Re: [netmod] I-D Action: draft-wwx-netmod-event-yang-04.txt

[Qin Wu <[e-mail]>]

Hi, Igor

发件人: Igor Bryskin [mailto:[e-mail]]
发送时间: 2019年11月5日 22:02
收件人: [e-mail]; [e-mail]; Lou Berger <[e-mail]>; Qin Wu <[e-mail]>
抄送: [e-mail]
主题: Re: I-D Action: draft-wwx-netmod-event-yang-04.txt

Hi Qin,

Good discussion. Please, see in-line.

Igor

On Tuesday, November 5, 2019, 3:09:53 AM EST, Qin Wu <[e-mail]<mailto:[e-mail]>> wrote:



Hi, Igor:

发件人: Igor Bryskin [mailto:[e-mail]]
发送时间: 2019年11月5日 3:06
收件人: Qin Wu <[e-mail]<mailto:[e-mail]>>; [e-mail]<mailto:[e-mail]>; [e-mail]<mailto:[e-mail]>; Lou Berger <[e-mail]<mailto:[e-mail]>>
抄送: [e-mail]<mailto:[e-mail]>
主题: Re: I-D Action: draft-wwx-netmod-event-yang-04.txt



Hi Qin,



Thanks for the effort.



My general question is  what is the ultimate objective/ambition of this work? Is it



1. Modeling the imperative policy style network automation as stipulated by the SUPA framework

          or

2. Event scoping of PUSH machinery



If 2. is the case, it would certainly make sense and might prove useful for many use cases. However, in this case you have neither reason nor right to use well understood abbreviation ECA, nor to refer to the SUPA documents. Neither it would make any sense to merge our contributions IMHO



If 1. is the case, then

here is our comments/suggestions as to how the work should in our opinion evolve going forward:



[Qin]:Good question, I think we mostly focus on modelling imperative policy in which ECA is a typical example of ECA model. In addition, we see Event scoping of PUSH machinery is a special case of ECA without Action to be specified.
We clarified the relation with YANG Push, we think YAN PUSH model can be augmented with some grouping defined in ECA model. So ECA model doesn’t need to tie with YANG Push.


IB>> True, but if we model generic ECAs, things like  PUSH Event scoping, Smart Filters, etc. come naturally as trivial private cases. There is no need to focus on them.

 [Qin]: Exactly.

1.The Expression clause in an ECA could be very complex and hence requires a complex syntax to articulate. To address this in our contribution (https://datatracker.ietf.org/doc/draft-bryskin-netconf-automation-yang/) we proposed two methods:

a) When configuring Condition using XPath expression string. This allows expressing Conditions of arbitrary complexity, but does require servers to (sufficiently) support XPath language;

[Qin]:XPATH expression is supported in model proposed in draft-wwx, it is modelled as one of member of union, i.e., instance-identifier, in addition, we support model three other member types

Type yang:object-identifier;

Type yang:uuid;
Type string

IB>> Good. Please, note that we were told on many occasions that because of potentiality very complex syntax of the ECA Condition clause, the XPath expression string is realistically the only choice, all alternatives are introduced for model completeness more than anything else - too cumbersome to be useful.

[Qin]: Tend to agree, this is complexity we can consider to get rid of.

b) For the case of simpler servers we defined elementary logical primitives that could be used in building bottom up in hierarchical manner complex logical expressions



[Qin]: I believe you are talking about Condition Expression, which is corresponding to ietf-trigger.yang defined in draft-wwx-netmod-event-yang-04. We model them as three trigger conditions

1.       An existence test monitors and manages the absence, presence, and change of a data object

2.       A Boolean test compares the value of the monitored object with the reference value and takes action according to the comparison result.

3.       A Threshold trigger condition regularly compares compares the value of the monitored object with the threshold values.
In each trigger condition, we will break down them into policy variable and policy value based on RFC3460, policy variable is renamed as target, policy value is renamed as value in proposed ECA model

IB>> IMHO this is not  sufficient, not even close.

