4 ms·

> If I have a distributed microservice architecture and I want to keep multiple datastores in synchronization or "consistent" what's the industry best practice?

• mhuffman 3y ago
  This is the answer. But if they are going to, I would recommend a book called "Designing Data-Intensive Applications" by Martin Kleppmann. It is a book with the most clear communication on this topic and others that I have ever read. Even if you have a degree in CS, I think it is eye-opening to see how very complex ideas can be communicated vs the same ideas in textbooks.
• matthewsinclair 3y ago
  So much this. I was part of an engineering team that built a payments switch from scratch in the early oughts. We built it in Java, on commodity hardware and OS, on top of our own replicated, stateful, distributed computing platform. This was a bonkers thing to do then, pre-cloud. It’s probably still a bonkers thing to do _today_. Anyway, we did a lot of work with 2PC and other consensus mechanisms and came to the conclusion they 2PC wasn’t up to scratch for what we needed (it’s actually provably less than ideal). We ended up building (again, from scratch) one of the earliest (that I know of) implementations of the Virtual Synchrony protocol. VS has some robust maths behind it that you can use to make some stronger consistency claims than you can with 2PC. These are important when you’re dealing with interbank settlements for payments switching. If we started again today I’d say that we might use something like Raft, but I’ve been away from the space for ~20 years now so I’m not entirely sure. However, one thing I do know is this: distributed consensus is Very Hard™ to get right. If the answer to your question involves any kind of multiple master implementation using distributed consensus I can unequivocally guarantee that (other than in a few very specific circumstances that you almost certainly do not have) you’re asking the wrong question.