Merge queue observations: - A <https://github.com/...
# kotest-contributors
o
Merge queue observations: • A PR Test's job failed flakily with
java.lang.OutOfMemoryError: Java heap space
. • Re-running that job was successful, but did not make the merge queue add the PR back to the queue. • Merging had to be re-started manually with the entire PR Test running again in order to complete the merge. Could the GH infra do better?
a
It would help if Kotest had remote Gradle Build Cache. The actions cache is over the limit "21.79 GB of 10 GB Used" https://github.com/kotest/kotest/actions/caches so it's probably effectively not able to cache anything
o
Hmm, the info box says "Least recently used caches will be automatically evicted to limit the total cache storage to 10 GB". The cache contains entries more than 24 hours old, so with an LRU eviction policy this should not affect the above (two consecutive action runs). What I was thinking is: Here we have a job re-run without any change to the repo at all, so if that succeeds, why not let it complete the merge process?
Correction: "more than 24 hours old" should have read "last used more than 24 hours ago". Some entries have actually been created long ago "cached last month".
a
looking at the build scan for the failed build you linked, it shows that none of the tasks could be loaded from cache https://scans.gradle.com/s/lqnrbqbdc7zja/performance/build-cache#local-miss I'd expect some tasks couldn't be loaded because the inputs have changed (after all, it's a PR, so the code has changed at some point), but it's suspicious that nothing could be cached.
o
Yeah, for the failed task it's OK to see cache misses, but the re-run should not be affected.
a
I couldn't see any entries created a month ago, but maybe they've been removed by GitHub by now
o
They are still there on page 3: https://github.com/kotest/kotest/actions/caches
Search for "last month".
a
hmmm weird, I sorted by 'least recent' and it's only showing caches 2 days old
o
Least recent means "used", not "cached".
a
and oldest is 1 hour old
o
No, look at the fine print on the left, below the artifact: "120 KB cached last month"
a
ahh right, I see, I was looking at the column
👍 1
e
I'll see if I can get it to fail locally and analyze the heapdump a bit..
👍 1
Hard to replicate tbh.. Perhaps this is good enough, albeit blunt? https://github.com/kotest/kotest/pull/3948
👍 2
o
Seems entirely reasonable to me.