- Memory shortage. Montior with vmstat/sar.
2GB memory, Squid takes up 700MB. It's the only thing running on the machine.
- I/O bottleneck. Monitor with vmstat/iostat/sar.
Cache hits are fast as usual, it's just cache misses and cache near misses that go kablooey. If there was an IO bottleneck, wouldn't cache hits be affected too?
- Overload. Monitor CPU usage with vmstat/iostat/sar, and filedescriptor usage in Squid using snmp polling. Also keep an eye on cache.log.
No fd shortage whatsoever, happens when there's only 2500 fd's in use (with a maximum of 8192, I occasionally get spikes to about 5000, that's why it can't be much lower than 8192)
I decided not to automatically restart (which could result in all kinds of risks), but to send a mail when the median response time in the last 5 mins is larger than 0.35 seconds, or the fd count is larger than 6000.
Joost -- Du hast mein Herz zerrissen, meine Seele geraubt Das es so enden würde hätt` ich nie geglaubt [Aus der Ruinen -] Ohne Rücksicht auf Verluste, hast Du meine Welt zerstört [L'Âme Immortelle] Eine Welt, die vor kurzem nur uns beiden hat gehört
