On Tue, 22 Dec 1998, Andrew Cormack wrote:
> I would hope that cache digests are associated with the IP address of
> the host they came from, or at least its unique canonical name ?
Cache digests are associated with whatever one specifies in cache_peer line.
If the host the bogus request came has just one cache_peer line for your
caches, there is no way a digest module can guess that you have not one but
more caches hiding behind a single name. Recall that digests use HTTP as a
distribution mechanism.
A solution is for your peer to list two caches rather than one. And yes, this
will make your DNS RR useless for that peer. An alternative is to allow miss
access on the second cache. Finally, you can wait for the next release of
Squid -- AFAIK, it will handle false misses.
Does the same problem exist for ICP or do we hack around it by storing the IP
address of the cache we got an ICP_HIT reply from?
A similar hack will not work for the current implementation of digests as a
digest for cache A can be fetched from cache B if cache C decides that B is
"closer" than A while looking for A's digest.
Just a yet another example of bad things happening when you fiddle with
protocols HTTP relies on.
Alex.
Received on Tue Dec 22 1998 - 10:39:54 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:43:42 MST