Re: [netconf] Private Candidates

Hi,

On Mon, Oct 31, 2022 at 7:00 PM Fengchong (frank) <frank.fengchong=
40huawei.com@dmarc.ietf.org> wrote:

> Hi James,
>
> Thanks for your effort on private candidate, it’s a very important
> feature, and provide a way to detect and solve conflicts.
>
>
>
> I have a suggestion on conflict detection.
>
>
>
> The current solution will check the conflict per node, it’s a good way,
> but more efforts will be payed if we try to implement this solution.
>
> For example, we MUST check all descendent nodes of a specified interface
> to determine whether the conflict occurs when this interface is deleted in
> running datastore.
>
>
>
> We may add a simple solution to detect conflicts, that is checking the
> conflicts globally. For example, we can use commit-id to determine whether
> the conflict occurs.
>
> The commit-id will be changed if the configuration of running datastore.
> When a private candidate datastore is created, the commit-id (e.g. 100) of
> current running datastore will be recorded,
>
> And if another session commits it’s configuration ,the commit-id of
> running datastore will be changed (from 100 to 101). And then this session
> want to commit its configuration, the conflict will be
>
> Detected (for commit-id  of  private candidate is different with the one
> of current running datastore).
>
>
>
> We can provide two ways (global and per node), the implementation can
> choose one way.
>
>
>

I am not going to dive into the solution details.
IMO a clear set of functional requirements is needed first.

The big issues:

1) resource management
     implicit vs. explicit private candidate creation deletion
      (pick 1 or let implementors pick)

2) RFC 7950 compliance
     A "valid" configuration datastore is clearly defined.
     There is no such thing as a "node commit".
     The entire datastore contents need to be valid  (e.g., leafref, must)

3) Backward Compatibility with :candidate capability and datastore locking
     How much? How to do it.

4)  Boundaries of the Standard
     Need to clearly separate functionality from implementation details
     e.g., how to do a private-commit (push) or a private-update (pull)

Andy

*发件人:* netconf [mailto:netconf-bounces@ietf.org] *代表 *Cumming, James (Nokia
> - US)
> *发送时间:* 2022年10月21日 22:17
> *收件人:* netconf@ietf.org
> *主题:* [netconf] Private Candidates
>
>
>
>
>
> Good morning/afternoon/evening,
>
>
>
> On behalf of the authors I would like to draw your attention to this draft.
>
>
>
> It deals with the use of Private Candidates over NETCONF.
>
>
>
> https://datatracker.ietf.org/doc/draft-jgc-netconf-privcand/
>
>
>
> We hope to present this for the first time in London.
>
>
>
> Kind regards.
>
>
>
>
> _______________________________________________
> netconf mailing list
> netconf@ietf.org
> https://www.ietf.org/mailman/listinfo/netconf
>