Re: Why are we PageInit'ing buffers in RelationAddExtraBlocks()?

Hi,

On 2019-01-28 16:40:36 -0800, Andres Freund wrote:
> On 2019-01-28 15:49:33 -0800, Andres Freund wrote:
> > On 2019-01-28 18:08:59 -0500, Tom Lane wrote:
> > > Andres Freund <andres(at)anarazel(dot)de> writes:
> > > > I'm inclined to put back the
> > > > LockBuffer(buf, BUFFER_LOCK_UNLOCK);
> > > > LockRelationForExtension(onerel, ExclusiveLock);
> > > > UnlockRelationForExtension(onerel, ExclusiveLock);
> > > > LockBufferForCleanup(buf);
> > > > if (PageIsNew(page))
> > > > dance regardless, just to get the buildfarm to green?
> > >
> > > The buildfarm's got half a dozen reports now of a failure of this ilk,
> > > so you'd better do something.
> >
> > Yea, I was working on a patch. Was trying to come up with an explanation
> > of how this can be realistically hit on the BF, but failed. I've pushed
> > something now, let's see whether that fixes it.
>
> It has not. Given that I don't understand what's happening here I'm
> going to revert both commits unless I figure it out in the next ~30min.

I did that now. I couldn't reproduce it locally, despite a lot of
runs. Looking at the buildfarm it looks like the failures were,
excluding handfish which failed without recognizable symptoms before and
after, on BSD derived platforms (netbsd, freebsd, OX), which certainly
is interesting. I asked Thomas Munro whether he could run on freebsd,
and he gave me a login, where I just now reproduced the issue (2 of 5
make check runs failing).

Greetings,

Andres Freund