I'm trying to evaluate my options for a persistent store according to the following usage constraints:
- service that has flow requirements of about 1000 tps.
- simple data structures (little relational atm)
- ACID guarantees & strong reading consistency
- ~ 2: 1 write / read
- about 10 million lines / day
- HA warranties
(for example, high-performance store for large amounts of highly sensitive transaction information, etc.)
My experience is almost solely with postgres. Over time, he has evolved a lot in different spaces. However, I wonder if postgres is best suited to my use case.
I essentially evaluate postgres against a NoSQL store with ACID warranties and high read consistency in a specified topology (for example, DynamoDB). I've heard horror stories about Dynamo's costs though …
I want to spread the store across a number of regions. Each region must therefore have a significant coherence in reading, but I agree with the possible coherence between the regions.
Can any one explain why I should not just use Postgres and maybe Citus if I have to evolve even more?