On Fri, Nov 20, 2015 at 7:03 AM, Robert Haas <robertmhaas@gmail.com> wrote:
> On Thu, Nov 19, 2015 at 7:19 AM, Thom Brown <thom@linux.com> wrote:
>> This bug still exists.
>
> Hmm. This probably should have been on the open items list.
I have added an open item for 9.5. That's not a cool bug to release
the next GA even if that's not limited to 9.5, it is easily
reproducible in older stable branches as well.
> I didn't
> pay too much attention this to this before because it seemed like
> Andres and Michael were all over it.
This completely fell off my radar. Sorry about that.
For back-branches, I tend to agree with what Horiguchi-san mentioned
upthread: we had better issue an unconditional fsync on the relation
when INIT_FORKNUM is used for it when replaying XLOG_FPI record. That
would rather localize the impact. An example of patch is attached that
applies on top of REL9_4_STABLE. That's rather ugly but it shows the
idea and fixes the issue.
For master and perhaps 9.5, I would think that a dedicated WAL record
type enforcing the fsync is the way to go. This special treatment
would go only for btree and spgist when they use INIT_FORKNUM and we
would not impact other relation types and code paths with this
behavior.
Thoughts?
--
Michael