Actions
Bug #56646
closedpacific: ECANCELLED errors within RGWRados::follow_olh should not bubble up to the caller
% Done:
0%
Source:
Tags:
Backport:
Regression:
No
Severity:
3 - minor
Reviewed:
Description
It's fine to ignore races because we only care that the BI OLH journal is eventually applied, not that it
is applied by the current thread. Even if we were to retry in these cases, there could be another OLH
change between the time that this function returns and the time that the caller uses the result. The caller
must, therefore, always assume that the resolved OLH target could be stale.
Allowing these to bubble up to the caller can unnecessarily cause "Unknown" errors to be returned by the S3 interface
when there are overlapping modification requests for the same OLH object. This is a discrepancy with AWS S3, and
we've seen it cause issues with some clients.
Updated by Cory Snyder almost 2 years ago
- Pull request ID changed from 47187 to 47266
Updated by Casey Bodley over 1 year ago
- Subject changed from ECANCELLED errors within RGWRados::follow_olh should not bubble up to the caller to pacific: ECANCELLED errors within RGWRados::follow_olh should not bubble up to the caller
- Backport deleted (
quincy, pacific)
pull request targets pacific, so no backports are necessary
Updated by Yuri Weinstein over 1 year ago
Actions