On Apr 2, 2012, at 3:04 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Seems like basically what you've proven is that this code path *is* a
> performance issue, and that we need to think a bit harder about how to
> avoid doing the fsync while holding locks.
Hmm, good idea. I wonder if we couldn't just hand off the fsync request to the background writer, as we do with buffer
fsyncrequests. AFAICS we don't need the fsync to happen right away; the next checkpoint cycle should be soon enough.
...Robert