More forwarding loops

Hi,

I am maintaining a bunch of squid hosts, and currently I am moving to 2.0.2
at a host every other day. There are five caches A..E in a logical domain
group, each has the other four configured as peers. Strangely enough I am
seeing more forwarding loops (in the logfile) than ever with the 2.0.2
squids. So far, there is just one 1.1 squid left in the group.

Anybody else experiencing this kind of noise - it is not really trouble, as
the loops are broken when detected, but still bothersome.

1) Peer cache B should not have answered positively to an ICP request
   from A if it doesn't have the object. A should fetch things directly,
   if there are no positive answers. Let's assume B still has the object
   as of the ICP query from A, and answered positively.

2) Peer B gets the HTTP forwarded from A. If the object expired in the
   meantime, B queries all peers again using ICP. Now A *should* *not*
   answer here positively, unless it has fetched the object
   *completely*. But somehow B concludes that A knows the answer and
   forwards the request back to A.

3) Sometimes the loops go A -> B -> C -> D before being detected.

- Of course, I newfs'ed the cache when moving from 1.1 to 2.0, the
  reason I am not switching all caches at once.
- There are no "unique_hostnames", and all visible hostnames are unique,
  as are the real hostnames (they map 1:1).
- There is no active "never_direct" nor "always_direct" component.
- Additionally, there are two parents F and G each for requests belonging
  to the opposite logical domain group, but those shouldn't matter here.

Le deagh dhùrachd,
Dipl.-Ing. Jens-S. Vöckler (voeckler@rvs.uni-hannover.de)
Institute for Computer Networks and Distributed Systems
University of Hanover, Germany; +49 511 762 4726
Received on Thu Oct 22 1998 - 05:56:38 MDT