Description:
Hi,
some statements/transactions are a barrier for Write Set. A type of such statements is DDL. When a DDL is executed, Write Set history is cleared and every following transaction will have their last_committed on that DDL (or later if there is a dependency).
Those statements (like DDL and probably transactions on non-transnational storage engines), generate a barrier on the slave, so they impact parallel replication performance. If such statements happen once every minute, it is no big deal, but if they happen 10 times per second, it can be a source of bad performance of parallel replication on slaves.
Such performance "killers" should be easy to identify, so I suggest adding a counter for monitoring statements/transactions that are causing Write Set barriers (a global status or some other value in Performance Schema would do).
Many thanks,
JFG
How to repeat:
Not a bug but a feature request.
Suggested fix:
Add a counter for monitoring statements/transactions that are causing Write Set barriers (a global status or some other value in Performance Schema would do).