Part of a series looking back at Build 2017, going over the 20+ pages of notes I took.
One of the big announcements at Build 2017 was the release of CosmosDB. This is a rebranding of DocumentDB, the document store in Azure, with some additional capabilities. You can also choose to store as
- Column-family (columnar)
The service is interesting in that it supports the MondoDB and Gremlin protocols for querying, which should allow some people to consider Azure instead of those platforms. The announcement was interesting from the data perspective. Microsoft focused a decent amount of time on this platform, and certainly people at the conference were very interested in the offering. Quite a few developers were thinking this would replace SQL Server, since it has some nice capabilities.
One of the big ones is that CosmosDB offers < 10ms latency for reads and < 15ms for writes. They scale up to millions of transactions/sec, and also grow to PBs in size. There are some good guarantees for the database.
They call it infinite, but that’s a marketing term. There’s a limit, and there certainly may be a limit you’re willing to store in the service. Jet.com was the featured customer and they talked about how much data they used and how they can scale the platform. I’m sure they went with DocumentDB and moved over, but they have their story in the keynote, if you want to watch. They get over 100 trillion, yes trillion, CosmosDB transactions/day. No matter what you do, a trillion of anything is a lot.
Perhaps the more interesting thing is that Azure is offering five different consistency models in CosmosDB. They are
- Bound Staleness
- Consistent Prefix
I don’t know how well these are supported in other platforms, but I like to see that developers have a choice. I’m not sure how many will understand the trade-offs and implications, and how many will get burned by choosing one over the others, but I’m glad the choice exists.
They are also saying a money back guarantee. What that means, or how you get money back, is going to be something to see. I haven’t always been thrilled with the disclosure in billing for new parts of Azure services, but perhaps there is going to be some way to request credit, though I would hope that any telemetry that shows issues results in some credit.
The other good thing is that the SLA is guaranteed across multiple dimensions: latency, throughput, availability, and consistency.
Rimme Nehme, who demo’d the product, had an interesting quote. It was something like “developers can concentrate on the app, not the database.” Forgive me if I’ve slightly misquoted as I’m going from notes. While I cringe a bit as a data person, I do know that most developers don’t want to really work deep in the database, especially for many simple apps. If I were building some simple mobile app, I’d seriously look at CosmosDB, and I plan to learn a bit about it.