关于Azure SQL与DocumentDB/CosmosDB成本对比及RU计量的技术咨询

算力阿峰

2026-5-21

Absolutely, plenty of engineers and cloud architects have run cost comparisons between Azure SQL Database and Cosmos DB—they’re two of Azure’s go-to data stores, so this is a super common question! Let’s break this down, starting with cost comparison basics and then diving into your RU (Request Unit)计量 question, since that’s where a lot of folks get tripped up.

Azure SQL vs. Cosmos DB Cost Comparison Basics

First, it’s important to remember their pricing models are fundamentally different, which makes direct apples-to-apples comparisons tricky without context:

Azure SQL Database: Charges based on compute (DTUs or vCores), storage (per GB/month), and backup storage. For steady, relational workloads with predictable traffic, this can be straightforward to budget for.
Cosmos DB: Charges primarily for RU/s (Request Units per second, either reserved or on-demand), plus storage (per GB/month) and data transfer. This model is built for global, low-latency, schema-flexible workloads where you need to scale throughput independently of storage.

In general:

If your workload is traditional relational, has low-to-moderate concurrency, and doesn’t need global distribution, Azure SQL will often be more cost-effective.
If you need global low-latency access, flexible schemas, or high-throughput read/write operations that scale seamlessly, Cosmos DB’s cost will depend heavily on your RU needs—but it’s often the better fit for those use cases.

Understanding Cosmos DB RU Logic (Your Key Question)

Great question about whether 1 request in a 1TB database vs. 1GB database uses the same RU. The short answer: Yes, if the requests are identical—RU计量 doesn’t tie directly to total database storage size.

Here’s why: RUs are a standardized measure of the resources (CPU, memory, I/O, index usage) needed to execute a single request. The factors that determine RU consumption are:

Request type: Write operations (insert, update, delete) use more RUs than reads. For example, inserting a 1KB document typically uses ~5 RUs, while a point-read of that same document uses ~1 RU.
Document size: Larger documents require more resources to process. A point-read of a 2KB document uses ~2 RUs, vs. 1 RU for a 1KB document.
Query complexity: Queries that scan many documents, use complex filters/aggregations, or cross partitions will use far more RUs than simple point-reads. This depends on how much data the query needs to process, not the total size of the database.
Consistency level: Strong or bounded staleness consistency uses more RUs than eventual consistency, since it requires more coordination across replicas.

So, if you run the exact same point-read (same document size, same consistency level) against a 1GB Cosmos DB database and a 1TB one, the RU cost will be identical. The total storage size affects your storage cost (charged per GB/month), but not the RU cost for individual requests.

A quick pro tip: Use Azure’s built-in Cosmos DB Capacity Calculator to estimate your RU needs based on your specific workload (document sizes, request rates, query patterns). It’s a great way to avoid over-provisioning and keep costs in check.

内容的提问来源于stack exchange，提问作者user3053247