bind in a non-killable state?

[Kevin Darcy]

Generally speaking, when a process is non-SIGKILL-able, this indicates a bug
at the kernel level, either in the kernel itself, or in a driver. What system
call was named stuck in? Debian has a system-call tracer, doesn't it
("strace" or something like that)?


- Kevin

Leif Neland wrote:

> Our main nameserver was hit with a powerfailure.
>
> Later, when it was up again, Big Brother (http://bb4.com) noticed ssh and
> smtp was not working on a remote server. This server also has its own local
> named running.
>
> I tried logging on, and after a long wait, which indicated a DNS-timeout, I
> got on.
>
> named was showing in ps, but didn't give any answers.
> I tried to kill -9 named, but it didn't want to die.
>
> I tried to do a reboot, but that was friday aftenoon, and now saturday
> evening I still haven't seen it...
>
> What can cause bind to be unkillable? Could it have been in the middle of a
> query to our main nameserver, when it got hit by the powerfailure?
> Is there an infinitely retrying loop somewhere?
>
> The remote server is running Debian 2.2 with the latest debian-packaged
> bind.
>
> Leif