Third attempt, this time only attached as a plain text file.
Rob
> -----Original Message-----
> From: Rob Wilton (rwilton) <[email protected]>
> Sent: 02 November 2020 19:42
> To: [email protected]; [email protected]
> Subject: AD review of draft-ietf-netmod-nmda-diff-07 (part 2)
>
> Part 2, unfortunately my original email (attached) was truncated somewhere
> along the way ...
>
> The remainder of my comments are:
>
>
> Suggest adding an example ", e.g., <operational>."
>
> o filter-spec: This is a choice between different filter
> constructs
> to identify the portions of the datastore to be retrieved. It
> acts as a node selector that specifies which data nodes are
> within
> the scope of the comparison and which nodes are outside the
> scope
Hi,
Here is my AD review for draft-ietf-netmod-nmda-diff-07. Apologies for the
delay.
Thank you for writing this document, I think that it is useful, and looks like
it is in good shape.
Main comments:
1. Should there be any text about how to find out what datastores are supported
by a device? E.g., pointing them to either YANG library, or protocol specific
mechanisms in the case of RESTCONF.
2. It might be helpful to add a comment about potential issues that could arise
by comparing <running> to <operational>, i.e., additional differences could be
reported due to inactive configuration and template processing between
<running> and <operational>.
3. I would prefer if 'exclude=origin' was in the reverse sense and perhaps
called 'report-origin' instead. With the reverse sense it seems to be safer if
new datastores are defined, where otherwise the behaviour could end being under
specified.
4. Should there be an option to filter on origin metadata? E.g., only include
values that come from intended. Otherwise, things like IP addresses learned
from DHCP may always turn up as differences.
5. I'm not that keen on the "Possible Future Extensions" section of an RFC.
Personally, I would prefer that this section is deleted, but if you wish to
retain it, then please can you move it to an appendix.
I've also included some minor comments inline below, and some nits at the end:
Abstract
This document defines an RPC operation to compare management
datastores that comply with the NMDA architecture.
The abstract is perhaps somewhat terse. Perhaps:
This document defines a YANG RPC operation to compare the
contents of network management datastores that comply with
the NMDA architecture and return the differences in the
YANG-Patch format.
1. Introduction
The revised Network Management Datastore Architecture (NMDA)
[RFC8342] introduces a set of new datastores that each hold YANG-
defined data [RFC7950] and represent a different "viewpoint" on the
data that is maintained by a server. New YANG datastores that are
introduced include <intended>, which contains validated configuration
data that a client application intends to be in effect, and
<operational>, which contains at least conceptually operational state
data (such as statistics) as well as configuration data that is
actually in effect.
I would suggest deleting "at least conceptually", since the <operational>
datastore does contain all operational state, but it may be implemented as a
virtual construct that spans multiple nodes (e.g., linecards) and processes.
NMDA introduces in effect a concept of "lifecycle" for management
data, allowing to clearly distinguish between data that is part of a
configuration that was supplied by a user, configuration data that
has actually been successfully applied and that is part of the
operational state, and overall operational state that includes both
applied configuration data as well as status and statistics.
"allowing to clearly distinguish" => distinguishing"
"status and statistics" => "status information and statistics"
As a result, data from the same management model can be reflected in
multiple datastores. Clients need to specify the target datastore to
be specific about which viewpoint of the data they want to access.
This way, an application can differentiate whether they are (for
example) interested in the configuration that has been applied and is
actually in effect, or in the configuration that was supplied by a
client and that is supposed to be in effect.
Perhaps reword the last sentence to match the logical data flow in the server:
For example, a client application can differentiate whether they are
interested in the configuration supplied to a server and that is
supposed to be in effect, or the configuration that has been applied and is
actually in effect on the server.
When configuration that is in effect is different from configuration
that was applied, many issues can result. It becomes more difficult
to operate the network properly due to limited visibility of actual
status which makes it more difficult to analyze and understand what
is going on in the network. Services may be negatively affected (for
example, breaking a service instance resulting in service is not
properly delivered to a customer) and network resources be
misallocated.
Perhaps change "actual status" to "actual operational status".
