Shutting Down Does Not Terminate Instance

[jrtc27]

Steps:

1. Launch SRI-Cambridge | FreeBSD | bluespec_p2 instance
2. Connect via SSH
3. Run su -
4. Run halt -p

This will cause FreeBSD to shut down and, ultimately, power off by telling the Connectal host-side software to stop, so fett.py sees the process as having exited.

Expected outcome:

Instance is terminated (after the same delay as it normally takes to manually terminate an instance).

Actual outcome:

Instance is still regarded as running.

Whilst liveness checks would catch this, it would be good to have the tools gracefully handle this non-exceptional case as if it were a normal termination.

[jrtc27]

(for reference, this was I-0BC3F5404037B3D0F)

[kiniry]

We have other discussions going about watchdogs on instances. I'll tag @kurthopfer on this so that this can be cross-referenced and we can possibly chase this up as a possible hot-fix.

[kurthopfer]

Yes - as it stands, neither the portal nor fett are actively monitoring target after successful launch and bootstrapping.

[mattlebeau-galois]

Implementation would be in FETT-Target to spit back a msg that the instance went down. Effort assessment in FETT-Target is next.

[andrew-bivin]

Adding to this, we need to verify that the instance is actually terminated (via EC2) when a shut down/halt occurs. We will take a look at this in FETT-Target.