Skip to content

A lightweight system to automatically scale Kinesis Data Streams up and down based on throughput.

License

Notifications You must be signed in to change notification settings

quanta-computing/kinesis-auto-scaling

 
 

Repository files navigation

Kinesis Data Streams Auto Scaling

A lightweight system to automatically scale Kinesis Data Streams up and down based on throughput.

Kinesis_Auto_Scaling

Event Flow

  • Step 1: Metrics flow from the Kinesis Data Stream(s) into CloudWatch Metrics (Bytes/Sec, Records/Sec)
  • Step 2: Two alarms, Scale Up and Scale Down, evaluate those metrics and decide when to scale
  • Step 3: When a scaling alarm triggers it sends a message to the Scaling SNS Topic
  • Step 4: The Scaling Lambda processes that SNS message and…
    • Scales the Kinesis Data Stream up or down using UpdateShardCount
      • Scale Up events double the number of shards in the stream
      • Scale Down events halve the number of shards in the stream
    • Updates the metric math on the Scale Up and Scale Down alarms to reflect the new shard count.

Features

  1. Designed for simplicity and a minimal service footprint.
  2. Proven. This system has been battle tested, scaling thousands of production streams without issue.
  3. Suitable for scaling massive amounts of streams. Each additional stream requires only 2 CloudWatch alarms.
  4. Operations friendly. Everything is viewable/editable/debuggable in the console, no need to drop into the CLI to see what's going on.
  5. Takes into account both ingress metrics Records Per Second and Bytes Per Second when deciding to scale a stream up or down.
  6. Can optionally take into account egress needs via Max Iterator Age so streams that are N minutes behind (configurable) do not scale down and lose much needed Lambda processing power (Lambdas per Shard) because their shard count was reduced due to a drop in incoming traffic.
  7. Already designed out the box to work within the 10 UpdateShardCount per rolling 24 hour limit.
  8. Emits a custom CloudWatch error metric if scaling fails, you can alarm off this for added peace of mind.
  9. Can optionally adjust reserved concurrency for your Lambda consumers as it scales their streams up and down.

Deployment

  1. Download kinesis_scaling_cf.yaml CloudFormation template
  2. Download kinesis_scaling.zip and upload it to your S3 Bucket
  3. Deploy the kinesis_scaling_cf.yaml CloudFormation template and point it to the kinesis_scaling.zip file in your S3 Bucket.

Testing

To generate traffic on your streams you can use Kinesis Data Generator.

Modifying / Recompiling the Lambda

Simply edit the scale.go file as needed and run ./build to genrate a kinesis_scaling.zip suitable for Lambda deployment. Go 1.15.x is recommened.

Update (12/1/2021)

Amazon Kinesis Data Streams On-Demand mode is now the recommended way to natively auto scale your Amazon Kinesis Data Streams.

About

A lightweight system to automatically scale Kinesis Data Streams up and down based on throughput.

Resources

License

Code of conduct

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Go 81.2%
  • HCL 18.8%