bug with ndc and bind 8.2

WOL - Odinn Sorensen os at wol.dk
Sat Jun 5 22:26:49 UTC 1999


Hello bind-users at isc.org.

We are an ISP with over 5500 domains and we are also secondary for the dk
tld. We are planning to upgrade our nameservers from bind 4.9.7 to bind 8.2 
(with patch1), but during our testning we have encountered a few problems.
We currently run our nameservers on RedHat Linux, intel platform. The tests 
were done with RedHat 6.0 and the bind 8.2 rpm that came with the
distribution, which includes patch1.

We can crash named by commenting out one zone in named.conf and then running
"ndc reload". Then named hangs unresponsive for about a minute and finally
crashes with these lines in the log:

Jun  5 22:13:56 voyager named[439]: Ready to answer queries.
Jun  5 22:13:56 voyager named[439]: ctl_wrtimeout: /var/run/ndc: write
timeout, closing
Jun  5 22:13:56 voyager named[439]: ctl_srvr.c:660: REQUIRE(sess->state ==
writing) failed.

The crash does NOT happen if instead we run "killall -HUP named". Named
still hangs unresponsive for a while, but comes to life again without
crashing.

Interestingly, if we comment out the "dk" slave zone (which is over 12MB in 
the backup file), "ndc reload" doesn't crash named.

Is there a fix for this bug?

Is it normal that named hangs for a while when reloading and the only change
is that a (simple, just a few hosts) zone is commented out?

Another (not very serious, but annoying) problem is that named does not
answer queries while it is (re)loading zones. This is annoying if there are 
a large number updated zones. We were under the impression that bind 8.x
could handle queries and zone loads at the same time?

-- 
Med venlig hilsen / Best Regards
  Odinn Sørensen (System developer) / World Online Denmark A/S
  Peter Bangs Vej 26, 2000 Frederiksberg, Denmark
  Tlf. (+45) 38 14 70 00 - Fax (+45) 38 14 70 07



More information about the bind-users mailing list