Re: Large shared_buffers freezing computers

I read an earlier post by Tom where he recommends 1/4 of physical ram.
I will go to 1/5 to be safe and I assume it will be ok. I'm guessing
my 50% was probably overkill.

--Michael

Michael G. Martin wrote:

> I've had this happen on 2 seperate servers now.
>
> After reading the docs, I bumped up shared_buffers. On one machine
> with 2G pyhsical ram, I set the param to use 1G of memory ( 131072
> value), on another machine with 800M of RAM, I set the value to about
> 500M ( 64000 ). ipcs shows the correct amounts allocated.
>
> Both servers run fine for a bit, then at some point, the entire box
> freezes. Pings work, but nothing else does, so a hard reboot is
> necessary.
>
> Any ideas. Any limits on what you can set these to. I thought these
> values would leave plenty for the other stuff to run on the server.
>
> Here is a top output before freezing:
>
> 9:14pm up 38 days, 12:47, 2 users, load average: 4.78, 5.12, 4.91
> 101 processes: 99 sleeping, 2 running, 0 zombie, 0 stopped
> CPU0 states: 24.1% user, 7.2% system, 0.0% nice, 68.1% idle
> CPU1 states: 28.0% user, 6.4% system, 0.0% nice, 64.4% idle
> Mem: 898892K av, 897300K used, 1592K free, 0K shrd,
> 0K buff
> Swap: 819272K av, 65792K used, 753480K free
> 805924K cached
>
> PID USER PRI NI SIZE RSS SHARE STAT %CPU %MEM TIME COMMAND
> 3370 postgres 9 0 382M 382M 381M S 0.1 43.5 10:53 postmaster
> 32762 postgres 10 0 104M 104M 103M S 15.2 11.9 65:25 postmaster
> 1226 postgres 9 0 54372 53M 52852 S 0.0 6.0 0:08 postmaster
> 1334 postgres 9 0 47756 46M 46240 S 0.0 5.3 0:03 postmaster
> 1181 postgres 9 0 46184 45M 44592 S 0.0 5.1 0:12 postmaster
> 1227 postgres 9 0 39796 38M 38328 S 0.0 4.4 0:06 postmaster
> 1228 postgres 9 0 25072 24M 23580 S 0.0 2.7 0:05 postmaster
> 9082 postgres 10 0 16608 16M 15180 D 5.0 1.8 0:00 postmaster
> 9084 postgres 10 0 14700 14M 13316 S 4.6 1.6 0:00 postmaster
> 3244 postgres 9 0 13376 13M 12052 S 0.0 1.4 0:00 postmaster
> 32668 postgres 9 0 11488 11M 10224 S 0.0 1.2 0:02 postmaster
> 32669 postgres 9 0 11136 10M 9888 S 0.0 1.2 0:55 postmaster
> 9085 postgres 15 0 10820 10M 9520 S 2.5 1.1 0:00 postmaster
> 9087 postgres 18 0 10796 10M 9496 R 2.9 1.1 0:00 postmaster
> 9086 postgres 16 0 10696 10M 9400 S 2.3 1.1 0:00 postmaster
>
> Thanks,
> Michael
>
>
>
> ---------------------------(end of broadcast)---------------------------
> TIP 4: Don't 'kill -9' the postmaster