> On Jun 28, 2018, at 9:47 AM, Vijaykumar Jain <vjain(at)opentable(dot)com> wrote:
>
> Thanks Scott.
>
> fsync is on.
>
> postgres=# show fsync;
> fsync
> -------
> on
> (1 row)
>
Yeah, I agree with Scott; it sounds like fsync() is being shorted somewhere in the I/O path. Are you using a RAID controller; if so is the cache setting — write through or write back? If it’s write back then writes are being cached and a power outage will lead to data corruption — the only saving grace at that point would be the battery backup logic on the controller which may or nay not be trustworthy.