[Qin]: Actually it can be extended, the essence of trigger condition is <target><relation><arg> which is similar to <arg1><relation><arg2> in draft-bryskin
would you like to provide an example which can not be expressed by these trigger conditions?
I am open to the better design choice.


I feel you change the meaning of policy variable, since in bryskin’s draft, policy variable is described as an output parameter of an RPC which is not consistent with the definition in RFC3460, in my opinion.

IB>> No, I have not. In our definition a PV is a variable where an ECA thread stores results of computations and output of algorithms/RPCs, so that the results could be used within a single thread or between multiple threads of the same or different ECAs, could provide input for automatic re-configurations and RPCs, could be used in Condition evaluations, could be exposed directly to the client via notifications, etc. In short, this is the place where ECAs store and accumulate the results of their work

 [Qin]: I thought PV is corresponding to target defined in draft-wwx, or data object to be monitored, we will reflect the change of data object or target in the action definition of ECA model.

I see the only difference on model design, is target or policy variable is separated from ietf-event, or part of ietf-event. If the reason why we should have a separate policy-variable is we should store state on policy-variable or target, I think put policy-variable into ietf-event, you still can store state related to policy-variable in ietf-event, No?

2. Your model seems to suggest for ECA Action  not much more than PUSHing a notification (triggered by a certain event and satisfying the configured condition) to the client with the hope that the client will subsequently request some device/network re-configurations ro react to the event.



[Qin]:Igor, the ECA action proposed in the model of draft-wwx-netmod-event-yang-04 can do more than PUSHing a notification, it have supported the following capabilities:
1)Configuration data object reconfiguration

IB>> Good, but keep in mind that the parameters of such configurations could not be limited to values specified by the client at the time of ECA configuration ( such values we call Policy Constants (PCs)). It is imperative to allow for the results of the ECA thread computations to be also used as values to configure (i.e. PVs along with PCs)

[Qin]: Yes, I have been aware that Policy constant is different from Policy variable, Are both pointing to the same monitored data objects?
I think whether it is policy constant or policy variable, it should be set or configured only when certain conditions hold.
I am wondering where do you store the results of computations(e.g., mean/variance) or some tempo value of monitored data object?
You use policy variable itself or you have somewhere else to store these tempo results?

2) ECA Log report Notification

3)Invoke another Event

IB>> 2) and 3) are (albeit important) auxiliary  functions, rather than ECA Actions, strictly speaking



[Qin]: Agree.

It can be extended to support more advanced features if needed.



There are situations, however, when the said re-configurations must be applied immediately after the event detection with no time to loose on network- client communications. Furthermore, there are cases when the necessary re-configurations are known a priory (at the time of the ECA configuration), and the client may want to pre-configure them along with configuring  the ECA's Event and Condition, and then rely on what we call close loop network automation, rather than to be involved in device/network micro management in real time. To this end our contribution suggests the flowing ECA Action configuration options:

a) Network re-configuration (in the form of per-configured Netconf edit config statements); [Qin]: We support this.

b) PUSHing notifications to the client (the same as you suggest) [Qin]: Correct.

c) Enabling/disabling notification streams (pre-configured as PUSH subscriptions); [Qin]: Do you propose to allow netconf server send notif to the client and instruct client to enable or disable notification stream or the network server can enable or disable some

event stream and inform the client the result?

d) Invoking local network intelligence (configured as YANG RPCs defined in supported by the server YANG models). For example, calling local TE path computation (defined as Path Computation RPC by the te-tunnel  or Path Computation model) could be configured within ECA as Action in order to discover more optimal path for a TE tunnel after the configured Event is fired.
[Qin]: Usually the RPC is sent from NETCONF client to NETCONF server ,do you propose the other way around and allow the netconf server send RPC request to the NETCONF client? I am not sure we can do this

