Hi, On Sat, Sep 02, 2006 at 09:58:58PM +0200, Ralf Hildebrandt wrote: >>>I am using diskd, with 5 cache dirs on each server (each dir containing >>>quite a lot of directories). If I turn off diskd and use plain ufs cache >>>dirs, the instabilitity problem goes away, but then the servers become >>>very slow to respond. :-( Which I/we can not live with. As 2.6 is >>>supposed to be faster than 2.5 because of kqueue, downgrading doesn't >>>seem like a good option either. > I "fixed" that by running squid under "runit" Squid doesn't die (and stay dead) non me, it just restarts itself. But when it happens, I lose the cache in RAM, which is the main problem. Also, the bug occurs more often when under high load, making cache hit ratio fluctuate a lot. So, runit wouldn't help me. I need to protect my backend servers from load as much as possible. Squid needs to run constantly and stable for day after day, keeping a constant high cache hit ratio (which it does fine as long as it is stable). Anyhow, I've done preparations to produce backtraces, and will get back on this once I have more to report. Cheers, -- Anders.
