-
Notifications
You must be signed in to change notification settings - Fork 3.8k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
sql/stats: TestDefaultColumns failed #81513
Comments
sql/stats.TestDefaultColumns failed with artifacts on master @ 7dd1c51f6b5802e32bafd82e46747f349836592f:
Help
See also: How To Investigate a Go Test Failure (internal)
|
It looks like TestDefaultColumns is flaking, but I haven't been able to reproduce it with the random seeds above or through stress testing 150 runs. |
sql/stats.TestDefaultColumns failed with artifacts on master @ 48e48db89eb3cd6bc38f3631364c516181811280:
Help
See also: How To Investigate a Go Test Failure (internal)
|
I wonder if that check should be wrapped in a retry thingee. |
This commit adds some logging to the TestDefaultColumns test in order to help debug a test failure. Informs cockroachdb#81513 Release note: None
81737: cli/demo: fix enterprise features for multitenancy r=rafiss a=knz First commit from #81762. Fixes #80270. Fixes #40434. This patch achieves two things: - it ensures that enterprise features work properly in multi-tenant `cockroach demo` when `--disable-demo-license` is not passed as argument. - it ensures `cockroach demo` does not require a working licensing endpoint to run (by not retrieving a license over the network on startup). The net result will be less spurious failures in CI, as well as avoiding startup delays/errors during interactive uses, which provide poor UX. Release note: None 81797: sql: assert that table writers use root txns r=yuzefovich a=yuzefovich We assume that all mutation statements use the Root txn, but this was never enforced in practice. This commit adds an explicit check when initializing the table writer that the root txn is provided. Release note: None 81800: stats: add logging to TestDefaultColumns to help debug flake r=rytaft a=rytaft This commit adds some logging to the TestDefaultColumns test in order to help debug a test failure. Informs #81513 Release note: None Co-authored-by: Raphael 'kena' Poss <[email protected]> Co-authored-by: Yahor Yuzefovich <[email protected]> Co-authored-by: Rebecca Taft <[email protected]>
sql/stats.TestDefaultColumns failed with artifacts on master @ 1e2cc61b58dc14386bb68dca59814874648931c2:
Help
See also: How To Investigate a Go Test Failure (internal)
|
Yes, auto stats are definitely enabled here. I'm not sure what exactly is causing the race condition, but I'll update the test to use the new table-level auto stats settings. |
81782: sql: use declarative schemachange for add column sequence exprs r=fqazi a=fqazi Fixes: #81781 Previously, the declarative schema changer was disabled for add column expressions with sequence references (i.e. default, on update, computed) because we were missing telemetry from the legacy schema changer for during backfill related failures. This was inadequate because the lack of telemetry from these areas could be seen as a usability issue. To address this, this patch adds in support for missing telemetry and enables supports for add column operations with sequence operations. Release note: None 82388: sql: remove date/intervalstyle_enabled from code r=otan,mgartner a=rafiss fixes #81529 No release note is needed, since v22.1 already included a release note about how these are both hardcoded to true and cannot be changed. The session variables and cluster settings remain in the code, but are marked as retired. Release note: None 82451: kvserver/rangefeed: fix off-by-one in `NewCatchUpIterator()` r=miretskiy a=erikgrinaker **kvserver/rangefeed: fix off-by-one in NewCatchUpIterator()** `NewCatchUpIterator` created the `MVCCIncrementalIterator` with `RangeFeedRequest.Header.Timestamp.Prev()`, because it assumed the given timestamp was inclusive. However, the rest of the rangefeed code treats this timestamp as exclusive. This patch uses the correct, exclusive timestamp when creating the `MVCCIncrementalIterator`. This change has no externally visible effect: even though the previous code would cause `MVCCIncrementalIterator` to emit keys at `Timestamp`, the `CatchUpScan()` method would discard those events if they were at or below `Timestamp`. An integration test is also added to verify that the start timestamp of a rangefeed catchup scan is exclusive. The test passes both with the new and the old code, as expected. Release note: None **rangefeed: emphasize that start time is exclusive** This patch adds comments for rangefeed-related APIs and components emphasizing that the start timestamp of rangefeeds is exclusive. In other words, the first possible emitted event (including catchup scans) will be at `RangeFeedRequest.Header.Timestamp.Next()`. Several parameters are also renamed to emphasize this. There are no functional or semantic changes. Touches #82488. Release note: None 82466: dev: have `doctor` advise to set a particular `tmpdir` r=rail a=rickystewart Bazel's default behavior of rooting the `tmpdir` to an "in-sandbox" directory has been a point of confusion for CRL developers. The sandbox directory does not exist after the test is run (unless `--sandbox_debug` is provided), which is sometimes confusing for folks who expect their test's temp files to be present where the logs suggest they should be (see #82413). Furthermore, the long `tmpdir` used in these cases breaks tests that create Unix sockets on OS's where Unix sockets have a maximum path length. Avoid these problems by having `doctor` just tell you to manually set a `test_tmpdir`. We add `/tmp` to `gitignore` in case people want to root it at the `tmp` directory in their checkout. Closes #72918. Closes #82413. Release note: None 82476: update cluster-ui to v22.2.0-prerelease-1 r=maryliag a=maryliag Update cluster-ui to latest published version Release note: None 82519: stats: fix flaky test TestDefaultColumns r=rytaft a=rytaft TestDefaultColumns creates statistics on a table with 110 columns using the command `CREATE STATISTICS s FROM t.a`. It then checks that there are exactly 101 column statistics on table t.a with statistics_name = 's' (one stat for the primary index, plus 100 other column stats). However, this test may be flaky if automatic statistics are running, since each new automatic stat will cause other stats to be deleted. Although the test disables automatic stats at the beginning, it seems that some sort of race condition may cause it to be reenabled. This commit fixes the problem by disabling automatic stats using the new table level settings, ensuring that the 101 column stats are not deleted after they have been created. Fixes #81513 Release note: None Co-authored-by: Faizan Qazi <[email protected]> Co-authored-by: Rafi Shamim <[email protected]> Co-authored-by: Erik Grinaker <[email protected]> Co-authored-by: Ricky Stewart <[email protected]> Co-authored-by: Marylia Gutierrez <[email protected]> Co-authored-by: Rebecca Taft <[email protected]>
sql/stats.TestDefaultColumns failed with artifacts on master @ ce66acdbba801de88f1dd645eaedeb3834f23dbd:
Help
See also: How To Investigate a Go Test Failure (internal)
Parameters in this failure:
This test on roachdash | Improve this report!
Jira issue: CRDB-15109
The text was updated successfully, but these errors were encountered: