[squid-users] Fault tolerant accelerator

Hi all,

I'm looking to place a (hopefully simple) squid accelerator array in front
of one of my http/xml app-servers.

currently we are running one squid, and are quite happy. but would like to
have two in case the hardware on one fails. the plan is to have a BIG/IP
round robin requests to each if they are alive.

the problem is this.

we would like to prevent as few requests from reaching the origin server as
possible (the xml/app-server).

we need to invalidate the cache on all members of the array (currently two
proposed machines) when something becomes pre-maturely stale.

when invalidating the object, would prefer to hit the 'system' and not each
individual squid instance (just make one request to clear the object)

so, if I understand things correctly. in an array of two squids as siblings:

1) if squid A gets a normal GET request, it does one of three things:
- serve a cache hit, or
- ask it's neighbor squid B and serves from it's cache (and store local), or
- serve from the origin server (and store local)

2) but, if I do a PURGE and if squid A gets the request:
- it clears the cache (does not notify squid B)

3) or, if I do a GET with No-Cache header against squid A:
- it serves from the origin server (and stores locally, does not notify
squid B)

assuming those assumptions are correct, it looks like I'm sol. or is there
still some config twiddling I can perform to effectively get me what I'm
looking for.

for example, squid A does not cache neighbor cache hits, so when it receives
a GET with a no-cache, it hits the neighbor (instead of the origin) since
it's a local miss, forcing the neighbor squid B to re-fetch the object,
store it, and return it to squid A. (long shot)

sorry for the long email. but I would appreciate any pointer or references
if there are any.

cheers,
chris
Received on Thu Oct 07 2004 - 13:54:15 MDT