Is asychronous DNS lookup worth keeping at all?

Chris Johns chrisj at ntpsec.org
Thu Dec 3 23:36:33 UTC 2015


On 12/02/15 08:13, Mark Atwood wrote:
> Ok, I think our voices of wisdom are right.  Don't rip it out just yet,
> let Chris weigh in, and then when the time is right, replace it.

I think we need to have asynchronous support for any operation that runs 
in a synchronous manner to the main loop processing and could.

The basis of this assertion is the effect these types of operations have 
on the main processing loop. My concern is jitter effects on the time 
and flow on effects based on slow DNS or failing DNS responses causing a 
back log of events or network data. This is now in an area I am not 
familiar with in NTP, ie it's protocols and time management, and I am 
happy with being told otherwise.

In relation to DNS it is configuration changes, errors and delays which 
seem to be the issue. We have DHCP lease changes or other interface 
related changes to manage. There is a 1 second (or so) poll of the DNS 
configuration and a DNS back off which is present in the DNS 
asynchronous processing. They must be present for a reason.

Finally looking at this from the other direction, how do we make a case 
that removing the support will not effect all users from big to small, 
from complex to simple? Would it be more painful to incrementally add it 
back in based on freedback from important users?

Chris


More information about the devel mailing list