Release 5 sieze-up, followed by crash on replica promotion about keydb HOT 3 CLOSED

This has been fixed in change a19d0f6 which is now in the RELEASE_5 branch.

The crash occurs when all conditions are satisfied:

• The keydb instance is connecting with its master
• The master is preparing to send the RDB dump of the database but has not started yet
• The replication sequence is aborted with "replicaof no one" from a client on any thread other than the main thread.

from keydb.

(I think there's a typo in the AE_ASSERT macro output :)

from keydb.

I've found the flaw in my test setup. SO_REUSEPORT likes to cluster clients to as few threads as possible, and test scenarios necessarily have fewer than the real world. So xthread scenarios aren't getting exercised as much as I had thought.

In any case this flaw is fairly simple to fix, this particular codepath assumes replication teardown will only happen on the main thread which was a very early assumption and is no longer true.

from keydb.