Azure Cosmos DB Consistency Levels Explained

Cosmos DB offers five consistency levels, from strongest to weakest. Choosing the right one balances performance, availability, and consistency guarantees.

The Five Levels

1. Strong

• Reads always return most recently committed write
• Global ordering of operations
• Highest latency, lowest throughput
• Use when: Financial transactions, inventory systems

2. Bounded Staleness

• Reads lag behind writes by at most K versions or T time
• Global ordering guarantee
• Use when: Gaming leaderboards, stock quotes

3. Session (Default)

• Strong consistency within a session
• Monotonic reads, writes, and read-your-writes
• Best balance for most applications
• Use when: User profiles, shopping carts

4. Consistent Prefix

• Reads never see out-of-order writes
• No gaps in sequence
• Use when: Analytics, social feeds

5. Eventual

• No ordering guarantees
• Lowest latency, highest throughput
• Use when: Likes/retweets, non-critical counters

Setting Consistency

from azure.cosmos import CosmosClient, PartitionKey

# Account-level default
client = CosmosClient(url, credential, consistency_level='Session')

# Per-request override (weaker only)
container.read_item(
    item='123',
    partition_key='partition1',
    consistency_level='Eventual'
)

Cost Impact

Level	RU Cost	Latency
Strong	2x	Highest
Bounded Staleness	2x	High
Session	1x	Medium
Consistent Prefix	1x	Low
Eventual	1x	Lowest

Strong and Bounded Staleness cost double because reads wait for quorum.

Practical Guidance

Start with Session - it provides intuitive consistency for users while maintaining good performance. Only go stronger when you have proven requirements.