Closed problame closed 2 months ago
Just a snarky reminder that we did have a discussion about it on the PR which introduced it, but I still agree, it would be fine as info!
.
The existing "ignoring" code path is rather indiscriminate -- probably need to make it more fine-grained and do things like explicitly ignoring cancellation, rather than just ignore all possible errors
There was a potential PageReconstructError buried behind this log issue, which we shouldn't be ignoring.
Log message
ignoring failure to find gc cutoffs: timeline shutting down
should be atinfo!
level IMO.Logged here:
https://github.com/neondatabase/neon/blob/7e60563910936cf6643edb686a8163b0b03c7108/pageserver/src/tenant.rs#L2932
Two tests have it in
allowed_errors
, so, as part of working on this ticket, one should clean upallowed_errors
.Found this while checking prodlike cloudbench logs for this week's release: https://neondb.slack.com/archives/C06K38EB05D/p1718117355604419?thread_ts=1718115200.692949&cid=C06K38EB05D
Please do not act on this before bug triage meeting has decided this should be info! level.