When a DS stops without sending a replica offline message, it blocks computing of the change number on the replication servers. Various reasons exist for a DS to stop abruptly:
- kill -9
- linux OOM killer
- when the machine where a DS is installed dies and is irrecoverable
To unblock computing of the change number, we could add a subcommand to dsreplication that would (in the end) send a ReplicaOfflineMsg on behalf of the DS that is gone. This will tell the replication servers to not wait for it when computing change numbers.