Weird Named error...
Tim Peiffer
peiffer at umn.edu
Thu Apr 21 11:38:37 UTC 2005
What is the complete command line you are using? Are you running
chroot, and/or in a sandbox? The sigsuspend() system call is
understandable if named is launched as foreground and ^Z suspended.
;=) But, you don't need to run in the foreground. I am currently only
running named as
<path to named>/named -c <path to configs>/named.conf.
We have run into similar behaviour using Bind8.4.7 on Red Hat AS3.0. In
our case, the behaviour came as we were rotating logs. The system load
was light ~300qps. We log all of our transactions and shuttle them out
using syslog. Anyway, as logs were renamed, new files created, and a
SIGHUP sent to syslog to re-open all file handles, named seemed to block
on IO. The process seemed to be wedged and had to be killed off and
started fresh. Since then, we have not been able to recreate/simulate
the problem. We do believe that the problem is OS and Syslog
implementation dependent. NB; Synchronous file system writes (default
RedHat syslog behaviour) carry a high penalty and is easily seen under
load - Turn off with '-' (minus) in front of the file name for syslog.
/etc/syslog.conf:
local0.* -/var/log/named.log
On an offline box using Bind9.3.1 and queryperf, we were driving the
box to 4700 qps and found that things got pretty spongy when the input
queue (netstat -an | grep 53) got to about 65000 packets in the queue.
But still, named was always able to dig itself out. Bind performance
returned to normal when the input queues got to below 62000 packets.
Tim Peiffer
University of Minnesota
Braun Brelin wrote:
>Hello all,
>
>I have a strange named problem which I'm hoping someone can help me solve.
>I'm running a RedHat 7.2 system with Bind 9.2.1.
>
>Currently, BIND only resolves names if I run it with the -g option.
>
>I.e. named -g -u named will resolve names via nslookup or dig.
>
>If I run it normally i.e. via /sbin/service or /etc/init.d/named start
>it will still show up as a process (actually it shows up as 4 processes in
>the process table) but nslookup timesout with a no connection error.
>
>I ran strace named -f -u named just to see if I could find out what was
>happening. It seems as though named is stuck in a sigsuspend() system
>call apparently waiting for some signal or another to continue.
>
>The only thing that has changed on the box was that the /var fs filled up
>yesterday (It's since been cleared). All of the zone files look normal.
>
>Has anyone seem this sort of behavior from named before?
>
>Thanks,
>
>Braun Brelin
>
>
>
More information about the bind-users
mailing list