Re: [netconf] Adoption-suitability for draft-unyte-netconf-distributed-notif

Hi Mahesh,

Please see in line about my thoughts.

Thanks,
Tianran

From: Mahesh Jethanandani [mailto:mjethanandani@gmail.com]
Sent: Tuesday, September 22, 2020 11:10 PM
To: Tianran Zhou <zhoutianran@huawei.com>
Cc: Kent Watsen <kent+ietf@watsen.net>; Andy Bierman <andy@yumaworks.com>; netconf@ietf.org
Subject: Re: [netconf] Adoption-suitability for draft-unyte-netconf-distributed-notif

[As a contributor]

Hi Tianran,

I am confused by the value statement. Maybe you can help clarify.

It would help if we can use a single terminology of publisher/subscriber instead of mixing it with client, which I understand to be the subscriber and server which I understand to be the publisher. Note, if HTTPS is used, the role of client and server would be reversed, so it is best to avoid using those terms.

See inline with [mj].

On Sep 21, 2020, at 7:04 PM, Tianran Zhou <zhoutianran@huawei.com<mailto:zhoutianran@huawei.com>> wrote:

Thanks Andy for your opinion. And thank Kent for your expansion and further question.
The goal is to allow the client to know if it has received all the data pieces. Furthermore, the client know which publisher failed to send data. Then it depend on the client to do the action, e.g., ignore, require the data one more, …
We just use the generator-id to indicate the publisher, as we find it in ietf-netconf-notification-messages, and it’s very convenient for this usage.

[mj] What usage would be that? The usage of generator-id to indicate who is the publisher and thereby have the subscriber request for any missing data? I would like to note that nothing prevents usage of a reliable transport, e.g. HTTPS or TCP, with this draft. Would you still need a generator-id?

One use case is exactly as Kent mentioned “Answering myself here, perhaps it enables two servers on the same IP address to send to the same receiver, as then the receiver doesn’t solely rely on source-IP address (assuming unauthenticated push) to designate the publisher”.
It’s also OK for us to use another indicator if the generator-id is not feasible.

[mj] Precisely. If you use a reliable transport, you would have other means to identify the source of the notifications, in which case you would not need the generator-id.

ztr> Yes. But I just think if we can use the generator-id, we do not need to develop different ways for each transport. So that the procedure could be consistent.

Which goes back to the question of why we need a generator-id. If we do not want data loss, we should be using a reliable transport, in which case we do not need a generator-id. If we can tolerate data loss, e.g. with statistics, we do not care for generator-id because we are not going to ask for missing data. What use case requires us to use unreliable transport and yet need a generator-id?

ztr> Here we want to check the integrity, not the precise packet loss. For reliable transport, the receiver know how many sessions need to establish for one subscription. On the other hand,  if one session get lost, which subscriptions are impacted.
For the unreliable transport, one more case is, if one session is really bad (say 50% packet loss), the subscriber may need to do some action/adjustment. Yes we can tolerate data loss, but I think it should be a threshold.

Thanks.

“what seems to be missing is an ability for the client to determine which generator-id is used by which publisher (e.g., line card). ”
If necessary, maybe we can extend this mapping in some initial exchanges. To be clear, how do you want to name the publisher?

Thanks,
Tianran

From: netconf [mailto:netconf-bounces@ietf.org] On Behalf Of Kent Watsen
Sent: Tuesday, September 22, 2020 6:55 AM
To: Andy Bierman <andy@yumaworks.com<mailto:andy@yumaworks.com>>
Cc: netconf@ietf.org<mailto:netconf@ietf.org>
Subject: Re: [netconf] Adoption-suitability for draft-unyte-netconf-distributed-notif

[As an individual contributor]

The overall approach to the binary push features seems a bit incoherent to me.
I don't see much value in this draft, but no harm either so I do not have an objection
to adoption.  Perhaps there is some debugging value here but since the architecture
really does not define message generators as sub-components of a configured subscription,
a client cannot expect any sort of consistent implementation of this field.

Good point. If I understand you correctly, what seems to be missing is an ability for the client to determine which generator-id is used by which publisher (e.g., line card).  The solution enables the client to partition notifications coming from different publishers, but that is it.  What value does this have to the client, I don’t know.

Answering myself here, perhaps it enables two servers on the same IP address to send to the same receiver, as then the receiver doesn’t solely rely on source-IP address (assuming unauthenticated push) to designate the publisher?

Can the authors help resolve the value-proposition question here?

K.

_______________________________________________
netconf mailing list
netconf@ietf.org<mailto:netconf@ietf.org>
https://www.ietf.org/mailman/listinfo/netconf

Mahesh Jethanandani
mjethanandani@gmail.com<mailto:mjethanandani@gmail.com>