Right now, after a test fails we still try to do clean shutdown of all servers in the cluster under test. This results in slower turnaround and a lot of useless log messages after the failures that we need to scroll past to find the failure message. This is particularly bad for tests that use a larger number of servers, like sharding tests. There is also a risk that the servers were left in a state where they hang while shutting down (eg if some failpoints were left active). Instead, once we know we have a definite failure, we should just abort the servers as quickly as possible using SIGKILL.

I think this applies both to servers launched by the test itself (eg by ShardingTest), as well as for externally managed servers launched by resmoke.