For future reference, it appears that Gnodals don't always behave right when port-channel/LACP state changes occur. TL;DR: do 'shut / no shut' on the Gnodal Ethernet interfaces and possibly also the port-channel interface, and sanity will return.
Setup: I have a few Juniper EX4200 in a virtual chassis setup, basically all stacked together and managed like one switch. I originally added a 2x10G LACP link between my Gnodal 0018 and two Junipers, and it worked fine.
Last week, I added a couple more ports to the LACP link, and discovered that (a) both switches thought that all links were up and negotiating correctly, and (b) I had lost the ability to talk to ~half of the servers across the link. I ended up shutting down the new links (in software, on the Juniper end) and everything returned to normal.
Tonight, I rebooted the Juniper switch with the turned-down LACP links, and *all* traffic across the LACP link stopped flowing. Again, both ends showed LACP negotiating correctly, with 2 active links. I deleted all of the non-active LACP configs from both ends; still no traffic. What fixed it was shutting down both of the underlying Ethernet interfaces on the Gnodal and then turning them back up. As soon as I did that, the link popped to life. Once that was done, I was able to re-add the additional pair of interfaces and everything seems to be working properly, with more bandwidth.
This looks like a pretty clear Gnodal bug to me, and obviously it's not going to get fixed in a future software update. Hopefully this won't happen every time I have to reboot a switch.
