, but this is now fixed by explicitly canceling the context of workers. Fixes: #146320. Release note: None Co-authored-by: Chandra Thumuluru <chandra.thumuluru@cockroachlabs.com> Co-authored-by: Mira Radeva <mira@cockroachlabs.com> Co-authored-by: Tobias Grieger <tobias.b.grieger@gmail.com> Co-authored-by: Yahor Yuzefovich <yahor@cockroachlabs.com> Co-authored-by: Kevin Cao <kevin.cao@cockroachlabs.com>

nightly: Random Syntax Tests often times out #146320

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions