Comment 5 for bug 1692745

I Can also confirm the issue on PXC 5.6, though we "only" experience 1 out of 3 nodes to crash in a 3-way-master setup (same error though).

We were able to narrow the issue down to 2 consecutive transactions (first deletes from table that is referenced by foreign key which seem to be applied out of order by the 2 slaves. The transaction usually fails only on one node after 4 tries (which crashes the node then, because he thinks he is no longer consistent). The other slave node tries 1 or 2 times and then gets it right. I suspect some kind of out-of-order-committing, maybe there is a race condition in the code that makes sure transaction will be applied in order?

Btw, we double checked repl.commit_order to make sure it has been set to never apply out-of-order on any of the nodes... :)

Would be glad if someone from Percona could have a look if there is a configurable workaround until no patched version is released.