Re: TCP_SWAPFAIL_MISS and asyncufs/2.3

On Thu, Apr 13, 2000 at 10:46:06AM +1000, Kevin Littlejohn wrote:
>
> Check your cache.log for frequent restarts. Squid seems to dislike being
> started twice, or killed and restarted in unfriendly ways - if it's
> crashing and restarting, it may be corrupting it's index, which would lead
> to that problem. For it to occur for that timeframe, I'd guess it might be
> happening semi-regularly.

  well kevin I woudl agree but, I can't add credence to this in this
case. I did some searching of logs this morning for more details. The
sequence was as follows:

<02/07 running 2.2.STABLE2 (my cache logs only go back to here)

03/03 upgraded 2.3STABLE2

03/08 cleared cache (I had for gottend about this event details
      follow)

03/15 restart on system reboot. (clean shutdown and restart for other
      maint)

04/10 cache access seemed slow, cache stopped and restarted. problem
      found to be a dns issue on local system.

04/11 cache access seemed really slow when I came in the morning, I
      stopped and restarted. after restart ran for a short time (15
      minutes and and response dropped again) at this point I
      discovered the TCP_SWAPFAIL_MISS messages. cleared the cache and
      restarted, been running fine since.
      
      the TCP_SWAPFAIL_MISS messages seemed to start with frequency
after the restart on 4/10. There were sporatic occurances of
TCP_SWAPFAIL_MISSes on 3/15 (after restart) and on pertty much on a
daily basis but not inordante: ranged from 2 entires to maybe 30 per
day. (I assume what Henrik termed as a "few"). This is a farily active
cache 400K accesses/day or so) I haven't gone back prior to 3/15 I
assume it would be the same.

      The 3/8 event, I had stopped the cache and inadvertently
restarted squid as root. ** do NOT try this at home kids ** It is not
recommended **
     This caused a great deal of havoc. It rewrote the pid file, among
others with root access, stumbled around crashed then restarted itself
several times. when I discovered the problem, shutdown and restarted
under squid's user the afore mentioned files were read only to the
squid user. I ended up sweeping the cache directories restoring the
proper ownership, cleared the cache and restarted. It has run a month
without issue.
     I don't recall if I cleared the disk and rebuilt with squid -z,
or just nulled swap.state. I assume the latter as it is a whole lot
faster and I do remember running a chown sweep... but I really don't
remeber the details. I must have blocked it. ;)
     I may clean off the cache dir structure and rebuild it just to
be confident that that there isn't a trashed file somewhere in L2.

      Thoughts?

fj..

-- 
	"The days are just packed!"   Calvin & Hobbs