You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Expected behavior
I would have expected CI runs on both examples above to mostly fetch from cache.
Environment
OS name + version: See respective CI pipeline
Bazel version: 2.1.0
Version of the rules: See respective PR
Additional context
I have seen cache misses like these on rules_haskell CI for a while now, while not experiencing such cache misses on projects using rules_haskell.
Something that stands out on the buildkite CI runs is that the log is filled with error messages of the form
WARNING: Writing to Remote Cache:
403 Forbidden
<?xml version='1.0' encoding='UTF-8'?><Error><Code>RetentionPolicyNotMet</Code><Message>Object overwrite or deletion is not allowed due to retention policy</Message><Details>Object 'tweag-bazel-cache-buildkite/cas/a4f5b8cffb3889f90cb7fe44ed954eb9e62772a44c2430cafa28d3cd853df3b4' is subject to bucket's retention policy and cannot be deleted, overwritten or archived until 2020-06-20T19:44:29.358643482-07:00</Details></Error>
The text was updated successfully, but these errors were encountered:
Is this still relevant? AIUI, CI has moved to GitHub Actions approx. 1 year ago; there's been no activity on BuildKite or CircleCI in the meantime. Have cache misses been observed in Bazel's upstream BuildKite? The scheduled builds complete very quickly, so presumably not...
Agreed, this seems no longer relevant. I believe rules_haskell's own CI encounters cache misses with BuildBuddy, but, to my understanding (and corroborated by the discussion here) this is likely caused by cached artifacts being GC'ed on BuildBuddy's side.
Describe the bug
Builds and tests on rules_haskell CI seem to be mostly missing the cache even when a near 100% cache hit rate would be expected.
To Reproduce
Not a reproduction as such, but two examples:
CI https://buildkite.com/tweag-1/rules-haskell/builds/1465 and https://app.circleci.com/pipelines/github/tweag/rules_haskell/219/workflows/ae71ee59-ed0d-4d41-83c0-d44f587770c9/jobs/9301/steps
CI https://buildkite.com/tweag-1/rules-haskell/builds/1467 and https://app.circleci.com/pipelines/github/tweag/rules_haskell/221/workflows/b65184ae-d9ec-407d-81ff-79d01cee4306/jobs/9305/steps
Both these PR's changes should not invalidate most cached items. Yet, when mergify rebased on latest master before merge their CI runs built most targets from scratch instead of fetching them from cache.
Expected behavior
I would have expected CI runs on both examples above to mostly fetch from cache.
Environment
Additional context
I have seen cache misses like these on rules_haskell CI for a while now, while not experiencing such cache misses on projects using rules_haskell.
Something that stands out on the buildkite CI runs is that the log is filled with error messages of the form
The text was updated successfully, but these errors were encountered: