Re: cache 'log' file

ihc@ukc.ac.uk writes:

>Hi folks,
>
>I had a rather 'interesting' experience this morning.
>
>I shut down the cache (squid -k shutdown)- which I'm assuming writes a
>new version of the 'log' file (the one in the first directory of
>cache_dir) - to try 1.NOVM.6 again.
>
>Everything appeared OK after restarting. I'd checked that the
>information was being read back from the 'log' file.
>
>Unfortunately I had to take squid back down later, only to find:
>
>97/02/24 12:27:57| 258048 Lines read so far.
>97/02/24 12:27:58| Shutting down...
>97/02/24 12:27:58| storeWriteCleanLog: Not currently OK to rewrite swap log.
>97/02/24 12:27:58| storeWriteCleanLog: Operation aborted.
>...
>97/02/24 12:27:58| Squid Cache (Version 1.NOVM.6): Exiting normally.
>
>
>Upon restarting, I didn't see any '... Lines read so far' messages.
>Looking in the 'log' file directory, I noticed a log.new growing.
>Further investigation (using the cache manager) suggested that my
>cache contents had been zapped.

It sounds to me like this was a slow restart (a VERY SLOW restart).
If log.new is growing, then Squid is rebuilding the metadata.
Since the previous shutdown was not clean, Squid must stat(2) every
cache file, and since yours is so busy, this takes a very long time.

>PS - Regaring my earlier query about degradation, it appears that the
>problem is primarily concerned with swapping. Is anyone out there
>handling >500000 requests per day on similar amounts of memory (128MB
>RAM)??

The busiest NLANR box is doing about 400,000 these days, with 256M physical memory.

Duane W.
Received on Mon Feb 24 1997 - 14:41:28 MST