Re: :-) Squid as a Load Balancer, . . . Yes!

In message <199705152300.QAA17567@mailsun3-fddi.us.oracle.com>, "TBRIGLIA.US.OR
ACLE.COM" writes:
>of the "redirector" program. Unfortunately to make this work you need to issue
> a SIGHUP so that Squid will reload the redirector program. This functionality
>will suffice to keep an online mirror available for redundancy, yet I'm not
>sure how practical it will be if you had the thing frequently switching back
>and forth between servers, each time having to send a SIGHUP.
>
>Can any of you experts enlighten me on how detrimental frequent or occaisonal
>SIGHUPs will be? On an idle system with a small cache, Squid recovers very
>quick from a SIGHUP as far as I can tell. How noticeable to users would it be
>if there was a SIGHUP every 2-3 minutes on a busy site???

sending a SIGHUP to a running squid, although good to make it re-read its
configuration file, has other detremental effects. it makes it close all
its FD's (ie. dnsservers, ftpgets, open sockets). (it does wait for up
to 30 seconds before doing this). this effectively means that if someone
has an active request still being served, it just stops in the middle,
incomplete.

(this is a likely scenario: a flow-analysis of http traffic here has shown
an average web request tcp stream lasts 57 seconds).

whilst it may work for short-running requests, try starting a large transfer
over a slow link and do a changeover. if the codepath of the accelerator
mode follows that of the cache-mode, you'll find you'll get a
'connection reset' message, or worse-yet, just a file that is prematurely
ended (truncated).

cheers,

lincoln.
Received on Thu May 15 1997 - 16:43:10 MDT