This is related to HEAD-2281. As described there, the health checker does not distinguish between a process that terminated normally with a non-zero exit status and one that terminated abnormally because it was killed by a signal. As a result, if the consumer has configured "ignoreExitStatus" to "true" (as we do for mahi, for example), then registrar will not consider it an error if the process was terminated by a signal either. In practice, such a health check will probably still fail because the stdout will not match what the checker is looking for, but that's not necessarily the case, and registrar should really be crisper about the specific conditions that will result in failure.