DynamoDB - utkaln/dynamodb-helloworld Wiki

DynamoDB Fundamentals

  • Usually a single table suffices all the needs for the project. Hence a good idea would be separate tables based on environments, such as DEV.TABLE_NAME, TEST.TABLE_NAME etc.
  • Tables names must be unique within a region name.

Data Types Supported in DynamoDB

  • Scalar : String, Number, Boolean. This type data can be used as Partition key, Sort key or Index
  • Partition key must be less than 2KB, Sort key must be less than 1KB
  • Map : Unordered list of objects of same type. Can not have duplicate values
  • List : Ordered list of objects, can be of any type data, can have duplicate values. Similar to JSON array
  • Map : Unordered list of key value pairs, can have any type data. Similar to hashmap

Consistency of Data

  • DynamoDB maintains 3 different copies internally, with near realtime replication (< 1s)
  • Eventually consistency is 50% cheaper than strong consistency
  • Default choice is eventual consistency. Strong consistency can be achieved by on demand

Throughput and Scaling

  • 1 capacity unit (cu) = 1 request / second
  • Calculation is done for Read Capacity Unit (RCU) or Write Capacity Unit (WCU). These are the major factors for budgeting. Though storage has cost too
  • Important to consider scaling based on RCU and WCU
  • RCU
    • 1 Strong Consistent Read Request / second
    • 2 Eventual Consistent Read Request / second
    • Can read upto 4 KB data / second
    • Example:
      • Avg Item Size : 10 KB
      • Provisioned Capacity: 10 RCU
      • Strong Consistency Throughput: 10 * 4 = 40 KB / second
      • Eventual Consistency Throughput: 10 * 2 * 4 = 80 KB / second
  • WCU
    • 1 Write Request / second
    • Can write upto 1 KB data / second
    • Example:
      • Avg Item Size : 10 KB
      • Provisioned Capacity: 10 WCU
      • Write Througput: 10 * 1 = 10 KB / second
  • DynamoDB provides burst capacity upto 300 seconds of unused capacity, this is used occassionally
  • Scale up can happen on demand anytime
  • Scale down can only happen max. 4 times a day
  • 1 partition supports upto 1000 WCUs or 3000 RCUs
  • 1 partition can store up to 10 GB of data
  • With more data requirement or througput requirement than 1 partition, DynamoDB will get more partitions, but the partition won't go away if the data size or throughput requirement comes down
  • It distributes the RCU and WCU needs evenly on all available partitions
  • Hence while planning it is important to go gradually up instead of coming to a requirement of going down

Index, Partition Key, Sort Key

  • Partition Key : Mandatory, It is a hashkey. Using a hash function DynamoDB internally decides what partition to send to. Must be a scalar type data.
  • Sort Key: Not mandatory, but to make sure that the data can be queried, it must be either part of Partition key or part of sort key
  • Simple Index - Only uses partition key, must be unique partition key for this to work
  • Composite Index - Combination of partition key and sort key. Typical case. Combination should be unique
  • Local Secondary Index - Another column that is queried over is included alongside the Partition key. Must be created during creation of database. Does not create data replica. Supports eventual consistency and strong consistency.
  • Global Secondary Index - Secondary partition and secondary sort key together create global secondary index. Creates replica of data. Can be created after DB creation. Only supports Eventual Consistency