> On 19 Dec 2016, at 12:31, Jan Lindblad <[email protected]> wrote:
> 
>> But once you only have machines programming the config, then they should be 
>> able to easily construct a single message for each edit of the config (which 
>> can be applied/failed in its entirety). Candidate and locking don't seem to 
>> help here.
>> 
>> Further, if you want to update multiple devices at the same time then you 
>> end up in the realm of distributed transactions which get very complicated 
>> and are hard to get right in a fully robust fashion.
> 
> Harder still without the candidate. Any case where transaction phases or 
> timing is involved would be harder to get right without the candidate and 
> confirmed commit. Being far from perfect, it's the best standardized 
> configuration protocol out there.

Per-user candidate datastores are much easier to deal with, and this is what we 
do in our RESTCONF implementation:

https://github.com/CZ-NIC/jetconf

Lada

> 
> /jan
> 

--
Ladislav Lhotka, CZ.NIC Labs
PGP Key ID: E74E8C0C




_______________________________________________
netmod mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/netmod

Reply via email to