On 5/3/16 10:50, Benoit Claise wrote:
I am of the opinion that I2Rs-relevant preemptive changes to the
running datastore should behave like an I2RS Client. I know this
isn't the case now.
"I'll create a new management interface, and since all changes, ever,
will come through my new management interface, I won't have any problem!"
You and I heard that story a couple of time in our OPS careers, right
:-) Did that behavior ever end up as expected?
Of course. I do everything via SNMP :-).
The current traceability model requires logging to be done by the I2RS
Agent. If an I2RS-relevant change to the running datastore is passed
to the Agent, then the Agent MUST log it. In this manner I would say
we should add some text to the description of Client Address that
makes it clear that the Client can be the network element itself.
What about:
"...an IPv4 or IPv6 address. In the case where changes to the network
element's running datastore preempt state set by the I2RS Agent, the
Client Address will be an address of the network element itself."
That solves one aspect of the problem.
The fact is that we have multiple management interfaces to manage what
I2RS care about. At the very minimum two: CLI and the I2RS agent.
The I2RS agent will log what it knows. Obviously, it doesn't know what
it doesn't know. Assuming that someone uses the CLI, the log will be
useless.
At the very minimum, we need a big capital letter warning: your log
might not be as useful as they pretend to be.
Is that warning really required? Architecturally, the I2RS Agent is
clearly in the path. If the I2RS Agent knows about a change (either via
a northbound Client or the network element preempting state), then it
logs it. If a network element chooses not to notify the Agent, then
like you say, we can't log what we don't know.
If you feel strongly that this be explicitly called out, I think the
best section might be "Trace Log Creation". It could be stated that
preemptive changes outside the I2RS protocol in which the network
element does not notify the Agent of state will not be logged.
Joe
_______________________________________________
i2rs mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/i2rs
