Re: icp_hit_stale & allow-miss

ons 2006-04-05 klockan 02:07 +0300 skrev srnoam@t2.technion.ac.il:
> Hi,
> I have several questions about using "icp_hit_stale on" & "allow-miss" :
> 1)when no icp sibling in use: If a client ask proxy server for a stale object
> would it return it direcly or treat it as a miss.

See below.

> 2)similar to question 1: If I have one proxy server (using ICP but all siblings
> are dead) & my cache is stale, whould I return my object or go to origin
> server?

Depends. It will ask the origin server about the objects freshness. If
the origin server says your copy is fresh or Squid fails to contact the
origin then the cached copy is sent, otherwise the new response from the
origin.

> 3)When I get a stale response from a sibling: do I think it's HIT or I know it's
> a stale response?

You never get stale responses from a sibling. The request your Squid
sends to the sibling has cache constraints only allowing responses which
you consider fresh. (via cache-control: max-age)

> If I know it's stale then, do I take the first response as in regular hit?

The icp_hit_stale directive makes this Squid return HIT responses to ICP
queries on stale objects. Only used by ICP peers.

The sibling cache_peer allow-miss option makes Squid forward the request
as if it was a parent relation, i.e. allows the contacted peer to not
only serve the request from cache but also to go to the network. Mainly
useful in a sibling relation where the sibling have icp_hit_stale
enabled.

The miss_access directives controls who is allowed to ask for cache
misses via this Squid.

If you have a stale copy this is first revalidated before given to the
client. In a sibling relation without allow-miss or when miss_access is
denied then revalidation is not possible and an error message is
returned instead.

Regards
Henrik