PROVIDER Β· Globally distributed multi-model cloud database

Azure Cosmos DB

Azure Cosmos DB is Microsoft's globally distributed multi-model database, replicating to any Azure region with multi-region writes and offering five consistency levels plus multiple compatible APIs as a low-latency data layer for multinational SaaS.

Backend & Database United States 2026.11.15
Backend & Database United States Globally distributed multi-model cloud database
4.4pts

Pros

4 items
One-action global distribution with multi-region writes and automatic failover
Five consistency levels tune the latency-versus-consistency trade-off precisely
Multiple APIs including NoSQL, MongoDB, Cassandra, and Gremlin lower migration effort
Vendor SLA promises 99.999 percent availability with single-digit millisecond P99 latency

Cons

4 items
RU billing is query-shape sensitive, making cross-partition cost high and hard to forecast
A wrong partition key is hard to change, requiring container rebuild and data migration
Compatible APIs differ from native versions, so migration needs item-by-item validation
Confined to Azure regions, preventing true cross-cloud data locality

Product Information

azure-cosmos-db
Category Backend & Database
Company Azure Cosmos DB
Service Region United States
Recommendation Score 4.4 pts
Published 2026-11-15

Detailed Notes

Overview

Azure Cosmos DB launched in 2017 from the 2014 DocumentDB, Microsoft's globally distributed multi-model database supporting NoSQL, MongoDB, Cassandra, Gremlin, Table, and PostgreSQL APIs with vendor-claimed single-digit millisecond P99 reads and writes and 99.999 percent availability.

For Backend & Database work the core is local reads and writes worldwide plus five consistency levels, since data replicates to any Azure region with one action and multi-region writes enabled, choosing among strong, bounded staleness, session, consistent prefix, and eventual per workload, hard to match cheaply in multinational SaaS.

Be careful with RU billing and partitioning, because every operation costs request units, cross-partition queries and large result sets multiply consumption and can blow the budget, and a wrong partition key creates hot partitions whose fix requires rebuilding the container.

Core Strengths

  • One-action global distribution with multi-region writes and automatic failover
  • Five consistency levels tune the latency-versus-consistency trade-off precisely
  • Multiple APIs including NoSQL, MongoDB, Cassandra, and Gremlin lower migration effort
  • Vendor SLA promises 99.999 percent availability with single-digit millisecond P99 latency

Product Ecosystem

NoSQL API and partitioning

The native API distributes containers by partition key with RU consumption tightly tied to query shape and customisable indexing policy.

MongoDB, Cassandra, and Gremlin APIs

Protocol-compatible forms ease migration of existing apps though some advanced features differ from native versions.

Multi-region and consistency

Read and write priorities plus failover order are set per region, with consistency adjustable at account and container level.

Drawbacks

  • RU billing is query-shape sensitive, making cross-partition cost high and hard to forecast
  • A wrong partition key is hard to change, requiring container rebuild and data migration
  • Compatible APIs differ from native versions, so migration needs item-by-item validation
  • Confined to Azure regions, preventing true cross-cloud data locality

Use Cases

  • Multinational SaaS (β˜…β˜…β˜…β˜…β˜…) - Local reads and multi-region writes
  • Global profile and session data (β˜…β˜…β˜…β˜…β˜…) - Low-latency point reads
  • Existing MongoDB apps moving to cloud (β˜…β˜…β˜…β˜…) - Protocol compatibility cuts rework
  • Complex joins and reporting (β˜…β˜…) - Use analytical storage

Pricing

Plan Price Scope
Serverless Per RU/s and storage Intermittent load with autoscale
Provisioned throughput Per RU/s-hour Better for steady load
Multi-region writes Throughput times region count Replication doubles or triples cost

Cosmos DB cost is almost entirely RU driven, with point reads near one RU while cross-partition queries can consume dozens or hundreds, so the levers are partition key choice, avoiding container scans, and short TTLs for cold data, while multi-region writes scale cost linearly and should follow real locality needs.

FAQ

  • Cosmos DB or MongoDB Atlas? Cosmos DB for deep Azure investment with multi-region low-latency writes, Atlas for fuller MongoDB features and cross-cloud flexibility.

  • How is RU cost controlled? Prefer point reads on the partition key, avoid cross-partition sorts and aggregations, page large result sets, set TTLs on cold data, and cap autoscale maximums.

  • Does migration require code changes? Most CRUD carries over with the MongoDB API, though transactions, aggregation pipelines, and index semantics differ enough to need regression testing and RU cost modelling.

Visit Azure Cosmos DB Website β†’ https://azure.microsoft.com/products/cosmos-db/
Visit Website β†—