Travis shutdown didn't result in failed job #8103

#8049 had a bug that would cause some RPC tests to fail, but travis didn't catch it, because the machine was shutdown(!):

https://travis-ci.org/bitcoin/bitcoin/jobs/129694378

Test 122: PASS [581]
Test 123: PASS [581]
Test 124: PASS [582]
Test 125: PASS [582]
Stopping nodes
Cleaning up
Tests successful

Pass: True, Duration: 37 s
..................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
Broadcast message from root@testing-gce-017e1b04-0cfc-4a39-9cf0-a92f5dae8c5d
    (unknown) at 16:47 ...
The system is going down for power off NOW!
........

Seems like it's worth investigating what happened here, and if its possible to make sure that jobs will be marked as failed when something like this happens?

Separately, I think we can also improve the rpc test framework to fail quicker and more obviously in this particular situation; I'll PR something soon.

Wow, thanks for noticing! I have marked this issue with 'upstream', as this should be handled by travis?

Counting the number of dots indicates that the previous test passed 10 minutes before the shutdown. Travis fails after 10 minutes of empty stdout/stderr, but it should not shut down...

Looks like this was a one-off upstream error: https://github.com/travis-ci/travis-ci/issues/4924#issuecomment-221993222