release-20.1: kv: elide truncation decision assert #47346
Conversation
See the discussion in cockroachdb#47143. We previously asserted on the invariants we expected the truncation decision to uphold. This was unfortunately more fragile than anticipated, and it's safer to remove it in the v20.1.0 release as we expect the panic to fire under normal operating conditions. A more hardened panic condition is going bake in master in the interim (cockroachdb#47143). As we build confidence with that patch, we can consider re-adding the assert to v20.1.1 if needed. Release note (bug fix): Remove a panic that would fire under normal operating conditions. Would result in "invalid truncation decision" error messages.
|
This change is |
|
Feel free to reject this patch. My argument here is the asserts were introduced without any behavioral changes around log truncations this release. See #42495. Removing the panics, I don't think, causes any other faulty downstream behavior. |
There was a problem hiding this comment.
Reviewable status:
complete! 1 of 0 LGTMs obtained (waiting on @nvanbenschoten)
There was a problem hiding this comment.
s/them for the first patch release/them until the first patch release/
Reviewable status:
|
Excellent, will just have to remember to cherry-pick #47143 when cutting out v20.1.1. |
See the discussion in #47143. We previously asserted on the
invariants we expected the truncation decision to uphold. This was
unfortunately more fragile than anticipated, and it's safer to remove it
in the v20.1.0 release as we expect the panic to fire under normal
operating conditions. A more hardened panic condition is going bake in
master in the interim (#47143). As we build confidence with that patch,
we can consider re-adding the assert to v20.1.1 if needed.
Release note (bug fix): Remove a panic that would fire under
normal operating conditions. Would result in "invalid truncation
decision" error messages.