At Wed, 25 Sep 2019 12:24:03 +0900, Michael Paquier <michael@paquier.xyz> wrote in <20190925032403.GF1815@paquier.xyz>
> On Tue, Sep 24, 2019 at 02:48:16PM +0900, Kyotaro Horiguchi wrote:
> > Of course I found no *explicit* ones. But I found one
> > ereport(DEBUG1 in register_dirty_segment. So it will work at
> > least for the case where fsync request queue is full.
>
> Exactly. I have not checked the patch in details, but I think that
> we should not rely on the assumption that no code paths in this area do
> not check after CHECK_FOR_INTERRUPTS() as smgrtruncate() does much
> more than just the physical segment truncation.
Agreed to the point. Just I doubted that it really fixes the
author's problem. And confirmed that it can be.
regards.
--
Kyotaro Horiguchi
NTT Open Source Software Center