On Sat, Mar 9, 2019 at 12:33 AM Julien Rouhaud <rjuju123@gmail.com> wrote:
On Sat, Mar 9, 2019 at 12:35 AM Magnus Hagander <magnus@hagander.net> wrote: > > On Mon, Mar 4, 2019 at 11:31 AM Julien Rouhaud <rjuju123@gmail.com> wrote: >> >> On Fri, Feb 22, 2019 at 3:01 PM Magnus Hagander <magnus@hagander.net> wrote: >> > >> > It tracks things that happen in the general backends. Possibly we should also consider counting the errors actually found when running base backups? OTOH, that part of the code doesn't really track things like databases (as it operates just on the raw data directory underneath), so that implementation would definitely not be as clean... >> >> Sorry I just realized that I totally forgot this part of the thread. >> >> While it's true that we operate on raw directory, I see that sendDir() >> already setup a isDbDir var, and if this is true lastDir should >> contain the oid of the underlying database. Wouldn't it be enough to >> call sendFile() using this, something like (untested): >> >> if (!sizeonly) >> - sent = sendFile(pathbuf, pathbuf + basepathlen + 1, &statbuf, true); >> + sent = sendFile(pathbuf, pathbuf + basepathlen + 1, &statbuf, true, >> isDbDir ? pg_atoi(lastDir+1, 4) : InvalidOid); >> >> and accordingly report any checksum error from sendFile()? > > > That seems it was easy enough. PFA an updated patch that does this, and also rebased so it doesn't conflict on oid. > > (yes, while moving this from draft to publish after lunch, I realized that you put a patch togerher for about the same. So let's merge it)
Thanks! Our implementations are quite similar, so I'm fine with most of the changes :) I'm just not sure about having two distinct functions for reporting failures, given that there's only one caller for each. On the other hand it avoids to include miscadmin.h in bufpage.c.
Yeah, and it brings "cosistence" to at least the calling point(s) within regular backends.
That's just a detail, so I'm marking it (again) as ready for committer!