[c-nsp] CSCti84025

[Phil Mayers]

On 05/16/2011 05:00 PM, Tim Durack wrote:
> Anyone run into CSCti84025 on a C6K, VS-S720-3C, 12.2(33)SXI3?
>
>
> VRFs hardware re-mapping causing MLS/CEF inconsistencies

We've had something very like this on SXI3 - routing changes trigger 
MPLS label mis-programming resulting in CPU punt and high packet loss on 
affected LSPs. TAC logged in and did some poking at the TCAM and fixed 
it, then I reloaded the box a week or so later.

We're on SXI5 and these seem to have gone away.

> On a Catalyst 6500 series switches, connectivity in VRFs may be affected after
> a large reconvergence. Another symptom of the issue is the presence of invalid
> mls mpls entries :
>
> Router#show mls cef mpls | i INVALID
> 583 731 INVALID punt
> 1377 675 INVALID punt
> 1379 9742 INVALID punt
>
> Conditions:
>
> This is observed on a PE running a fair amount of ebgp sessions between its
> own VRFs.

That wasn't the trigger for us (although TAC was unable to determine 
exactly what was) hence "very like" rather than "exactly" this.

> Interested in knowing what it "feels" like. We are seeing these
> invalid entries, and have been experiencing random/cyclic packet loss.
> I'm guessing invalid labels punting to control plane with CoPP
> configured might result in seemingly random packet loss.

That's exactly what we experienced.

It seems to "go away" for a while after reload i.e. the TCAM 
mis-programming doesn't seem to happen until the box has been up a 
while. But that was very much an impression/gut feeling, no hard evidence.