Bind server crashing (lots of EAGAIN, ENOENT, ...). With strace log.

[Mike Hoskins (michoski)]

-----Original Message-----

From: Alan Clegg <alan at clegg.com>
Date: Tuesday, November 5, 2013 9:28 PM
To: "bind-users at lists.isc.org" <bind-users at lists.isc.org>
Subject: Re: Bind server crashing (lots of EAGAIN, ENOENT,	...). With
strace log.

>
>On Nov 5, 2013, at 7:31 AM, K L <kl.forwarder at gmail.com> wrote:
>
>> Here is a strace log from when it happens:
>>http://pastebin.com/raw.php?i=7i0PgALG . Example:
>> 6500 recvmsg(518, {msg_name(16)={sa_family=AF_INET, sin_port=htons(53),
>>sin_addr=inet_addr("10.0.101.50")},
>>msg_iov(1)=[{"~\223\201\200\0\1\0\1\0\5\0\6\3ns3\5cymru\3com\0\0\1\0\1\30
>>0"..., 4096}], msg_controllen=32, {cmsg_len=32, cmsg_level=SOL_SOCKET,
>>cmsg_type=0x1d /* SCM_??? */, ...}, msg_flags=0}, 0) = 252
>> 6500 recvmsg(518, 0x7fd4b6588900, 0) = -1 EAGAIN (Resource temporarily
>>unavailable)
>
>What about more "normal" bind logging?  Anything useful in there?

It's been awhile since I looked at a strace of named under load, but seem
to recall this being normal for non-blocking sockets...take that with a
grain of salt, it's was long ago when I went looking for EAGAIN/ENOENT on
google.  :-)  I'm sure someone from ISC can provide a lot more insight
into "normal" operation.

Aside from logs, how does rndc status look under load?  Do you have
anything like cacti or graphite pulling numbers out of statistics output,
to possibly spot any trends?  What options are you running in named.conf,
and what is the BIND version?