We only set engine binlog position in ordered_commit path flush phase
and not all transactions go through them (such as table copy in DDL).
So in cases where a DDL statement implicitly commits earlier transaction
and starting a new one, the new transaction could be “leaking” the
engine binlog pos. In order to avoid that and accidentally overwrite
binlog position with previous location, we reset it here.
Do you have any test case for this ?
Avoid and accidentally overwrite binlog position with previous location.
is it a bug ?