Re: Solaris and running out of file descriptors

On Fri, 28 Jan 2000, Jens-S. Voeckler wrote:

> On Fri, 28 Jan 2000, Peter K wrote:
>
> ]squid-2.2.STABEL5 including all relevant patches available on
> ]squid.nlanr.net
> ]
> ]configured --enable-gnuregex --enable-async-io --disable-icmp
> ] --enable-kill-parent-hack --enable-snmp
> ] --disable-cache-digests --enable-heap-replacement
> ] --disable-forw-via-db
>
> DO USE poll with Solaris - though it should be selected by configure. The
> Solaris select() only allows for 1024 FDs, regardless of any higher
> limits.

Hmmm. Wonder why it didn't. Yet squid limits itself to 128.

> ]/etc/system contains
> ] set rlim_fd_max = 1024
> ]
> ]dns_children is set to 8, which seems just enough to cater for peek
> ]times.
>
> During startup, Squid reports in cache.log the number of FDs it can
> actually open: "With XXXX file descriptors available"

The cache.log attached to the original message reports 128.

> ]It is only occasionally that 'WARNING! Your cache is running out of
> ]filedescriptors' occurs in the cache.log files (19 times during the last
> ]9 days)
>
> Depending on the load of your Squid, 1024 may not be enough. Setting the
> hardlimit in etc/system to a higher value is (usually) without trouble for
> other applications. Frequently look into your "client mgr:info" page, or
> better, let MRTG do it for you, and you might see thightnesses.

Currently 128 seems "Just not enough during peak" (TM) at times.

I'd like to know how the hard limit of 1024 set in /etc/system will be
honoured by squid.

Peter Kooiman | Voice : +27-12-547-2846
                                        | Cell : +27-82-321-3339
Box 81214, DOORNPOORT, 0017, RSA | e-mail : pko@paradigm-sa.com
Received on Fri Jan 28 2000 - 02:32:41 MST