Re: BUG #15636: PostgreSQL 11.1 pg_basebackup backup to a CIFS destination throws fsync error at end of backup

On Sat, Feb 23, 2019 at 11:28 AM Andres Freund <andres(at)anarazel(dot)de> wrote:
> On 2019-02-23 11:20:02 +1300, Thomas Munro wrote:
> > So, for this patch tolerating EINVAL for fsync of directories (but not
> > files), enabling eg pg_backbackup on CIFS on eg Linux 3.10 to work,
> > tested by John, we have:
> >
> > Stephen: +1
> > Michael: -0.1
> > Thomas: -0.1
> >
> > Any other opinions?
> >
> > Just in case you think it's strange that I'm voting against my own
> > patch: I'd probably be for it if I hadn't discovered that they've
> > fixed this in Linux 3.16 and later so that it succeeds. It's
> > apparently not in the default RHEL 6 and 7 kernels, though, and the
> > latter could be around for a while. I'm not entirely sure what amount
> > of work we should be doing to tolerate problems that are fixed in a
> > newer versions. One argument is that a 3.10 user who cares about this
> > should petition RH to back-port the fix into that kernel. (The nearby
> > WSL thread has some things in common but is a more clear cut case IMV
> > because there is no fix available on the WSL side.)
>
> Hm, given the fact that EINVAL doesn't appear to be triggerable by data
> level issues, I don't see much reason not to allow it. I mean it's
> really stupid that cifs ever returns it, but there's going to be a lot
> of users running things on older kernel for the forseeable future.

Thanks. Pushed.

FTR here is the discussion about the pre-existing tolerance of EBADF
for directories:

https://www.postgresql.org/message-id/flat/20100215005057.5AC8F7541C5%40cvs.postgresql.org

--
Thomas Munro
https://enterprisedb.com