-
Notifications
You must be signed in to change notification settings - Fork 577
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
tests: use rpk producer for even record distribution #4928
tests: use rpk producer for even record distribution #4928
Conversation
kafka tools cli used in some of the tests could produce non uniform message distribution, causing errors in assertion because the segment is not uploaded to SI. using rpk producer produces better distribution as it uses random keys. this commit also fixes a log message and adds a topic specific pattern to the matching condition of topic_manifest.json. This is to avoid the verification matching files for topics which are not part of the test.
This looks good. I wonder if it's work converting all the cases in topic_recovery_test to use this producer? I think they all have a similar risk of partition imbalance, although I'm not sure why we saw more failures of test_missing_partition than the others. I did just see a failure of test_fast2 here (https://buildkite.com/redpanda/redpanda/builds/10542#0180faea-d582-4f72-8f1e-69f25b0eec77), which I haven't dug into but maybe has the same underlying cause? |
It is possible. Looking at the logs for that failure, partition index 2 of topic 1 never got enough records to upload to S3 and so the condition was not reached (1/0 is uploaded twice and 1/2 is never uploaded):
for this partition all logs show the candidate segment open like below:
although in this case the bytes difference does not seem very large in the partitions. Probably because there are six partitions to upload. |
494541c
to
ef38553
Compare
use rpk producer for all TopicRecoveryTest instances. Additionally some cleanups according to python linter.
ef38553
to
4329fa5
Compare
I used rpk producer for most tests but left size and time based retention tests alone for now, they do not seem to work well with that producer and more investigation is required into why they fail, will keep this PR focused on the tests where the record count is not uniform across partitions. |
/backport v22.1.x |
Failed to run cherry-pick command. I executed the below command:
|
/backport v21.1.x |
Branch name "v21.1.x" not found. |
/backport v21.11.x |
Failed to run cherry-pick command. I executed the below command:
|
this branch is really far out of sync with dev |
Cover letter
kafka tools cli used in some of the tests could produce non uniform message distribution, causing errors in assertion because the segment is not uploaded to SI. Using rpk producer produces better distribution as it uses random keys.
fixes #4886