All, we are running a few BIND 9.3.5 servers on Solaris and
occasionally run into issues with servers that do not apparently like
EDNS0 queries (boeing.com's DNS servers for example).

What seems to be happening is that BIND's ENDS query failure detection
seems to stop working. With a packet sniffer I can see queries with
the extra EDNS payload going out and observe a "Format error" response
from the remote DNS server. BIND then cycles onto the next DNS server.
Typically after failure from all the available NS servers, BIND is
supposed to retry again *without* EDNS enabled. However, there clearly
are times when BIND does not do this and simply stops after the
rejections from the remote server. No additional packets are sent.

If I restart BIND, it seems that the proper behavior is resumed.

I only just stumbled across this problem (we couldn't successfully
resolve any boeing.com DNS records) so I'm not sure how reproducible it
is.

Has anyone else run into anything similar? As a workaround I'm simply
disabling edns completely for remote servers that I know do not support
EDNS.

I see a mention to a bug:

2383. [bug] named could double queries when they resulted in SERVFAIL
due to overkilling EDNS0 failure detection

In the latest 9.3.5 release notes as an upcoming bug fix. Is ISC's bug
tracker publicly available? Perhaps this is the issue I'm running
into?

Thanks!
Ray