Restart=always leads to the following loop:
1. scylla terminates abnormally
2. scylla-jmx sees that, and terminates
3. systemd sees that scylla-jmx terminated, and restarts it.
4. scylla-jmx requires scylla, so systemd starts it.
5. goto 1.
To prevent the loop, set Restart=on-abnormal; systemd will restart scylla-jmx
if some JVM bug got it killed, but not otherwise.
The downside to this patch is that if scylla-server goes down, so does
scylla-jmx, but if scylla-server is then restarted, scylla-jmx stays down.
To get scylla and scylla-jmx to start together, we need to create
scylla.service that requires both of them.
24cebcc9a1