Kb257

Emulab FAQ: Testbed Operations: Problem with events not firing on plab nodes.

> > Which experiment is this?  It turns out that plab450 is the same node
> > Kevin reported problems with.
>
> It's tbres/nstest. I just swapped it in this morning. I'll turn off
> ide-swap so that you can look at it w/o it getting swapped out.

OK, it turns out that the messages complaining about mapping to agents like __plab_XXX are expected. You will find several of these scattered about in most any event scheduler log. Essentially, any time a plab event proxy first comes online, it will broadcast an UPDATE event destined to all schedulers so that it can get itself subscribed to all experiments that have a vnode (sliver) residing on it's respective host. If your experiment isn't one of these, then your experiment's event scheduler will complain thusly. We can squelch these errors pretty easily, but they aren't doing any harm so I'll punt for now.

However, one of the nodes (plab122) you are using is in the same state I found plab450 in: The proxy is connected, but not receiving events from ops. I'm going to hack up a little program to dump the set of subscriptions from an elvind to help track this down further (if someone has already done this, please speak up!).