[Pdns-users] Intermittent recursion failure due to timeout

[bert hubert]

Hi Brian,

On Tue, Jan 23, 2018 at 02:00:58PM -0700, Brian T wrote:
> My recursor config has 'network-timeout' set to 5500ms (instead of the
> default 1500ms), so I can understand the "timeout resolving after
> 5535.35msec" message, but I was surprised by the timestamps of the messages
> all being the same.  Are these messages flushed to syslog all at once or is
> the timeout really happening immediately?

This is exactly why we note the 'observed' time passed in the log line. 
When running with trace=servfail, we indeed buffer the log with one single
timestamp.

> leading up to the timeout that look suspicious?  This has been happening
> about 10-15 times per day and started happening on Jan 18th.  I see similar
> timeouts to 3 other sites, but this one most frequently.

91.189.95.68 is the host causing timeouts.  It answers very reliably for me
from here, but it is close to me (London).  It may be that this address
isn't very well served to you somehow.

> Lastly, is there any way to retry here?  I'd rather lower the network
> timeout and attempt a retry if possible.

"Jan 23 05:47:55 n422 pdns_recursor[8739]: [414333]                                                                                                          
nova.clouds.archive.ubuntu.com: Cache consultations done, have 1 NS to                                                                                      
contact"

Usually domains have at least two nameservers.  That generates two attempts,
which should really be enough.

For domains with only a single nameserver with a single IP address, we might
indeed consider trying twice. But there is no way to make that happen right
now from the configuration.

For now, I think you are mostly observing a somewhat unreliably hosted
domain name. 

	Bert