Thread: Is that VACUUM normal ?

Is that VACUUM normal ?

From
Jean-Christophe Boggio
Date:
Hi all,

3 hours ago I started :

myplace# vacuum verbose analyze gains;
NOTICE:  --Relation gains--
NOTICE:  Pages 44547: Changed 406, reaped 1941, Empty 0, New 0; Tup 4099959: Vac
 61464, Keep/VTL 0/0, Crash 2, UnUsed 1281, MinLen 67, MaxLen 267; Re-using: Fre
e/Avail. Space 6900268/6899660; EndEmpty/Avail. Pages 0/1917. CPU 8.71s/23.58u s
ec.
NOTICE:  Index ix_gains_idmembreidcondition: Pages 18532; Tuples 4099959: Delete
d 10846. CPU 3.09s/4.39u sec.
NOTICE:  Index actegen_gains_key: Pages 10291; Tuples 4099959: Deleted 10845. CP
U 1.98s/4.11u sec.
NOTICE:  Index idcondition_gains_key: Pages 8420; Tuples 4099959: Deleted 10845.
 CPU 1.93s/4.14u sec.
NOTICE:  Index dategain_gains_key: Pages 15090; Tuples 4099959: Deleted 10845. C
PU 3.46s/4.18u sec.
NOTICE:  Index gains_idgain_key: Pages 10124; Tuples 4099959: Deleted 10845. CPU
 2.33s/4.11u sec.

and now it's stuck here for more than 2 hours now, top says :

  PID USER     PRI  NI  SIZE  RSS SHARE STAT %CPU %MEM   TIME COMMAND
 9790 postgres  16   0 19628  19M 18536 R    99,2  2,0 212:15 postmaster


I use PG 7.03 on a 2x PIII-800 with 1Gb RAM with UWSCSI disks.

Table is about 4M tuples (as you can see), data file is :
-rw-------    1 postgres postgres 364929024 fév 20 08:15 /usr/local/pgsql/data/base/mybase/gains

It shouldn't be THAT long ! I think a few days ago it was 10mn long

Our site is down until I either :
- stop the vacuum but then with no analysis information what will be
the performances ?
- wait until it finishes but what year will that be ?

Anyone has an idea of what's going on ?

--
Jean-Christophe Boggio
cat@thefreecat.org
Independant Consultant and Developer
Delphi, Linux, Perl, PostgreSQL



Re: Is that VACUUM normal ?

From
Tom Lane
Date:
Jean-Christophe Boggio <cat@thefreecat.org> writes:
> 3 hours ago I started :
> myplace# vacuum verbose analyze gains;

> It shouldn't be THAT long ! I think a few days ago it was 10mn long

Seems odd.  What file is it currently modifying?  (Look at the
timestamps in the database directory)  Can you attach to the running
backend with gdb and get a stack trace?

            regards, tom lane