checkin from rogue runners

[jb098]

In the mite/scenario.py, ScenarioManager.checkin_data works perfectly under normal circumstances.

If a runner is hagning around waiting to check back in from a previous test, this causes an exception.

It's not a problem that it causes an exception, it's a problem that it's not clear what has caused it. We shoudld try and gracefully exit and log out what the issue might be to give the user a hint.

[jb098]

@aecay is this still a problem? I haven't put any fixes in for it but I know we've changed a little bit. I tried to fix once but my solution hamstrung the performance and I couldn't work out why.

[aecay]

It still is. My sketch of a fix is this:

• have each controller process give itself an ID (CID, controller ID)
• put the CID on all the messages from controller -> runner
• make the runners include the CID whenever it responds to a message
• make the controller ignore messages with an unknown CID (currently it sometimes throws in this scenario, because it mixes invalid data from an old message into its valid data)

This will have some performance overhead by swelling slightly the controller <-> runner traffic, but I hope it will not be too bad. I think it's a simple change, but I haven't gotten around to implementing it because busy, and it will require careful testing for correctness and (especially) performance impacts.

Do you remember what your earlier attempt at a fix was? If you had a problem here, that could be an indication that it's not as simple as I'm imagining it to be... 😟