DynamoDB Consistency Models and ACID Properties

DynamoDB offers two read consistency models:

1. Eventually Consistent Reads:

   • Default read model

   • Provides best read performance

   • May not reflect the results of a recently completed write

   • Typically consistent within a second

2. Strongly Consistent Reads:

   • Returns a result that reflects all writes that received a successful response prior to the read

   • Might have higher latency

   • May be less available in case of network delays or outages

   • Consumes more throughput capacity than eventually consistent reads

ACID Properties in DynamoDB

ACID stands for Atomicity, Consistency, Isolation, and Durability. DynamoDB supports ACID transactions, but with some nuances:

1. Atomicity:

   • Supported through TransactWriteItems and TransactGetItems operations

   • All specified actions succeed, or none do

2. Consistency:

   • Immediate consistency within a single item

   • Cross-item consistency achieved through transactions

3. Isolation:

   • Transactions are serializable

   • Isolated from other operations

4. Durability:

   • Data is synchronously replicated across multiple facilities in an AWS Region

Key Points

1. Read Consistency Choice:

   • Eventually consistent reads are the default

   • Strongly consistent reads must be explicitly requested

   • Strongly consistent reads use more RCUs

2. Performance vs. Consistency Trade-off:

   • Eventually consistent reads offer better performance and availability

   • Strongly consistent reads offer up-to-date data at the cost of higher latency

3. Global Tables and Consistency:

   • Multi-region replication uses eventually consistent reads

   • Strongly consistent reads are only available within the same region

4. ACID Transactions:

   • Supported across multiple items and tables

   • Can span up to 25 items or 4 MB of data

5. Use Cases:

   • Eventually consistent: High-throughput reads where slight inconsistency is acceptable

   • Strongly consistent: Financial transactions, inventory systems where immediate consistency is crucial

6. Capacity Consumption:

   • Strongly consistent reads consume twice the RCUs of eventually consistent reads