On Thu, Jan 31, 2008 at 08:36:05PM +0900, Adrian Chadd wrote: > On Thu, Jan 31, 2008, Amos Jeffries wrote: > > If the delay between starting and stopping is long enough and can be > > done with X + random-time offset. Squid will cope with some helpers > > simply stopping and resumes them. > > Yeah, the trouble is that you can't kill them all or Squid will die > complaining the helpers are dying too quickly. Yep, I've definitely run into that condition. =) Squid recycles itself, but this causes the same (if not worse) downtime as the one I'm trying to avoid. > Hm, I've always wanted to fix that pause during reconfigure and > rotate. Of course, reconfigure's "deny accepting connections" is > probably it closing and re-opening all its listen() sockets.. That's what my reading of cache.log suggested - that Squid closes its listening socket, restarts its children (which can be slow), and only then re-opens the socket. Anyone trying to connect during the restart gets a TCP RST. --Emil
