Wednesday, May 22, 2019

PAGP etherchannel caused stack-mgr to freak out.

Maybe there is an obvious gotcha, but so far I haven't found it.

The story so far...

I was creating a LAGG between a 9300 two member stack and a 3650 two member stack. I ran out of 10G interfaces on the 3650 end, so I thought I'd toss a 1G interface into the LAGG group as sort of a standby (I know you can't mix 10G and 1G interfaces since they run at different speeds).

If I configure the member interfaces with `channel-group 1 mode active` they come up as LACP, as expected, no problem. The problem is that sometimes the 1G would be `(P) - bundled in port-channel` while the 10G would be `(s) - suspended`, or vice versa.

Then I got the bright idea of giving 10G a higher priority... somehow.

PAGP's "port-priority" sounds promising. So I set the 10G interface to `pagp port-priority 1`, the 1G interface to `pagp port-priority 255`. Thinking everything looks ok, I `no shut` the 1G interface, was about to `no shut` the 10G, when the 3850 seems to go offline (console dead, can't ping it). I eventually get on the serial console, and before doing anything else, I `shut` the 1G interface, since that was the last change. Right after doing that, I see a bunch of stack-mgr message scrolling by... WTF?

Both stacks are using stacking cables, not virtual stacking. So i'm at a loss.

Is there a footnote on PAGP port-priority on stacked switches I'm missing?



No comments:

Post a Comment