Serverless Plugin Warmup

Keep your lambdas warm during Winter.

user  

juanjoDiaz

Serverless WarmUp Plugin ♨

Serverless npm version npm monthly downloads Build Status Coverage Status license

Keep your lambdas warm during winter.

Requirements:

  • Serverless v1.12.x or higher (Recommended v1.33.x or higher because of this).
  • AWS provider

How it works

WarmUp solves cold starts by creating a scheduled lambda (the warmer) that invokes all the selected service's lambdas in a configured time interval (default: 5 minutes) and forcing your containers to stay warm.

Installation

Install via npm in the root of your Serverless service:

npm install --save-dev serverless-plugin-warmup

Add the plugin to the plugins array in your Serverless serverless.yaml:

plugins:
  - serverless-plugin-warmup

Configuration

The warmup plugin supports creating one or more warmer functions. Warmers are defined under custom.warmup in the serverless.yaml file:

custom:
  warmup:
    officeHoursWarmer:
      enabled: true
      events:
        - schedule: cron(0/5 8-17 ? * MON-FRI *)
      concurrency: 10
    outOfOfficeHoursWarmer:
      enabled: true
      events:
        - schedule: cron(0/5 0-7 ? * MON-FRI *)
        - schedule: cron(0/5 23 ? * MON-FRI *)
        - schedule: cron(0/5 * ? * SAT-SUN *)
      concurrency: 1
    testWarmer:
      enabled: false

The options are the same for all the warmers:

  • folderName Folder to temporarily store the generated code (defaults to .warmup)
  • cleanFolder Whether to automatically delete the generated code folder. You might want to keep it if you are doing some custom packaging (defaults to true)
  • name Name of the generated warmer lambda (defaults to ${service}-${stage}-warmup-plugin-${warmerName})
  • role Role to apply to the warmer lambda (defaults to the role in the provider)
  • tags Tag to apply to the generated warmer lambda (defaults to the serverless default tags)
  • vpc The VPC and subnets in which to deploy. Can be any Serverless VPC configuration or be set to false in order to deploy the warmup function outside of a VPC (defaults to the vpc in the provider)
  • memorySize The memory to be assigned to the warmer lambda (defaults to 128)
  • events The event that triggers the warmer lambda. Can be any Serverless event (defaults to - schedule: rate(5 minutes))
  • package The package configuration. Can be any Serverless package configuration (defaults to { individually: true, exclude: ['**'], include: ['.warmup/${warmerName}/**'] })
  • timeout How many seconds until the warmer lambda times out. (defaults to 10)
  • environment Can be used to set environment variables in the warmer lambda. You can also unset variables configured at the provider by setting them to undefined. However, you should almost never have to change the default. (defaults to unset all package level environment variables. )
  • tracing Specify whether to enable/disable tracing at the function level. When tracing is enabled, warmer functions will use NPM to install the X-Ray client and use it to trace requests (It takes any of the values supported by serverless as boolean, Activeor PassThrough and defaults to the provider-level setting)
  • prewarm If set to true, it warms up your lambdas right after deploying (defaults to false)

There are also some options which can be set under custom.warmup.<yourWarmer> to be applied to all your lambdas or under yourLambda.warmup.<yourWarmer> to overridde the global configuration for that particular lambda. Keep in mind that in order to configure a warmer at the function level, it needed to be previously configured at the custom section or the pluging will error.

  • enabled Whether your lambda should be warmed up or not. Can be a boolean, a stage for which the lambda will be warmed up or a list of stages for which your lambda will be warmed up (defaults to false)
  • clientContext Custom data to send as client context to the data. It should be an object where all the values are strings. (defaults to the payload. Set it to false to avoid sending any client context custom data)
  • payload The payload to send to your lambda. This helps your lambda identify when the call comes from this plugin (defaults to { "source": "serverless-plugin-warmup" })
  • payloadRaw Whether to leave the payload as-is. If false, the payload will be stringified into JSON. (defaults to false)
  • concurrency The number of times that each of your lambda functions will be called in parallel. This can be used in a best-effort attempt to force AWS to spin up more parallel containers for your lambda. (defaults to 1)
custom:
  warmup:
    default:
      enabled: true # Whether to warm up functions by default or not
      folderName: '.warmup' # Name of the folder created for the generated warmup 
      cleanFolder: false
      memorySize: 256
      name: warmer-default
      role: WarmupRole
      tags:
        Project: foo
        Owner: bar 
      vpc: false
      events:
        - schedule: 'cron(0/5 8-17 ? * MON-FRI *)' # Run WarmUp every 5 minutes Mon-Fri between 8:00am and 5:55pm (UTC)
      package:
        individually: true
        exclude: # exclude additional binaries that are included at the serverless package level
          - ../**
          - ../../**
        include:
          - ./**
      timeout: 20
      tracing: true
      prewarm: true # Run WarmUp immediately after a deploymentlambda
      clientContext:
        source: my-custom-source
        other: '20'
      payload: 
        source: my-custom-source
        other: 20
      payloadRaw: true # Won't JSON.stringify() the payload, may be necessary for Go/AppSync deployments
      concurrency: 5 # Warm up 5 concurrent instances
    
