Re: [I2rs-proto-dt] couple YANG details

["Susan Hares" <[e-mail]>]

Andy:  

 

I’d also like to understand the concepts of panes of glass versus sibling databases. 

 

Sue 

 

From: I2rs-proto-dt [mailto:[e-mail]] On Behalf Of Andy Bierman
Sent: Wednesday, October 28, 2015 12:21 PM
To: Susan Hares
Cc: Jeffrey Haas; [e-mail]
Subject: Re: [I2rs-proto-dt] couple YANG details

 

Hi,

 

I will be there Sat. night, so anytime from Sunday to Friday is OK

(except WG meetings I have to attend).

 

The YANG validation should fall out for free if the architecture

is good enough.  Something as simple as YANG "max-entries"

should be obvious, but that does not seem to be the case.

 

There was a slide from Jeff way back that had "running" and

"ephemeral" as "sibling datastores".  The intended config

was derived by the I2RS agent somehow.   This is how Phil described

the Junos implementation to me.  The router picks data from the

dynamic datastore to activate.  It is not a direct API to operational state.

 

Understanding the differences between panes of glass and sibling datastores

might be useful. Something to talk about next week.

 

 

Andy

 

 

 

On Wed, Oct 28, 2015 at 6:02 AM, Susan Hares <[e-mail]> wrote:

Andy: 

 

The two options were a choice – not both.  I’m not sure how to take the document toward this major change. I think I need in person time at IETF to talk about this. 

 

When do you arrive at Yokohama? 

 

Sue 

 

From: I2rs-proto-dt [mailto:[e-mail]] On Behalf Of Andy Bierman
Sent: Monday, October 26, 2015 4:18 PM
To: Susan Hares
Cc: Jeffrey Haas; [e-mail]
Subject: Re: [I2rs-proto-dt] couple YANG details

 

 

 

On Fri, Oct 23, 2015 at 1:47 PM, Susan Hares <[e-mail]> wrote:

Andy: 

 

Two options: 

1)      Ephemeral obeys the overall configuration, 

2)      Ephemeral sets its own ephemeral limit. 

 

 

It is not that clear what it means for both (1) and (2) to be true.

 

I am ready to give up on panes of glass.

Nobody is clear on the details of client-facing integration of running

and ephemeral datastores.

 

Instead, I think the ephemeral datastore should be completely separate

from the running datastore.  At boot-time, it is completely empty.

 

To override "next-hop",  the I2RS client has to provide all the ancestors,

and (at least) ancestor key leafs.  These data nodes will not be created automatically

or mirror running config automatically, or do anything at all automatically,

wrt/ the running datastore.

 

The router is responsible for reconciling the running and ephemeral datastores

to produce the current intended config.

 

A collision can only occur in the ephemeral datastore between the proposed edit

and the existing ephemeral datastore contents.

 

Running datastore validation and operation is completely untouched

and unaffected by the ephemeral datastore.   

 

Ephemeral datastore validation and operation is completely untouched

and unaffected by the running datastore.   

 

The client will not be able to derive the intended configuration by retrieving

both datastores and combining them. A special operation is probably

needed for that.

 

Validation on the ephemeral datastore is not really required, although

any deviation from the running datastore (especially pick and choose statements)

is likely to cause user astonishment.

 

IMO it is least astonishing to say the ephemeral datastore is not validated

at all, but the intended configuration in use by the agent SHOULD be valid

at all times.

 

 

Andy

 

 

 

 

If you do #2, then are you suggesting we put a limit in the original data model. 

 

Sue 

 

From: I2rs-proto-dt [mailto:[e-mail]] On Behalf Of Andy Bierman
Sent: Friday, October 23, 2015 11:29 AM
To: Jeffrey Haas
Cc: [e-mail]; Susan Hares
Subject: Re: [I2rs-proto-dt] couple YANG details

 

 

 

On Fri, Oct 23, 2015 at 8:19 AM, Jeffrey Haas <[e-mail]> wrote:

Andy,

On Wed, Oct 21, 2015 at 02:49:12PM -0700, Andy Bierman wrote:
> We need to figure out for each YANG constraint:
>  1) does it apply at all?
>  2) is it MUST, SHOULD, or MAY enforce?
>  3) Does the constraint apply the same in running vs. ephemeral?
>  4) Does the constraint apply to the combined panes of glass or
>      each pane independently?

I'm going to avoid answering your question to let you decide what this
functional requirement means:
Ephemeral nodes SHOULD NOT introduce their own constraints.
The presence of ephemeral nodes does not trump validation or constraints for
persistent state.

 

So if the YANG list has "max-elements 5",

and the config has 5 entries.  What happens when

5 or 10 entries are added in the ephemeral datastore?

 

 

-- Jeff

 

Andy