...But, I'm not even sure if it is a problem with postgres as
"Unexpected EOF on client connection." indicates a problem with the
client (i.e. pnfs).
You could consider trying to load the atlas DB into a fresh install of
pnfs/postgres on a spare node.
Greig
On 06/12/07 00:17, Matt Doidge wrote:
> Hello,
>
> We're just recovering from a postgres disaster here at Lancaster and
> are having trouble getting pnfs back up . Long story short, we had a
> database wraparound catastrophe (despite autovacuum being on)- I made
> the mistake of restarting pnfs rather then noticing what the error was
> straight away and that caused some nasty write errors that left things
> unhappy with the postgres database. To make matters worse I found out
> our backups have been broken for a few months (the end of August to be
> exact), so a restore from dump is the absolute last resort.
>
> I'm managed to soothe postgres to actually work now, and most of the
> databases are there (ops had to be dropped but they weren't keeping
> anything interesting), but pnfs is still not starting properly, with a
> few vo directories and the admin and data1 directories not starting.
>
> In the pnfs logs I see errors like:
> Can't open Database (read/write) <dbname> (/opt/pnfsdb/databases/data1):
> <dbname> can't determine DB ID: -1012 (0)
>
> In the postgres logs I see for each failed pnfs db:
> Unexpected EOF on client connection.
>
> Any help appreciated, I'm stuck now. I've made mutliple backups of the
> current postgres database so I'm free to try things. I'm also pretty
> sure I've been unable to avoid some data loss, but would like to try
> to keep it to a minimum- however maybe it's already too late for that?
>
> Any help or advice appreciated, I'm a little desperate now as I'm
> completely stuck.
>
> cheers,
> Matt
|