handle create repl_dev failure

[Besroy]

Creating a replication device (repl dev) can fail and leave garbage repl devs on certain members. For example:

• Leader adds member F2, but the operation times out.
• Leader assumes F2 is not in the group, but F2 successfully joins.
• When the group is destroyed, F2 is not included, leaving an orphaned group on F2.

Related Logs

Leader:

• Sent join request to F2:

  [09/20/25 09:01:12.178923] [I] [76] [handle_join_leave.cxx:149:invite_srv_to_join_cluster] sent join request to peer 984278082, 98f2b032-095d-4d78-a069-d1e21f95603d [group=e3be8382-63f2-4edd-85ed-8851e8e65641]
  

• Timeout occurred:

  [09/20/25 09:01:14.179209] [I] [61] [raft_server.cxx:1639:handle_ext_resp_err] receive an rpc error response from peer server, Deadline Exceeded 12 [group=e3be8382-63f2-4edd-85ed-8851e8e65641]
  

Follower (F2):

• Received the join request after the leader timed out:

  [09/20/25 09:01:14.390810] [I] [65] [handle_join_leave.cxx:188:handle_join_cluster_req] got join cluster req from leader 2114978300 [group=e3be8382-63f2-4edd-85ed-8851e8e65641]
  

For more context, refer to the related discussion: GitHub PR #136 - Discussion.

[Besroy]

@JacksonYao287 Could you help check if it can be closed when eBay/HomeObject#432 merged

[JacksonYao287]

they are different problems.

eBay/HomeObject#432 aims to redo pg resource cleanup if crash occurs after pg is marked as destroyed and before completing pg resource cleanup.

this issue is more about raft level. the problem is how to deal with orphaned group on a member