Overview of Dynamodb-mapper

DynamoDB is a minimalistic NoSQL engine provided by Amazon as a part of their AWS product.

DynamoDB allows you to stores documents composed of unicode strings or numbers as well as sets of unicode strings and numbers. Each tables must define a hash key and may define a range key. All other fields are optional.

Dynamodb-mapper brings a tiny abstraction layer over DynamoDB to overcome some of the limitations with no performance compromise. It is highly inspired by the mature MoongoKit project

Requirements

• Boto = 2.6.0
• AWS account

Highlights

• Python <–> DynamoDB type mapping
• Deep schema definition and validation with Onctuous (new in 1.8.0)
• Multi-target transaction (new in 1.6.0)
• Sub-transactions (new in 1.6.2)
• Migration engine (new in 1.7.0)
• Smart conflict detection (new in 1.7.0)
• Full low-level chunking abstraction for scan, query and get_batch
• Default values
• Auto-inc hash_key
• Framework agnostic

Logging

Dynamodb-mapper uses 3 loggers:

• model
• model.database-access
• transactions

Known limitations

• Dates nested in a dict or set can not be saved as datetime does not support JSON serialization. (issue #7)

Important performance notice

To avoid superfluous DescribeTable requests, Table objects are cached after their first request. As it removes an extra overhead of ~20ms that was before each requests, it might let you go faster than your provisioned throughput. This will trigger an exponential backoff retry algorithm in Boto. While this algorithm is well done, be informed that the resulting effective throughput will be up to 40% slower than what you could have expected with your configured throughput.

If you ever reach this situation, you have 2 options:

• either increase (if bandwith allows) your provisioned throughput.
• either introduce manual, carefully chosen sleeps.

With the second option, I have been able to reach the theoretical throughput on an extremely intensive and non representative benchmark.