[clug] Alarming SuSE/Postgres problem

[Alex Satrapa]

On 17 Jun 2005, at 11:07, Antti.Roppola at brs.gov.au wrote:

> A bit of reading suggests that disk caching settings in Posgres and/or
> the filsystem (ext3 on LVM on IDE) may be why (i.e defaults for  
> fsync).
> That or how WAL is set up. I'd really like to know what's going on  
> before
> trying to induce a repeat of the event. Ideas?

People love blaming caches... how long was your system running, and  
how big was the database that disappeared?

If the system was running for more than a few hours (heck, five  
seconds of idle time is usually enough), then it's not the WAL,  
filesystem or IDE caches that caused your problem.

I would start looking at things like an update moving the database  
directory from somewhere like /usr/local/mydatabase to the default  
of /var/lib/postgres/data. Perhaps your database is hiding somewhere?  
The PID file might be in the old location too. Check /etc/postgresql  
for any old config files. Search the file system for PG_VERSION files.

You may also have put the PostgreSQL databases on a volume that is no  
longer mounted, or mounted a different volume over the top of the  
PostgreSQL directories. Check what volumes you've got described in  
your /dev/lvm/ (?) directory, and see where they're mounted (versus  
where they're supposed to be mounted).

It would surprise me to find that PostgreSQL had been caching stuff  
in memory for a few weeks!