Sorry for replying to myself, but I had to point out that:
>> I'm guessing that the new table/index files created during the CLUSTER
>> didn't get released after the PANIC. Look for files that are not
>> referenced by any relfilenode value in pg_class.
> Using the oid2name tool I get this:
> I suspect all the files with "No tables with that oid found" can be
> removed, but I don't know exactly how oid2name works, so I'll check them
> against pg_class before removing them.
This is ofcourse not a good idea. There is quite a difference between the
oid and the relfilenode. Comparing the output of:
SELECT DISTINCT relfilenode FROM pg_class ORDER BY relfilenode;
to the directory listing gave a much better result.
The files I found in this way also had an atime which was the same as the
time of the crash.
Just to prevent someone else from making the mistake I made in my previous
In response to
pgsql-admin by date
|Next:||From: Dick Davies||Date: 2005-01-16 09:10:30|
|Subject: Re: PAM ldap|
|Previous:||From: Sander Steffann||Date: 2005-01-15 22:21:20|
|Subject: Re: Less available diskspace after crashed CLUSTER |