As part of SQL Server in-memory database recovery, a parallel task is started to recover in-memory objects. In certain scenarios, this task could fail. There is a race condition during log recovery that, if the failure happens at the last phase of log redo, the parallel task misses the failure notification and does not exit. This translates to a database hang during recovery.
Note This is a transient-race condition issue, and a server restart would unblock the database.
This fix is included in the following update for SQL Server:
This update fixes the missed notification so that the recovery task detects the failure and leaves the database as recovery failed. Then an ALTER DATABASE SET ONLINE statement can be issued to restart the database recovery process.
Microsoft has confirmed that this is a problem in the Microsoft products that are listed in the "Applies to" section.