Comment 7 for bug 1516933

Krunal Bauskar (krunal-bauskar) wrote :

Dom,

I am not sure why you expect a innodb_deadlock count to increase with the given use-case.

Let me get some background/semantics:

a. innodb_deadlock is used to track the deadlock traced by the innodb engine on the given node.

b. blog use-case is not inducing a deadlock on node level but at cluster level. innodb_deadlock is not meant to capture cluster level deadlock. From the node perspective there wasn't any deadlock or can't be as node-1 has only one active session.
node-1 trx failed due to certification failure at cluster level.

So let's not mixup use of innodb_deadlock to monitor cluster deadlocks. They are 2 different semantics.

Also, note that if the same use-case is tried on same node InnoDB locks will not allow the use-case to proceed but cluster is allowing it due to pessimistic locking protocol (this is even mentioned in blog)

------------

Hope this clarifies the issue.