functions:
  myColdfunction:
    handler: 'myColdfunction.handler'
    events:
      - http:
          path: my-cold-function
          method: post
    warmup:
      default:
        enabled: false

  myLowConcurrencyFunction:
    handler: 'myLowConcurrencyFunction.handler'
    events:
      - http:
          path: my-low-concurrency-function
          method: post
    warmup:
      default:
        clientContext:
          source: different-source-only-for-this-lambda
        payload:
          source: different-source-only-for-this-lambda
        concurrency: 1
   
  myProductionOnlyFunction:
    handler: 'myProductionOnlyFunction.handler'
    events:
      - http:
          path: my-production-only-function
          method: post
    warmup:
      default:
        enabled: prod
      
   myDevAndStagingOnlyFunction:
    handler: 'myDevAndStagingOnlyFunction.handler'
    events:
      - http:
          path: my-dev-and-staging-only-function
          method: post
    warmup:
      default:
        enabled:
          - dev
          - staging

Runtime Configuration

Concurrency can be modified post-deployment at runtime by setting the warmer lambda environment variables.
Two configuration options exist:

  • Globally set the concurrency for all lambdas on the stack (overriding the deployment-time configuration):
    Set the environment variable WARMUP_CONCURRENCY
  • Individually set the concurrency per lambda
    Set the environment variable WARMUP_CONCURRENCY_YOUR_FUNCTION_NAME. Must be all uppercase and hyphens (-) must be replaced with underscores (_). If present for one of your lambdas, it overrides the global concurrency setting.

Networking

The WarmUp function use normal calls to the AWS SDK in order to keep your lambdas warm. If you set up at the provider level or the warmer confir level that the wamer function should be deployed into into a VPC subnet you need to keep in mind a couple of things:

  • If the subnet is public, access to the AWS API should be allowed by Internet Gateway.
  • If the subnet is private, a Network Address Translation (NAT) gateway is needed so the warmers can connect to the AWS API.
  • In either case, the security group and the network ACLs need to allow access from the warmer to the AWS API.

Since the AWS SDK doesn't provide any timeout by default, this plugin uses a default connection timeout of 1 second. This is to avoid the issue of a lambda constantly timing out and consuming all its allowed duration simply because it can't connect to the AWS API.

Permissions

WarmUp requires permission to be able to invoke your lambdas.

If no role is provided at the custom.warmup level, each warmer function gets a default role with minimal permissions allowing the warmer function to:

  • Create its log stream and write logs to it
  • Invoke the functions that it should warm (and only those)
  • Create and attach elastic network interfaces (ENIs) which is necessary if deploying to a VPC

The default role looks like:

resources:
  Resources:
    WarmupRole:
      Type: AWS::IAM::Role
      Properties:
        RoleName: WarmupRole
        AssumeRolePolicyDocument:
          Version: '2017'
          Statement:
            - Effect: Allow
              Principal:
                Service:
                  - lambda.amazonaws.com
              Action: sts:AssumeRole
        Policies:
          - PolicyName: WarmUpLambdaPolicy
            PolicyDocument:
              Version: '2017'
              Statement:
               # Warmer lambda to send logs to CloudWatch
                - Effect: Allow
                  Action:
                    - logs:CreateLogGroup
                    - logs:CreateLogStream
                  Resource: 
                    - !Sub arn:aws:logs:${AWS::Region}:${AWS::AccountId}:log-group:/aws/${self:service}-${opt:stage, self:provider.stage}/*:*
                - Effect: Allow
                  Action:
                    - logs:PutLogEvents
                  Resource: 
                    - !Sub arn:aws:logs:${AWS::Region}:${AWS::AccountId}:log-group:/aws/${self:service}-${opt:stage, self:provider.stage}/*:*:*
                # Warmer lambda to invoke the functions to be warmed
                - Effect: 'Allow'
                  Action:
                    - lambda:InvokeFunction
                  Resource:
                    - !Sub arn:aws:lambda:${AWS::Region}:${AWS::AccountId}:function:${self:service}-${opt:stage, self:provider.stage}-*
                # Warmer lambda to manage ENIS (only needed if deploying to VPC, https://docs.aws.amazon.com/lambda/latest/dg/vpc.html)
                - Effect: Allow
                  Action:
                    - ec2:CreateNetworkInterface
                    - ec2:DescribeNetworkInterfaces
                    - ec2:DetachNetworkInterface
                    - ec2:DeleteNetworkInterface
                  Resource: "*"

