Truncated names and DNS troubles in Squid 1.1.6

I run Squid 1.1.6 on a SGI Challenger S running IRIX 5.3, and I have
had two problems the last few days.

1. Truncated names

Our running Squid process suddenly truncated the name of one of our parents
to 'it.cache.nla' instead of 'it.cache.nlanr.net'. This caused a
ERR_DNS_FAIL for every request to a domain that this parent served.

The day after it truncated the name of the "cache.log" file to
"cache.l".

Both of these problems where solved by HUP'ing the process, and I have
not been able to reproduce these errors. But something is definitely
wrong when Squid starts truncating these internal variables.
 

2. DNS troubles

On the UNINETT web-cache server I have configured some of the
web-cache servers at nlanr.net as parents for several domains (.edu,
.gov, .mil, .us). But today the nameservers for the .net domain failed
(we lost contact with the .net domain and several .com domains) and
the following happened:

 * The Squid process had to re-resolve the hostname for one of our
   parents at nlanr.net, but this failed because the DNS server
   for the .net domain was unavailable, and nothing was cached in
   the DNS servers at our side.
 * Our Squid process then failed to fetch any URL for all the domains
   this server was a parent for, and it gave this error message to
   the clients:

        While trying to retrieve the URL:
        ftp://hyper.stanford.edu/pub/pcmcia/SUPPORTED.CARDS

        The following error was encountered:

             DNS name lookup failure

        The system returned:

            DNS Domain 'it.cache.nlanr.net' is invalid: Host not found
            (authoritative).

   As you can see the host it did not find was our parent, and not the
   host given in the URL. This was logged as a ERR_DNS_FAIL in the
   cache.log file. But the URL from the request was printed after the
   error-message, not a warning about a failed dns-lookup on a parent.
   (Exactly the same thing happened when the parent-name got truncated
   earlier on.)
 * By bypassing the cache I was still able to fetch objects from other
   domains (.edu) which this parent served.

Perhaps a conclusion from this should be that a parent is considered
dead if it is impossible to resolve it's name. And Squid should then
try to fetch the objects from the source, or from an alternative
parent? (In this case I had two alternative parents, both on the
unavailable .net domain, so contacting them would have been useless.)

PS: Error number 3? I tried to HUP the Squid-process when I discovered that
it was unable to resolve the name of the parent. I thougth the Squid
process had truncated the name again, but then the server crashed
with this in the cache.log:

        97/02/21 13:30:05| WARNING!!: DNS lookup for 'it.cache.nlanr.net' failed!
        97/02/21 13:30:05| THIS NEIGHBOR WILL BE IGNORED.
        97/02/21 13:31:27| Configuring Sibling dc.cache.nlanr.net/3128/3130
        97/02/21 13:31:27| WARNING!!: DNS lookup for 'dc.cache.nlanr.net' failed!
        97/02/21 13:31:27| THIS NEIGHBOR WILL BE IGNORED.
        97/02/21 13:32:40| Configuring Sibling pb.cache.nlanr.net/3128/3130
        FATAL: ipcache_add: somebody adding a duplicate!
        Squid Cache (Version 1.1.6): Terminated abnormally.

It restarted again before I had time to comment out the relevant lines
from the configuration file. But fortunately it did not crash this time,
it just ignored the parents (defined with neighbor_type_domain).

-- 
| Lars Slettjord             | EMAIL: larss@cc.uit.no
| University Computer Centre | URL  : http://www.cc.uit.no/~larss/
| University of Tromsoe      | PHONE: +47 77644115
| N-9037 Tromsoe             | FAX  : +47 77644100