Slow expireover (again)

Russ Allbery rra at stanford.edu
Sun May 23 06:51:02 UTC 2004


Mike Zanker <mike-sender-6677e0 at zanker.org> writes:

> I posted last week about slow expireover on an Ultra 10 with a single
> 80GB IDE disk. This has been getting longer by the day and was still
> running this morning (after nearly 7 hours). I've run vmstat to get some
> performance figures but I'm not a Solaris tuning expert so would
> appreciate it if anyone can see if there is anything obvious here.

Hm.  Is the amount of free swap you have in your system dropping slowly
over time by any chance?  (This is a real shot in the dark, but I did find
a Solaris kernel bug that caused the kernel to slowly consume all system
memory until the system didn't have enough memory to run INN, but cleared
up on a reboot.  That was a bug in an older version of the RAID Manager
driver for A1000s, though, and appeared to be related to memory mapping
raw devices.)

If you run truss on the running expireover process (with truss -p), you
should see a ton of stats intermixed with writes.  Is anything looking
particularly slow, or does the output stop scrolling at any particular
points and then start again?

-- 
Russ Allbery (rra at stanford.edu)             <http://www.eyrie.org/~eagle/>

    Please send questions to the list rather than mailing me directly.
     <http://www.eyrie.org/~eagle/faqs/questions.html> explains why.


More information about the inn-workers mailing list