Status: Closed (View Workflow)
As reported by Pavel Ivanov:
The scenario is that IO thread is using GTID to find the starting position
when connecting to the master. In case of re-connect in the middle of an event
group, the IO thread needs to skip already received events following the GTID
event of the reconnect starting position.
The bug is that when reconnect happens at the end of an event group, the IO
thread actually re-connects from the start of that event group, even though it
could start from the next event group. This does not cause wrong behaviour
(the event group is fetched and skipped as it should), however it causes
reporting of GTID position in error log and old-style positions read from teh
relay log to be a bit off, which could be confusing to users.