I also suggest changing the last sentence to:
Services may be negatively affected (e.g., degrading or breaking a customer
service) or network resources may be misallocated.
3. Definitions:
It should probably define that <intended>, <operational>, (and perhaps
<running>) are used to indicate names of datastores.
It should also explain that <compare> is used as the name of a YANG RPC.
4. Data Model Overview
At the core of the solution is a new management operation, <compare>,
that allows to compare two datastores for the same data.
Suggest rewording this first sentence to:
The core of the solution is a new management operation, <compare>,
that compares the data tree contents of two datastores.
o target: The target identifies the datastore to compare against the
source.
Suggest adding an example ", e.g., <operational>."
o filter-spec: This is a choice between different filter constructs
to identify the portions of the datastore to be retrieved. It
acts as a node selector that specifies which data nodes are within
the scope of the comparison and which nodes are outside the scope.
This allows a comparison operation to be applied only to a
specific portion of the datastore that is of interest, such as a
particular subtree. (The filter dow not contain expressions that
would match values data nodes, as this is not required by most use
cases and would complicate the scheme, from implementation to
dealing with race conditions.)
Perhaps "parts/part" rather than "portions/portion".
Suggest rewording the last sentence to:
Note, the filter does not allow expressions that match against data node
values since that may incur implementation difficulties and is not required for
normal use cases.
o ... When the target datastore is <operational>, "origin"
metadata is included as part of the patch. Including origin
metadata can help in some cases explain the cause of a difference,
for example when a data node is part of <intended> but the origin
of the same data node in <operational> is reported as "system".
I think that this test needs to refer back to the 'exclude-origin' or
'report-origin' options since the origin metadata isn't always included.
5. YANG Data Model
The key words 'MUST', 'MUST NOT', 'REQUIRED', 'SHALL', 'SHALL
NOT', 'SHOULD', 'SHOULD NOT', 'RECOMMENDED', 'NOT RECOMMENDED',
'MAY', and 'OPTIONAL' in this document are to be interpreted as
described in BCP 14 (RFC 2119) (RFC 8174) when, and only when,
they appear in all capitals, as shown here.
I couldn't see that RFC 2119 language is actually used in the YANG module, so
perhaps this can be deleted?
rpc compare {
description
"NMDA compare operation.";
Perhaps "NMDA datastore compare operation."
anydata source-value {
when "../operation = 'delete'"
+ "or ../operation = 'merge'"
+ "or ../operation = 'move'"
+ "or ../operation = 'replace'"
+ "or ../operation = 'remove'";
description
"The anydata 'value' is only used for 'delete',
'move', 'merge', 'replace', and 'remove'
operations.";
I'm not convinced how useful the when statement really is in this case, since
'source-value' isn't marked as mandatory, a server is allowed to omit it when
it doesn't apply anyway.
6. Example
<operational> does not contain object "description" that is contained
in <intended>. Another object, "enabled", has differences in values,
being "true" in <operational> and "false" in <intended>. A third
object, "name", is the same in both cases. The origin of the objects
in <operational> is "learned", which may help explain the
discrepancies.
I think that we should probably refrain from calling them objects, perhaps
"leaf instance" would be better? Rather than "differences in values", perhaps
"different values".
E.g.,
<operational> does not contain an instance for leaf "description" that
is contained in <intended>. Another leaf, "enabled", has different
values in the two datastores, being "true" in <operational> and
"false" in <intended>. A third leaf, "name", has the same instance
value in both datastores. The origin of the leaf instances in
<operational> is "learned", which may help explain the discrepancies.
//OPERATIONAL
<interfaces
xmlns="urn:ietf:params:xml:ns:yang:ietf-interfaces"
xmlns:or="urn:ietf:params:xml:ns:yang:ietf-origin">
<interface or:origin="or:learned">
<name>eth0</name>
<enabled>true</enabled>
</interface>
</interfaces>
There is an extra line, and dodgy indentation for "</interface>".
Nits:
"possibly for" => "possible for"
"reference for the" => "the reference data tree for the"
"is basis" => "is the basis"
Thanks,
Rob
_______________________________________________
netmod mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/netmod
