[Unbound-users] TCP random timeouts

[W.C.A. Wijngaards]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi Thomas,

On 11/15/2010 01:35 AM, Thomas Guthmann wrote:
> I am encountering some TCP issues on a RHEL5.5 with unbound 1.4.6
> 2010-11-14T20:58:50.266456+11:00 ns2 unbound: [11863:0] error: tcp
> writev: Broken pipe

This indicates the remote end closed read the read end of the tcp
channel.  The error printout is maybe a bit overly dramatic, perhaps
this specific error, since it only indicates another host on the
internet closed the connection, should not be printed at low verbosity.

But this does not seem to explain the tcp issues you say you have.  What
is the observed problem again?  What queries are giving resolution
problems and timeouts - are you sure they even use tcp - and can you do
a unbound-control dump_requestlist if the query times out (it may still
be in the requestlist and we can then see what is going on) and
unbound-control lookup (that says what unbound would probably do if it
were in the requestlist).  And possibly enable a higher verbosity (which
you can do very easily without a restart; unbound-control verbosity 4;
(wait) ; unbound-control verbosity 1 ) but you would want to actually
catch the offending action in the verbosity which may get tricky (higher
verbosity can slow down the server significantly and produce Gbs of logs
so on high traffic servers you really cannot leave it on).

> Could this be linked to the fact I compiled Unbound 1.4.6 with ldns
> 1.6.4 (saying that because Paul Wouters opened a thread about possible
> dangerous linking when building RPM - something you've just fixed in
> 1.4.7).

No, unbound mostly uses ldns to read-write DNS to ascii, tests and other
stuff, but not for TCP.

Best regards,
   Wouter
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org/

iEYEARECAAYFAkzg6KsACgkQkDLqNwOhpPhcdwCffWjPkq3v53QanAaA+u5FXdCK
frQAoLGk2ZSAhteQ6CsIiABv5Jc6Cxnw
=NG+2
-----END PGP SIGNATURE-----