At Tue, 11 Nov 2008 09:29:05 -0500,
"Davenport, Steve M" wrote:

> I'm running 9.3.5-p1 on a lightly loaded secondary caching server under
> Solaris 10 sparc. I noticed from a stats graph and snoop that when a
> resolver hits this server with closely spaced, rapid queries, Server
> Fails are generated.
> 13:45:51.51634 ns-p -> hhost DNS R Error: 2(Server Fail)
> 13:45:51.51652 ns-p -> hhost DNS R Error: 2(Server Fail)
> 13:45:51.51664 ns-p -> hhost DNS R Error: 2(Server Fail)
> 13:45:51.51677 ns-p -> hhost DNS R Error: 2(Server Fail)
> ...
> During this time (or at least when I am looking)
> idle cpu is around 93% and the bind process uses about 4%
> and the named process is using about 32M of memory.
> The "rndc status" shows recursive clients: 15/1000
> Open sockets from "lsof -n|grep -c named": 40
> "netstat -I 5" shows no interface errors
> Would 9.3.5-P2 reduce the server fails or should I be looking at another
> version? Has anyone running 9.3.5-p1 under Solaris 10 seen this issue?

9.3.5-P1 could cause server failures more often when it receives a
bulk of queries in a short period.

I'd strongly recommend you to try 9.3.6rc1, which should be much
better than 9.3.5-P1 and should be as stable as 9.3.5-P2. If you
still have the same problem with 9.3.6rc1, please report it again.

JINMEI, Tatuya
Internet Systems Consortium, Inc.