Creating missing remote databases during recovery can fail because the client ID is sent across the wire and looked up on remote nodes. If the client ID on a remote node maps to a non-recovery-daemon process then the missing database will not be attached - the attach will be deferred.
This can be fixed when attaching to a database by confirming that the source node is the current node before looking up a client ID.
This could be also fixed by clearing the client ID when forwarding a packet to another node. However, this change would be more invasive.
Created attachment 14284 [details]
Patch for 4.7
Created attachment 14285 [details]
Patch for v4.8
This is ready for v4.7 and v4.8.
(In reply to Amitay Isaacs from comment #3)
pushed to autobuild-v4-[7,8]-test.
(In reply to Karolin Seeger from comment #4)
Pushed to both branches.
Closing out bug report.