Umm - Lazarus files?

From: Dancer <dancer@dont-contact.us>
Date: Tue, 10 Mar 1998 23:36:19 +1000

This isn't properly a squid question. But it isn't exactly not one,
either.

Let me rebuild the chain of events. Maybe someone can shed some light on
what occured:

18:02 - One of my redirectors goes offline. Unusual. They do no dynamic
allocation and are compiled. They don't touch malloc() or free() once
running, and do no file accesses. They reload on a signal, and that's
it. They've caused no trouble since they were written, and I've never
had one die before.

18:03 - A second redirector goes offline. Two down out of five. Squid
terminates 'FATAL: All redirectors have exited!' Three redirectors were
still running at that point, as near as I can determine.

18:03 - (still) Squid is restarted by the watchdog, and now the
mysterious happens. I use machine-generated squid.conf files. I have
component files that I pass through the C preprocessor and then catenate
together with a Makefile to build the final squid.conf, sans comments
(except a 'do not edit me' warning). This squid.conf is _months_ old,
and specifies a cache that is several GB smaller than was allocated more
recently. It's not one of my machine-generated ones.

18:03-23:02 - Squid obediently starts throwing out cache-objects
willy-nilly, trying to get below the high-water mark, and bitching and
whinging in it's cache.log the whole time.

23:03 - Convinced that it's a lost cause, squid shuts down.

23:04 - Squid restarts (watchdog) and continues throwing out cache
contents.

23:06 - I glance at the squid graph, and notice that the last couple of
minutes have had no squid-requests. Odd time for it. Should be busy 'up
the wazoo' (as I think americans sometimes say). Double-check the time.
No it's one hour too early for log-rotation. Look at numbers next to
server entry on www-page. Only 176 connections since startup? Definitely
something wrong.

        Check cache.log. Find problem. Quick less of squid.conf reveals wrong
cache_swap size. Quick look at surrounding lines shows it to be one of
the old-style (or current-style for the rest of you) squid configs. '#'
comments, instead of '//', and so forth. Not one of my machine-generated
ones. Flustered, I jumped to the component directory and did a make,
writing a machine-generated one over the one in place, shutdown squid
and let the watchdog restart it.

        Why didn't I look at the timestamp on the rogue file, and maybe save a
copy? I panicked. It's peak-usage period. Stupid. Stupidstupidstupid.

        Many minutes of security checks. No, nobody's been into the box.
Nothing's touched, everything checks. Either that, or the archangel
Gabriel has lain down his horn and gone into system cracking.

        Just that one lousy file, returned from the grave to feast on the
caches of the living.

Baffled. Illumination, anyone?

D
Received on Tue Mar 10 1998 - 05:46:51 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:39:14 MST