Bind Recursion limitation?

[Simon Waters]

John Wingenbach wrote:
> 
> I am running bind 9.1.3 and have run into an interesting 

I tend to take the view that the first thing to do if you think
it is a bug is upgrade to 9.2.1.... Which is slightly faster
anyway I beleieve.

> Each query attempt fully utilizes a thread of bind?

Not a chance, out of the box 9 has 1024 limit of outstanding
recursive queries, this can be increased. It may have only one
thread, but the number of threads is related to number of CPU's
not number of queries.

> Am I off base anywhere here?

Yes, some penalties are applied for unresponsive name servers,
but if BIND is asked after a SERVFAIL it tries again.

I would check the back off strategy of the client application.

> Normally they handle the load with out an issue.
> However, they have experienced a few times where they completely got
> overloaded without actually performing many queries. 

Is anything logged?

> If my assumptions are
> correct, then a caching of dead name servers would help resolve this
> predicament.  Basically, when all name servers for a domain come back not
> answering, mark that domain unreachable for 60 seconds or something.

Hmm, that would delay recovery if a temporary network glitch
caused all remote name servers to be unavailable, quickly every
domain would be on a 60 second delay.

You could try increasing the recursive client limit, but I'd
suggest 9.2.1 first.

Why are so many domains failing SERVFAIL on an internal network?
Sound like a network redundancy problem....