Maintained by: NLnet Labs

[Unbound-users] Unbound Slow to Prime

W.C.A. Wijngaards
Tue Aug 9 09:26:02 CEST 2011


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

Hi Alan,

On 08/08/2011 01:55 PM, Alan Gutierrez wrote:
> 
> On Aug 6, 2011, at 7:45 PM, Paul Wouters wrote:
> 
>> On Sat, 6 Aug 2011, Alan Gutierrez wrote:
>>
>>> It used to be the case that I could flush the unbound cache and then resolve a name immediately. Now there is a long delay before I start to get results. It is especially bad on Fedora 15 running in VirtualBox on OS X Lion, but I'm also seeing it slow down on Fedora 15 at EC2. I updated root.hints, but that didn't seem to help. I can't make much sense of the debugging logging output. A couple weeks ago, flushing cache or restarting, you might notice a light delay on the first lookup, but now it takes a few minutes to get results other than timeouts.
>>>
>>> server:
>>> 	verbosity: 2
>>> 	num-threads: 1
>>> 	interface: 127.0.0.1
>>> 	do-ip4: yes
>>> 	do-ip6: no
>>> 	do-udp: yes
>>> 	do-tcp: no
>>
>> Why do you have do-tcp set to no? That will cause problems on large dnssec replies that get
>> truncated.

Yes that would be good to enable.

>>> 	use-caps-for-id: yes
>>
>> You can try disabling this, some servers don't do 0x20 properly and it might be causing delays.
> 
> I tried disabling this, but it was working fine before about two weeks ago. Now I'm finding a slow start everywhere. I've created a gist, with the output of the unbound logs, with verbose at 9. It goes from startup to the successful resolution of "dig NS .". The first time I invoke dig, it times out. Then I wait a while and invoke dig again, for a successful resolution.
> 
> https://gist.github.com/1131622
> 
> There is a lot there. I'm not sure what to look for. If there is anything I can do to make it easier to get feedback from the group, let me know.

What seems to be happening is this:
Aug 8 11:20:37 celje unbound: [6762:0] debug: timeout udp

You have only a DLV key.  The root.key file you read with the statement
trusted-keys-file (which is for BIND compatibility).  I would expect you
to use auto-trust-anchor-file: "....root.key". (and use unbound-anchor
to make that file in the rc/init script).  There are no trust anchors in
your root.key file?

Thus, because you do not have a root key.  But a DLV key.  And 0x20 and
harden-referral-path.  For me that creates a queue of 164 on the first
prime.  And it takes a long time to prime for me too.  (perhaps
something like an adsl modem is dropping connections because there are
too many?)  If I add a normal root trust anchor, the first prime
succeeds (in a second).

So, your combination of options creates a lot of work.  You can wait for
that.  Or you can add a root trust anchor, which I think you intend to?

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