IB>> No, this is about instructing the serer to invoke  its local intelligence with the identity and input/output of which  articulated by the client (as ECA Action) in the form of an RPC defined in a YANG model supported by the server. Think about it this way: when the client per-configures an automatic re-configuration, it does so in the form of edit-config NETCONF command, that is, in the form of a native NETCONF RPC. We simply extend this to allow for specifying RPCs defined by YANG models (e.g. Path Computation RPC).
[Qin]:

I thought you describe the RPC request can be sent from the managed device to the management system when certain condition satisfies. I think we could refer to the ietf-lmap-report
example in RFC8194. In this example, RPC did can be sent from measurement agent to collector or from netconf server to netconf client.
So RPC you proposed is pushed down to the managed device from the management system and invoked only when certain condition hold in the managed device.
For content-moving, I am wondering why not resue edit-config operation?
In addition, when we talk about how to use ECA model, are we focusing  using ECA model in the external interface between NMS and router or are you focusing on using ECA model as internal script to manipulate service logic?

IB>> The latter. This is what pushing (imperative or declarative) policies down to the network server usually means.

 [Qin]: I think both are needed to provided event driven network management, first, the management system put down ECA policy to the managed device using NETCONF interface, secondly, ECA script is generated from ECA policy in the managed device.

3. Evaluation of ECA Conditions, as well as input to ECA Actions may require not just instantaneous network states, but also accumulation/computation of thereof over periods of time (e.g. min/max/mean leaf values, history data, threshold overstep counters, results of various functions/computations/algorithms performed on network states over time, etc.) Hence there is a need for storage of intermediate results of such computations. Our contribution introduces such storage in the form of Policy Variables (PVs). PVs could be part of Condition expressions, as well as Action inputs along with instant network states. PVs also could appear in notifications PUSHed to the client.





[Qin]: If you follows https://tools.ietf.org/html/draft-bwd-netmod-eca-framework-00

You will see we have already considered what state needs to be held, current state and history state, and where this state is held.

Basic state of ECA include: Event Name, event occurrence time, start time, end time, threshold value, etc.
I think it is challenging to store all the states and it adds complexity of server implantation.

IB>> No, I am talking about defining /pushing by the client and executing by the server arbitrary logic in the form of ECAs. This logic, for example, may instruct the server how to recover from various network failures under extreme time constraints. It may also instruct the server how to identify and report "interesting" for the client  events and data, rather than stream raw data  99% of which to be parched, evaluated and discarded as uninteresting

[Qin]: yeah, network failure recovery and filtering unwanted data are two valid use cases we are aiming at also. I am fascinating on function-call you proposed, I am wondering where you store these computation results, why not defined it as mathematics function, just provide input

And then get output, but the problem where to store these output, in addition, how many policy-argument you can support? I seems only two policy-arguments are supported? If we support mathematics function, you can support more than two policy arguments, right?



4. Notifications triggered by ECA s require definition beyond what is defined by PUSH models, so that the notifications could be properly associated by the client with a given execution of a given ECA.  Said definition could be found in https://datatracker.ietf.org/doc/draft-bryskin-netconf-automation-yang/.


[Qin]:Good, we also provide a few use cases in the section 4 of draft-bwd-netmod-eca-framework-00 to discuss how notification is sent to the NMS to trigger another ECA policy execution, we also could support One event invoke another event, depends on use cases,

IB>> Note that ECAs is not about intense communication between the client and the server, rather, quite the opposite - it is about pushing ECAs down to the server and let the server perform the instructed event driven network management

[Qin]: We are aligned on this core case.

The use case we like to aim at is service assurance use case and network troubleshooting self-management use case.



We have more points to discuss, but what is above is a good starting point.



Regards,
Igor (and Xufeng)







On Saturday, November 2, 2019, 10:33:40 AM EDT, Lou Berger <[e-mail]<mailto:[e-mail]>> wrote:





Qin,
    Thanks for the update.

To answer your question as well as respond to the related thread, as
chair, I generally think it best to adopt once there is consensus in the
WG on a direction to take with respect to the topic covered by a draft.
That is not to say that a fully formed or documented solution is
required at adoption but that if there are several different approaches
available, that the adopted work reflects the direction that the WG will
pursue.

