Обсуждение: Stange IO error while working with large objects.

Hey all,

I've never experienced such problems before pefrorming some tests

on large objects. I am on Ubuntu and my HDD is whole encrypted

(LVM2). I've imported large object ~ 1.5 Gb of size. After this, entire

system lost performance dramaticaly and the disk activity becomes

anomalous.

After reboot everithing is fine with OS, but attempt to remove all

large object results in error:

dmitigr=# select lo_unlink(loid) from (select distinct loid from pg_largeobject) as foo;

ERROR:  could not read block 704833 in file "base/16386/11838.5": read only 4096 of 8192 bytes

What does this means -- hardware / OS / Postgres failure ?

Any suggestions ? 

--
// Dmitriy.

On Wed, Mar 30, 2011 at 7:37 AM, Dmitriy Igrishin <dmitigr@gmail.com> wrote:
> I've checked the disk with badblocs(8). The results are:
>
> File /pgsql/9.0/data0/base/16386/11838.5 (inode #3015588, mod time Wed Mar
> 30 13:13:13 2011)
>   has 50 multiply-claimed block(s), shared with 1 file(s):
>     <The bad blocks inode> (inode #1, mod time Wed Mar 30 15:23:19 2011)
>
> After this, I've dropped the database and create a new one. Problem is
> solved.
> All the same it is interesting, why there was such problem? I am disturbed
> because
> I intend to use large objects in production...

It's very unlikely the problem is postgres.  I give it equal odds on
hard disk issue (did you check smart?) or lvm bug.  I'd consider
stress testing this storage stack for a while.

merlin