Henrik Nordstrom schrieb:
> Jens-S. Voeckler wrote:
>
> > Nope, half-closed FDs are closed and released after some time of
> > inactivity on my machines. The only problem (still remaining since
> > squid-1.1) is that with a workload higher than the host can handle, I have
> > loads of FDs remaining in CLOSE_WAIT until I reach my FD limit (which I
> > set to 4096). More than 200 CW handles is a good sign to restart squid
> > (still found no other solution).
>
> Thats most likely half closed client connections. Unless you disable
> half closed clients, aborted client connections on stalled server
> connections will stay in CLOSE_WAIT until the server connection times
> out (which could be something like 15 minutes later).
Sigh, I have to find (yet) that particular timer to tweak (in Solaris). It
is *not* tcp_close_wait_interval (that was actually renamed to its real
descriptive name tcp_time_wait_interval in Solaris 7).
> half_closed_clients off
>
> If you continue to get sockets seemingly stuck in CLOSE_WAIT then there
> is a problem in Squid somewhere.
Nope, it went away. Though I have to admit that I like the idea of
half duplex transmits, e.g. I am frequently using Stevens' sock tool
in half-closed mode (because Apache supports it) to send queries
via my cache.
-- Le deagh dhùrachd, Dipl.-Ing. Jens-S. Vöckler (voeckler@rvs.uni-hannover.de) Institute for Computer Networks and Distributed Systems University of Hanover, Germany; +49 511 762 4726Received on Wed Dec 08 1999 - 06:44:57 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:49:47 MST