Hey Team,

When we have a node on a cluster that is a replica of another server, in case you have a read-only schema, and we execute some transaction that will give us an error on the server(like an alter user for a user that didn't exist), the node leaves the clusters. Error log below:

How to repeat:

Create a PXC cluster in which one node replicates another server or another PXC cluster.

Create the database on Source:

On replica, set it to read-only:

Then, create a user without replicating, then alter this user replicating:

Now we can check the expected behavior on the replica PXC node:

However, when you try to skip the alter wrong user transaction, the node has already left the cluster:

Note: The bug won't happen if the test database is read-only disabled. (read only = 0)