In our mysql cluster setup for eZ Publish we have the front end running using slaves.

We have noticed a problem when the user activates their account (using the user/activate view) where in some cases roles are not assigned for that object.

By some debugging we have concluded that the optimistic eZDB implementation will happily write to master and try to read the same object from slave. We now clearly see that replication will not always be able to finish in time for the next read of that object.

My proposal is to let the ezdb implementation be aware of writes and always use the master for further database actions for that request.

See https://github.com/ezsystems/ezpublish/pull/339 for a possible patch