On Tue, Jul 27, 2021 at 11:18 PM Michael Paquier <michael@paquier.xyz> wrote:
> I have been looking at that. Here are some findings:
> - In pg_archivecleanup, CleanupPriorWALFiles() does not bother
> exit()'ing with an error code. Shouldn't we worry about reporting
> that correctly? It seems strange to not inform users about paths that
> would be broken, as that could bloat the archives without one knowing
> about it.
> - In pg_basebackup.c, ReceiveTarAndUnpackCopyChunk() would not
> exit() when failing to change permissions for non-WIN32.
> - pg_recvlogical is missing a failure handling for fstat(), as of
> 5c0de38.
> - pg_verifybackup is in the wrong, as mentioned upthread.
>
> Thoughts? All that does not seem to enter into the category of things
> worth back-patching, except for pg_verifybackup.
I think all of these are reasonable fixes. In the case of
pg_basebackup, a chmod() failure doesn't necessarily oblige us to give
up and exit; we could presumably still write the data. But it may be
best to give up and exit. The other cases appear to be clear bugs.
--
Robert Haas
EDB: http://www.enterprisedb.com