named crashes/core dump problems

[Stephen L Johnson]

We have been experiencing problems with our named processes core dumping
with the following syslog messages:

Nov  9 10:05:20 achilles named[20026]: insist: critical: ns_main.c:537:
INSIST(e
rrno == EINTR): Invalid argument failed.
Nov  9 10:05:20 achilles named[20026]: ns_main.c:537: INSIST(errno ==
EINTR): In
valid argument failed.
Nov  9 10:05:20 achilles /kernel: pid 20026 (named), uid 0: exited on
signal 6 (
core dumped).

This started several weeks ago. We were running bind 8.1.2. Nothing in
syslog files are stands out as a glaring problem.

We have upgraded to servers 8.2.1 and 8.2.2 P4, but named still keep
core dumping.
The crashes can take place any time from minutes to weeks apart of the
same machine.

I tracked the problem down in the source code to the evGetNext function
call in ns_main.c. Somehow the function is being entered with the
EV_POLL and EV_WAIT flags set, even through evGetNext is called from
ns_main with only the EV_WAIT flag at the point of the core dump.

I can't seem to track the problem down a definitive event or action.