4 ms·

I don't know much about replication but what happens when a master dies just after the database is updated. Will the changes not be replicated? If the user see

• alfiejohn_ 15y ago
  I think spydum's comment is what you're looking for. These are things that can and do happen, and you have to plan for it. Designing your applications and infrastructure with eventual consistency in mind can do wonders for both performance and redundancy.
• jermy 15y ago
  It depends on your use case - either you accept there is a write hole that the update wont appear until that host comes back online and other hosts can relay the change, or you add application-level checking to not return to the user in the first instance until the change hits at least another host. In your scenario, yes. The two changes would be a conflict, but hopefully (if the user makes the same change) your conflict-resolution code would realise it is the same and merge them correctly. I fear I don't know how MySQL::Replication handles this, but it has to be application-level logic, so I assume it will allow you to override a default of, say, last-modification-wins.
• ww520 15y ago
  That's one of the problems of multi-master replication. It's easy to introduce conflicts and requires lots of application logic or operation manual work to handle them.