Re: false hit recovery?

Alex Rousskov wrote:

> Because a workaround can be ready tomorrow while false hit retry
> code is still in the works, and because there will be a need for
> some "migration" solution before everybody upgrades to the version
> of Squid that can retry false hits.

Is there anyone working on false hit recovery?

Whats needed:
1. forward.c needs to know more than one address.
2. The protocols (esp http) needs to be able to tell forward.c that the
request failed.

Regarding the list of addresses to forward.c:
It should be a sorted list of addresses to peers and the origin site,
all marked accordingly (proxy or origin site).

always_direct
1. origin site

never_direct
1. HIT siblings/parents
2. other parents

hierarcical
1. HIT siblings/parents
2. other parents
3. origin site

non-hierarchical
1. HIT siblings/parents (or perhaps not).
2. origin site

forward.c should then cycle throught the list of addresses, and fail
when the complete list has been tried two or three times.

Perhaps there should be a limit on the number of HIT siblings to try.
Imagine peering with 20 siblings, and all give a false HIT on the same
object (not to unlikely). We then do not want to probe all 20. 2-3 is
probably a reasonable limit on the number of HIT siblings to probe. Note
that this restriction only makes sense on siblings and not parents
and/or origin site.

/Henrik
Received on Tue Jul 29 2003 - 13:15:54 MDT