-
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
pkg/sql/sqlitelogictest/tests/local/local_test: TestSqlLiteLogic_testindexorderby_nosort1000slt_good_0_test failed #89635
Comments
I ran the test via this command, but it timed out after 1 hour, so need to increase the timeout:
|
I found a memory budget exceeded error in the logs:
I believe the test could be run with this command:
I tried running it once, but with the wrong config, and it timed out after an hour. |
Do we need to have the force production sizes logictest mode beef up memory for these tests? |
pkg/sql/sqlitelogictest/tests/local/local_test.TestSqlLiteLogic_testindexorderby_nosort1000slt_good_0_test failed with artifacts on release-22.2.0 @ 79756fd7476c970f763661a63824beaedc339bd3:
|
Finally got a failure on 3rd try running this:
|
So the big question here is why is this failure only happening on 22.2.0 branch? On master and release-22.2 this test passes even when workmem is 18kb so there must be something in the streamer that wasn't backported that fixes this? |
Okay false alarm, this fails on 22.2 and master as well I was just tricked by how flakey it is. |
Below ~70k these tests would flake like so: ``` pq: streamer budget: memory budget exceeded: 38112 bytes requested, 57344 currently allocated, 76801 bytes in budget ``` Fixes: cockroachdb#89635 Release note: None
Something is still bothering me here. Shouldn't the streamer be able to make progress while keeping memory usage below whatever budget is set? (With special behavior for individual rows that exceed the budget) I feel like there is an underlying streamer bug here of some kind. |
There is something weird going on because even if I force the workmem limit to something low, it doesn't fail every time, and crdb_test is off. Not sure how to debug, I'll keep banging on it. |
Do we think this is a release blocker? |
I looked at it enough to convince myself that the problem is the row based join reader thats throwing hundreds of spans at the streamer. From Enqueue doc comments:
I think its working by design and we should just bump up the limit but also feel like Yahor would know better. |
89851: roachtest: avoid using sudo to install python packages r=e-mbrown a=rafiss see also #88675 sudo can make it confusing which veresion of a package is being installed and used. This fixes the usage of virtualenvs, and avoid sudo. This also removes per-version blocklists for python tools' tests. Release note: None 90095: sql: bump workmem size to avoid sqllite flakes on index/1000 tests r=cucaroach a=cucaroach Below ~70k these tests would flake like so: ``` pq: streamer budget: memory budget exceeded: 38112 bytes requested, 57344 currently allocated, 76801 bytes in budget ``` Fixes: #89635 Release note: None Co-authored-by: Rafi Shamim <rafi@cockroachlabs.com> Co-authored-by: Tommy Reilly <treilly@cockroachlabs.com>
Below ~70k these tests would flake like so: ``` pq: streamer budget: memory budget exceeded: 38112 bytes requested, 57344 currently allocated, 76801 bytes in budget ``` Fixes: #89635 Release note: None
Ok, sounds good. Thanks for looking at it. |
Below ~70k these tests would flake like so: ``` pq: streamer budget: memory budget exceeded: 38112 bytes requested, 57344 currently allocated, 76801 bytes in budget ``` Fixes: cockroachdb#89635 Release note: None
pkg/sql/sqlitelogictest/tests/local/local_test.TestSqlLiteLogic_testindexorderby_nosort1000slt_good_0_test failed with artifacts on release-22.2.0 @ 776e9f889dec6fec9b4011e82c523c648b72254e:
Help
See also: How To Investigate a Go Test Failure (internal)
This test on roachdash | Improve this report!
Jira issue: CRDB-20345
The text was updated successfully, but these errors were encountered: