It's a little unclear from the article, but based on
Since the system supports asynchronous replication, our data centers eventually become consistent. Writes are considered complete as soon as the remote data center acknowledges it.
I assume its eventually consistent (so AP).
I wish they compared why normal mysql async replication fails for them and they needed to build this though.
Comments
Go ahead and downvote this; but where's the source?
Otherwise isn't this pointless bragging?
So according to this post they created geographically distributed, multi master (with a mesh topology) that is always consistent.
Either they created a holy grail of databases, or an opportunity to write a great Post-Mortem in the future.
It's a little unclear from the article, but based on
I assume its eventually consistent (so AP).
I wish they compared why normal mysql async replication fails for them and they needed to build this though.
Or it's just incredibly slow and not always available
It is not pointless for Uber. Please read it as a job advertisement for scalable database experts.