WCU & RCU

  • Provision

    • ( N = number of items per second; M = item size in KB )
    • WCU for Standard Write: N * ceil(M)
    • RCU for Eventually Consistent Read: (N / 2) * ceil(M / 4)
    • RCU for Strongly Consistent Read: N * ceil(M / 4)
    • Transactional Write / Read costs 2 * WCU / RCU

  • On-demand

APIs

  • PutItem
    • Create a new Item or replace an old item
  • UpdateItem
    • Update attributes of an existing item or add a new item
    • Can be used for Atomic Counters
  • GetItem
    • Read based on primary key (partition key or partition key + sort key)
    • ProjectionExpression can be specified to retrieve only certain attributes
  • Query
    • KeyConditionExpression
      • Partition key value (only = expression)
      • Sort key expression
    • FilterExpression
      • Used on non-key attributes
    • Limit
      • Limit number of retrieved items
    • Up to 1 MB of data each query. Use pagination to recursively retrieve data
  • Scan
    • Scan the entire table and then filter out data
    • Up to 1 MB of data each page
    • Can use Limit, ProjectionExpression, and FilterExpression
  • ParallelScan
  • DeleteItem
    • Ability to perform a conditional delete
  • DeleteTable
  • Batch Operations
    • BatchWriteItem
      • Up to 25 PutItem & DeleteItem in one call
    • BatchGetItem
      • Return items from one or more tables
      • Up to 100 items and 16 MB of data

Secondary Indexes

  • LSI
    • Alternative Sort Key
    • Up to 5 LSI per table
    • Must be defined at table creation time
    • Attribute projections (attributes included in the LSI)
      • KEYS_ONLY
      • INCLUDE
      • ALL
  • GSI
    • Alternative Primary Key
    • Can be added or modified after table creation
    • Must provision RCU and WCU for GSI
    • Attribute projections
    • Write to a table will consumed WCUs of the table and all GSIs of the table

Optimistic Locking (Conditional Writes)

  • Perform a write/update/delete only if conditions are met
    • In case that multiple uses update one table at the same time

DAX

  • Cache cluster in VPC
  • Up to 10 nodes
  • Multi-AZ

Streams

  • Triggers
    • Kinesis Data Streams
    • Lambda
    • KCL applications
  • Data retention can be up to 24 hours
  • Information in streams
    • KEY_ONLY
    • NEW_IMAGE
    • OLD_IMAGE
    • NEW_AND_OLD_IMAGE

Copy Table

  • Use Data Pipeline
    • Copy the data to S3 through a EMR cluster
    • Load the data from S3 through a EMR cluster
  • Backup to / restore from S3 on yourself