In this case, the current rev is certainly a step in that direction, but
the WG still as two different basic approaches available to it in this
draft and draft-bryskin-netconf-automation-yang.  I personally always
prefer it when individual draft authors can find common ground and come
to the WG with a single (unified) proposal rather than ask the working
group to choose one over the other.  I'm not sure who among the authors
will be in Singapore, but perhaps the authors can take the opportunity
to meet to discuss the possibly of such a unified proposal as well
report back to the working group on their progress/status.  Time
permitting, we should at least hear a summary of each approach so that
if a unified approach is not proposed that the WG is better informed on
the proposals.

Cheers,
Lou

On 11/1/19 11:02 PM, Qin Wu wrote:
> v-04 is posted to address chairs' comments,
> https://datatracker.ietf.org/doc/html/draft-wwx-netmod-event-yang-04
> the main changes include:
>    o  Add text in introduction section to clarify the usage examples of
>      ECA policy
>    o  Update objective section to align with use cases.
>    o  Clarify the relationship between target and policy variable.
>    o  Change variation trigger condition back into threshold trigger
>      condition and clarify the usage of three trigger conditions.
>    o  Remove Event MIB related section.
>    o  Add new coauthors and contributors.
> Chairs, what is the next step?
>
> -Qin (on behalf of authors)
>
> -----邮件原件-----
> 发件人: I-D-Announce [mailto:[e-mail]<mailto:[e-mail]>] 代表 [e-mail]<mailto:[e-mail]>
> 发送时间: 2019年11月2日 10:57
> 收件人: [e-mail]<mailto:[e-mail]>
> 主题: I-D Action: draft-wwx-netmod-event-yang-04.txt
>
>
> A New Internet-Draft is available from the on-line Internet-Drafts directories.
>
>
>        Title          : A YANG Data model for ECA Policy Management
>        Authors        : Michael Wang
>                          Qin Wu
>                          Chongfeng Xie
>                          Igor Bryskin
>                          Xufeng Liu
>                          Alexander Clemm
>                          Henk Birkholz
>                          Tianran Zhou
>     Filename        : draft-wwx-netmod-event-yang-04.txt
>     Pages          : 32
>     Date            : 2019-11-01
>
> Abstract:
>    RFC8328 defines a policy-based management framework that allow
>    definition of a data model to be used to represent high-level,
>    possibly network-wide policies.  Policy discussed in RFC8328 are
>    classified into imperative policy and declarative policy, ECA policy
>    is an typical example of imperative policy.  This document defines an
>    YANG data model for the ECA policy management.  The ECA policy YANG
>    provides the ability for the network management function (within a
>    controller, an orchestrator, or a network element) to control the
>    configuration and monitor state change on the network element and
>    take simple and instant action when a trigger condition on the system
>    state is met.
>
>
> The IETF datatracker status page for this draft is:
> https://datatracker.ietf.org/doc/draft-wwx-netmod-event-yang/
>
> There are also htmlized versions available at:
> https://tools.ietf.org/html/draft-wwx-netmod-event-yang-04
> https://datatracker.ietf.org/doc/html/draft-wwx-netmod-event-yang-04
>
> A diff from the previous version is available at:
> https://www.ietf.org/rfcdiff?url2=draft-wwx-netmod-event-yang-04
>
>
> Please note that it may take a couple of minutes from the time of submission until the htmlized version and diff are available at tools.ietf.org.
>
> Internet-Drafts are also available by anonymous FTP at:
> ftp://ftp.ietf.org/internet-drafts/
>
> _______________________________________________
> I-D-Announce mailing list
> [e-mail]<mailto:[e-mail]>
> https://www.ietf.org/mailman/listinfo/i-d-announce
> Internet-Draft directories: http://www.ietf.org/shadow.html <http://www.ietf.org/shadow.html%20> or ftp://ftp.ietf.org/ietf/1shadow-sites.txt
>