On Fri, May 21, 1999 at 10:24:00AM +0100, Robin Bowes wrote:
> Hi,
>
> Some time ago (1 month, actually) I sent the following message. I don't
> think it made it to the list (I was posting via my mail/news gateway) so
> I've re-submitted it.
>
> The reason for the renewed interest is that I've had exactly the same
> problem again. I've shut down my cache box to check the BIOS and, when I
> restarted, it began re-building swap.state and is still going...
>
> I use "shutdown -h now" to bring down the FreeBSD box. Should I shutdown
> squid first?
YES!
Add squid -k shutdown;sleep 10 somewhere to your /etc/shutdown.local if
something like that exists (don't know FreeBSD so well).
> Any ideas how I can prevent this from happening again in the future?
Do you rotate your logs regularly? (squid -k rotate)
This will "write clean logs", I'm not sure what this means but I see
this in my cache log. Maybe this will help peed up recovery after squid
has crashed or was killed or whatever.
> I've got squid 2.1.PATCH2 running on a FreeBSD 3.1-RELEASE box. The
> machine is a P133, (was 96M) 128M RAM, with a 1G system disk (256M swap)
> and 8G cache disk, although I've limited the cache size to 4GB. I've
> got soft-updates on both disks.
>
> It seemed to be working OK when I first installed but I've recently
> shutdown the box to replace the RAM - I've upped it to 128M.
>
> Everything came back up OK but it took forever to rebuild
> swap.state. I suspect that this is because the squid process is 145M in
> size and I've only got 128M of RAM so there was continual swapping.
It should not swap at all. With 4GB disc cache our squid uses
about ~35MB of RAM on our NetBSD/i386 box. (less during cache rebuilt)
We have only 64MB of RAM and this box is running quite a few other
services...
> Here's cache.log:
[...]
> 1999/04/20 14:13:19| 1376256 Entries read so far.
> 1999/04/20 14:13:58| 1441792 Entries read so far.
> 1999/04/20 14:14:38| 1507328 Entries read so far.
> 1999/04/20 14:15:27| 1572864 Entries read so far.
> 1999/04/20 14:19:15| 1638400 Entries read so far.
something bad has happen here (big slowdown now)
Has squid grown too big here?!?
> 1999/04/20 16:56:22| 1703936 Entries read so far.
> 1999/04/20 22:39:15| 1769472 Entries read so far.
> 1999/04/21 06:47:12| 1835008 Entries read so far.
> 1999/04/21 19:53:30| 1900544 Entries read so far.
[...]
> 122296 seconds is just under 34 hours!!! During this time, client
> performance was awful.
A complete rebuild (with deleted swap.logs) should not last
more than maybe *maximal* 2 hours for 4GB.
> Is this normal behaviour? Should this file grow so large? Will the
> squid process return to normal size once this file is built? Are there
> any configuration parameters I can tweak to reduce the size of this
> file?
No it's not.
Upgrade to 2.2.PATCH3 wich is the latest stable release AFAIK.
Maybe this will fix your problems. Wich file, btw?
-- Dies ist Thilos Unix Signature! Viel Spass damit.Received on Fri May 21 1999 - 09:38:14 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:46:23 MST