deadlock when replicating from admin port

[romange]

we have 100% reproducible deadlock bug when replica connects to a master via admin port.

the problem is that one of the commands in master (migrate) requires a listener to do the thread migration. We use hardcoded "main_listener" which is incorrect of course if we connect via admin.

I am going to fix a bug, but we need test coverage for admin port replication. Moreover, we need to scan where we use main_listener in the codebase and see if it makes sense for connections coming via admin_listener or uds listener etc

[adiholden]

I dont find from which flow we Call the DFLY THREAD command.. can you share where is this call coming from?

[romange]

we do not call it, it's a debug command for us to test thread/connections interactions.