I've recently upgraded form 1.1.5 to 1.1.16 (and subsequently to 1.1.17 on
one machine) and I've noticed a couple of oddities.
1. When you stop the cache, it waits the requisite number of seconds for
connections to close down, and then it waits even longer for the unlinkd
to finish doing whatever it's doing. At least this is what I assume. In
both the cases where I have noticed thuis behaviour, the unlinkd has been
busy reducing the size of my cache, and when I killed that process off
independantly, the cache continued it's shutdown cleanly.
It would be nice (TM) if the unlinkd could be stopped cleanly by the cache
when it is ready to close down.
2. Squid is using more memory than it did with 1.1.5. It is also using
more filehandles than it used to. Given the discrepancy between what the
cache manager accounts and what the process is actually using varies from
a factor or 2 to a factor of 4, [90M accounted, anywhere from 180M to at
one point 398M used], is now much worse than it used to be, my
rationalisation is that most of this 'lost' memory is being used by
in-transit data.
How can I tune this down? My system is memory limited. I have already
played with the connect_timeout and read_timeout, but it seems to have
made little difference. Or is my rationalisation wrong?
3. In the one worst case above, where squid reached 398M of memory, this
was immediately after I restarted it, which I'm doing periodically to
shrink the size of the running process. In this one case when I restarted
it it rapidly grew to twice it's usual size. It stopped and restarted it
again, and it grew to it's usual initial size of 110M or so.
Has anyone else noticed a peculiarity like this?
I'm now running 1.1.17 on one of our machines, but the ChangeLog doesn't
suggest any quick fixes for any of these. The number of filehandles used
by the 1.1.17 is much the same as it was with 1.1.16, though it hasn't
been running long enough for me to determine if it still has the same
memory acquisition abilities.
Thanks.
John
John Sloan
Systems Administrator
UUNET UK
Received on Wed Oct 22 1997 - 02:48:07 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:37:19 MST