[c-nsp] ME3600X 15.2S memory leak

[Mark Tinka]

On 27/Feb/16 11:16, Adam Vitkovsky wrote:

> Hmmm looks like the symptoms are the same just the cause seem to be unknown.

Not exactly.

With the (r)LFA issue, you could still reach the box remotely and either
reload it or reset the FIB. And remember this issue had to do with the
fact the (r)LFA on the ME3600X is not supported with Implicit-Null MPLS
labels.

With this particular case, you can neither reach nor transit the box,
even though all looks well. Nothing in the logs complaining of anything,
besides BFD and IS-IS reporting an outage.

> Are you using MPLS-TE, BGP-PIC(Core/Edge), load-sharing/multipath?

No MPLS-TE, no BGP-PIC.

Load sharing is standard through IS-IS, due to there being multiple exit
paths from the box on to the fibre ring.

> Maybe running out of HW resources?

I considered that, but even with BGP which takes up the most amount of
RAM, the box still has over 340MB of memory left. This is plenty, and is
not leaking. If I look at graphs over the course of a year, creep has
been linear with BGP table growth, and shows no signs of a memory leak.

I also considered an issue where someone once said they were running out
labels (or label space in the FIB, can't quite recall clearly), but that
does not appear to be the case here. Besides, we have several other
switches deployed exactly the same not exhibiting this issue.

We're going to move to 15.5(3)S2 tonight and take it from there.

Mark.