roachtest: ycsb/F/nodes=3/cpu=32 failed #71264
Labels
branch-master
Failures and bugs on the master branch.
C-test-failure
Broken test (automatically or manually discovered).
O-roachtest
O-robot
Originated from a bot.
release-blocker
Indicates a release-blocker. Use with branch-release-2x.x label to denote which branch is blocked.
Comments
cockroach-teamcity
added
branch-master
|
roachtest.ycsb/F/nodes=3 failed with artifacts on master @ cc6296c24ddb048215dabe5cc41339f306db4f41: |
|
roachtest.ycsb/F/nodes=3 failed with artifacts on master @ cc6296c24ddb048215dabe5cc41339f306db4f41: |
craig bot
pushed a commit
that referenced
this issue
Oct 7, 2021
71083: opt, parser: add support for nulls first, nulls last r=nehageorge a=nehageorge Previously, we only supported the default Postgres ordering for ORDER BY. That is, NULLS come first for ascending order and NULLS come last for descending order. Postgres also supports the keywords "NULLS FIRST" and "NULLS LAST", which explicitly places all the NULLS first or last, respectively. This PR adds support for this non-default ordering. Informs #6224. Release note (sql change): NULLS FIRST and NULLS LAST specifiers now supported for ORDER BY. 71219: sql: UserHasAdminRole should return true for the admin role itself r=rafiss a=tukeJonny This PR makes `planner.UserHasAdminRole` returns true on the `admin` role for improving introspection. This PR had checked by running `make testbaselogic` . Fixes #70779 Release note: None 71220: dev: update how we test for presence of `--dev` config in `doctor` r=rail a=rickystewart Some people are running Linux and can therefore use the `crosslinux` config on a day-to-day basis. This is OK and `dev doctor` should allow this. Instead of failing the `doctor` check if `--config dev` isn't set, we instead test whether the user has stamping enabled on a `bazel build`. The built `cockroach` binary won't start if stamping is disabled and both the `dev` and `crosslinux` configs enable stamping. Release note: None 71227: bazel: avoid `read-only file system` error when calling into `autom4te` r=rail a=rickystewart See 3f50217 for more context. Many people are encountering the error described in that commit and it's not feasible to ask everyone to install an `autom4te` shim. Instead we include the shim in the build definition. We also remove it from the builder image to demonstrate that the fix works. Closes #70986. Release note: None 71265: changefeedccl: don't warn when Timestamp() is called on flush event r=miretskiy a=stevendanna When investigating a recent test failure, I noticed the following log messages: ccl/changefeedccl/kvevent/event.go:153 ⋮ [-] 1830 setting empty timestamp for unknown event type ccl/changefeedccl/kvevent/event.go:153 ⋮ [-] 1831 setting empty timestamp for unknown event type ccl/changefeedccl/kvevent/event.go:153 ⋮ [-] 1832 setting empty timestamp for unknown event type ccl/changefeedccl/kvevent/event.go:153 ⋮ [-] 1833 setting empty timestamp for unknown event type ccl/changefeedccl/kvevent/event.go:153 ⋮ [-] 1834 setting empty timestamp for unknown event type One possibility is that these were generated by our new Flush event, which is an in-band signal between the kvFeed and the changeAggregator and has no associated timestamp. This PR adds handling for TypeFlush to cut down on the log spam. Not however, this raises the question of why we were hitting the condition that forces a flush so often. One possibility is that we were unable to allocate resources to add items to the queue because of some other user of our memory monitor and thus while we kept trying to flush, these flushes had no effect. We may consider tracking whether there has been a successful resource acquisition since the last flush and only resending a flush if there has been. Release note: None 71280: sql: clear (rather than emptying) bound account r=[Azhng,miretskiy] a=stevendanna Empty() leaves some number of bytes in the bound account as reserved. When we abandon a Container, those "reserved" bytes were never released back to the parent. I believe we currently abandon one Container at the end of each explicit transaction. Fixes #71262 Fixes #71263 Fixes #71264 Fixes #65279 We may want to provide two different methods. One that calls Empty() and one that calls Clear() in case there are users of this who really need to keep their reserved bytes. Release note: None Co-authored-by: Neha George <neha.george@cockroachlabs.com> Co-authored-by: tukeJonny <ne250143@yahoo.co.jp> Co-authored-by: Ricky Stewart <ricky@cockroachlabs.com> Co-authored-by: Steven Danna <danna@cockroachlabs.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
roachtest.ycsb/F/nodes=3/cpu=32 failed with artifacts on master @ cc6296c24ddb048215dabe5cc41339f306db4f41:
Reproduce
See: roachtest README
This test on roachdash | Improve this report!
The text was updated successfully, but these errors were encountered: