Open abel-von opened 3 months ago
after this change poststop hooks are now required to be idempotent, even existing ones, as they can be called multiple times. Won't existing hooks fail in most cases when called multiple times. Possibly they can misbehave and remove the wrong resources.
I wonder if this behavior should be configurable to not affect existing installations. We could have a flag for each hook to say whether a failure should be ignored or reported immediately (with the default being ignore
as it is currently documented)
If a poststop hook run failed(maybe because the resource is still not ready to be recycled), we expect that the delete operation could fail and be retried util all poststop hooks run succeed.
In the current runtime spec, when a
poststop
hook failed, runtime can not get the error, so that docker or kubernetes also can not get the error, this is easy to leave the resource residual.I think this is similar to the situation of
poststart
hook. #1262Also please refer to the PR in runc: https://github.com/opencontainers/runc/pull/4364