Lambda provisioned concurrency

What is Provisioned Concurrency for AWS Lambda?

AWS Lambda as a main building block of serverless world. Overall serverless computing is great for various use cases, but still has some limitations. One of them could be latency. When lambda hadn’t been used for some time an execution environment needs to be created. It can take from 200ms to even 600ms, depending on chosen language. So when latency is critical for your application, additional 600ms might be painful. So here Provisioned Concurrency comes to the rescue! It creates execution environment just after function deployment, and it keeps function in initialized state, so latency is no longer an issue.

How to setup provisioned concurrency

I think that it’s always good to follow best practices, so on my blog I will try present demos using infrastructure as a code approach. Additional benefit is that you can run template via SAM or AWS Cloudformation, instead manually clicking setup in console.

In this post I will use SAM for demo purpose. Make sure you have SAM cli installed

Setup is very simple. Let’s go through it.

sam init

Then, in SAM template we need to extend our function configuration by adding ProvisionedConcurrencyConfig property:

ProvisionedConcurrencyConfig:
  ProvisionedConcurrentExecutions: 5

But it’s not enough. When you’ll try to deploy your function in this stage, you’ll get following error:

Error: Failed to create changeset for the stack: lambda-provisioned-concurrency,
ex: Waiter ChangeSetCreateComplete failed: Waiter encountered a terminal failure state Status: FAILED.
Reason: Transform AWS::Serverless-2016-10-31 failed with: Invalid Serverless Application Specification document.
Number of errors found: 1. Resource with id [HelloWorldFunction] is invalid.
To set ProvisionedConcurrencyConfig AutoPublishALias must be defined on the function...

So as we clearly see we need to include an alias to our function config. Let’s add it:

AutoPublishAlias: live

So here’s our final configuration:

Resources:
  HelloWorldFunction:
    Type: AWS::Serverless::Function
    Properties:
      CodeUri: hello-world/
      Handler: hello-world
      Runtime: go1.x
      AutoPublishAlias: live
      ProvisionedConcurrencyConfig:
        ProvisionedConcurrentExecutions: 5
      Events:
        CatchAll:
          Type: Api
          Properties:
            Path: /hello
            Method: GET

So now we can deploy our function to AWS:

sam deploy --stack-name <you-stack-name> --guided --capabilities CAPABILITY_IAM

And after that, when deployment will be completed, you should be able to test your function.

Results

To test results you can follow approach from AWS blog article. They’re using ab tool to test function latency. Parameter n defines number of requests to perform. Parameter c defines number of multiple requests beeing performed at a time.

ab -n 1000 -c 10

You can run this function on two lambdas, one with enabled provisioned concurrency and the second with disabled provisioned concurrency. On AWS blog you could check results of similar experiment. Also you can run example that I created on my github account.

#aws #lambda #serverless