Does dynamoDB provide snapshot isolation backups

Question

Hi I am confused about what consistency level that dynamoDB backups can provide.

In https://docs.aws.amazon.com/amazondynamodb/latest/developerguide/CreateBackup.html, it says "DynamoDB backups do not guarantee causal consistency across items; however, the skew between updates in a backup is usually much less than a second.".

But in https://aws.amazon.com/blogs/database/set-up-scheduled-backups-for-amazon-dynamodb-using-aws-backup/, it says "On-demand provides you a fully consistent snapshot of your table data and stay around forever (even after the table is deleted).".

And in the paper "Amazon DynamoDB: A Scalable, Predictably Performant, and Fully Managed NoSQL Database Service" https://assets.amazon.science/33/9d/b77f13fe49a798ece85cf3f9be6d/amazon-dynamodb-a-scalable-predictably-performant-and-fully-managed-nosql-database-service.pdf#:~:text=Backups%20or%20restores%20don%E2%80%99t%20affect,DynamoDB%20table%20at%20any%20time, section 5.5 seems to indicate that it can achieve SI (snapshot isolation) backups but in second granularity.

My question is whether dynamoDB on-demand backup and PITR achieve SI backups? Thank you!

Leeroy Hannigan · Accepted Answer

Always consider official documentation to be your source of truth.

DynamoDB doesn't provide snapshot isolation, but it is typically consistent within 1 second granularity. This only becomes a concern if you're using Transactions, in which your snapshot could contain some, all or none of the items written in a transaction during the snapshot time-frame.

re:Post Agent · Answer

DynamoDB's backup capabilities have some nuances regarding consistency levels:

For on-demand backups, DynamoDB creates backups that are generally consistent but with some important caveats. While they are often referred to as "fully consistent snapshots" in some documentation, the technical reality is that they don't guarantee perfect snapshot isolation (SI) across all items. There may be a very small skew (usually less than a second) between updates in a backup.

Point-in-Time Recovery (PITR) provides continuous backups with per-second granularity, allowing you to restore to any given second within your configured recovery period (1-35 days). PITR backs up your table data automatically and continuously, which helps protect against accidental writes and deletes.

Both backup methods operate with zero impact on table performance or availability, which is a significant advantage. However, neither method provides perfect snapshot isolation in the strict database theory sense. The slight inconsistency is typically negligible for most use cases (less than a second of potential skew), but it's important to be aware of this limitation if your application requires absolute causal consistency across all items in a backup.

In practical terms, DynamoDB backups provide a high level of consistency that is sufficient for most backup and recovery scenarios, even if they don't meet the strictest definition of snapshot isolation.  
**Sources**  
[Using on-demand DynamoDB backup and restore - Amazon DynamoDB](https://docs.aws.amazon.com/amazondynamodb/latest/developerguide/backuprestore_HowItWorks.html)  
[Amazon DynamoDB - Big Data Analytics Options on AWS](https://docs.aws.amazon.com/whitepapers/latest/big-data-analytics-options/amazon-dynamodb.html)  
[Amazon DynamoDB point-in-time recovery (PITR)](https://aws.amazon.com/dynamodb/pitr/)

Does dynamoDB provide snapshot isolation backups

Add your answer

Relevant content