[j-nsp] Juniper Resiliency Interface

[Saku Ytti]

https://www.juniper.net/documentation/us/en/software/junos/flow-monitoring/topics/topic-map/resiliency-exception-reporting.html

Anyone running this?

This seems incredibly useful. We've been screenscraping PFE exception
counters for years, but with screenscraping you only get exception
name, packet counter, byte counter, linecard and NPU.
With this feature, you also get an ingress/egress interface and most
importantly, headers.

But even with just screenscraping PFE exceptions they can be immensely
useful when dealing with complex problem that doesn't immediately make
sense, take a peek at your PFE exception and if something has started
incrementing which usually doesn't, it can be a smoking gun that will
expedite troubleshooting massively.

But if you do collect them, you will constantly see exceptions
increasing that really shouldn't be increasing, so you gotta be
comfortable with 'well sometimes rarely we tickle some esoteric corner
cases and lose packets', since there is no point fixing everything,
you only gotta fix things that actually matter.



ASR9k has a constant circular buffer for EZChip exceptions (maybe
lightspeed too, haven't tried). Unfortunately circular buffer was the
wrong implementation choice, simpler and more correct would have been
buffer[exceptionType] = header.
Now when you do have exotic problem, chances are by the time you are
troubleshooting it, the circular buffer is full of exceptions you
expect, and you cannot anymore see the rare issue you are interested
in.



-- 
  ++ytti