Okay, I'll admit it: I've been side-eyeing Azure's Cosmos DB for a while. The power was undeniable, but the cost? Oof. It always felt like using a Formula 1 car to run errands. Great for high-performance, mission-critical stuff, but overkill for my marketing automation prototypes or low-volume, event-driven data.
The new serverless tier seems like they finally read my diary 😅. Pay-per-request pricing that actually scales to zero (or near-zero) when idle? That's the kind of flexibility I'm used to from DynamoDB or Firestore for smaller projects.
**My main question for the cloud-savvy crowd:** Has anyone run the numbers on real-world scenarios yet? I'm thinking about use-cases like:
- Storing user event streams for a new personalization engine (bursty traffic).
- Housing a small/medium-sized customer data platform that needs low-latency reads.
- Session storage for a web app with unpredictable user concurrency.
Specifically:
1. **Latency:** Is the performance hit noticeable compared than provisioned throughput for, say, sub-10ms P99 reads?
2. **Cost crossover point:** At what sustained request volume does it become more expensive than a modest provisioned RU/s setup?
3. **The "gotchas":** Are there hidden throttling limits or configuration quirks that bite you?
I saw the published $0.25 per million RU's, but translating that to actual monthly bills for spiky workloads is an art form. If anyone has migrated a test workload or done a head-to-head with something like AWS DynamoDB On-Demand, I'm all ears. A good serverless database means I can automate more data-heavy workflows without sweating the infrastructure cost.
It's not marketing, it's logic.