The permissions can also be added to all lambdas using setting the role to IamRoleLambdaExecution and setting the permissions in iamRoleStatements under provider (see https://serverless.com/framework/docs/providers/aws/guide/functions/#permissions):

provider:
  name: aws
  runtime: nodejs10.x
  iamRoleStatements:
    - Effect: 'Allow'
      Action:
        - 'lambda:InvokeFunction'
      Resource:
      - !Sub arn:aws:lambda:${AWS::Region}:${AWS::AccountId}:function:${self:service}-${opt:stage, self:provider.stage}-*
custom:
  warmup:
    default:
      enabled: true
      role: IamRoleLambdaExecution

If setting prewarm to true, the deployment user used by the AWS CLI and the Serverless framework also needs permissions to invoke the warmer.

On the function side

When invoked by WarmUp, your lambdas will have the event source serverless-plugin-warmup (unless otherwise specified using the payload option):

{
  "Event": {
    "source": "serverless-plugin-warmup"
  }
}

To minimize cost and avoid running your lambda unnecessarily, you should add an early return call before your lambda logic when that payload is received.

Javascript

Using the Promise style:

module.exports.lambdaToWarm = async function(event, context) {
  /** Immediate response for WarmUp plugin */
  if (event.source === 'serverless-plugin-warmup') {
    console.log('WarmUp - Lambda is warm!');
    return 'Lambda is warm!';
  }

  // ... function logic
}

Using the Callback style:

module.exports.lambdaToWarm = function(event, context, callback) {
  /** Immediate response for WarmUp plugin */
  if (event.source === 'serverless-plugin-warmup') {
    console.log('WarmUp - Lambda is warm!')
    return callback(null, 'Lambda is warm!')
  }

  // ... function logic
}

Using the context. This could be useful if you are handling the raw input and output streams.

module.exports.lambdaToWarm = async function(event, context) {
  /** Immediate response for WarmUp plugin */
  if (context.custom.source === 'serverless-plugin-warmup') {
    console.log('WarmUp - Lambda is warm!');
    return 'Lambda is warm!';
  }

  // ... function logic
}

If you're using the concurrency option you might want to add a slight delay before returning on warmup calls to ensure that your function doesn't return before all concurrent requests have been started:

module.exports.lambdaToWarm = async (event, context) => {
  if (event.source === 'serverless-plugin-warmup') {
    console.log('WarmUp - Lambda is warm!');
    /** Slightly delayed (25ms) response 
    	to ensure concurrent invocation */
    await new Promise(r => setTimeout(r, 25));
    return 'Lambda is warm!';
  }

  // ... add lambda logic after
}

Python

You can handle it in your function:

def lambda_handler(event, context):
    # early return call when the function is called by warmup plugin
    if event.get("source") in ["aws.events", "serverless-plugin-warmup"]:
        print("WarmUp - Lambda is warm!")
        return {}

    # ... function logic

Or you could use a decorator to avoid the redundant logic in all your functions:

def skip_execution_if.warmup_call(func):
    def warmup_wrapper(event, context):
      if event.get("source") in ["aws.events", "serverless-plugin-warmup"]:
        print("WarmUp - Lambda is warm!")
        return {}

      return func(event, context)

    return warmup_wrapper

# ...

@skip_execution_if.warmup_call
def lambda_handler(event, context):
    # ... function logic

Java

You can handle it in your function:

public ApiGatewayResponse handleRequest(Map<String, Object> input, Context context) {
  if ("serverless-plugin-warmup".equals(input.get("source"))) {
    System.out.println("WarmUp - Lambda is warm!");
    return ApiGatewayResponse.builder()
        .setStatusCode(200)
        .build();
  }
  
  // ... function logic
}

Ruby

You can handle it in your function:

def handle_request(app:, event:, context:, config: {})
  if event['source'] == 'serverless-plugin-warmup'
    puts 'WarmUp - Lambda is warm!'
    return {} 
  end

  # ... function logic
end

Lifecycle hooks

WarmUp plugin uses 3 lifecycles hooks:

  • warmup:addWamers:addWamers: This is where the warmers are added to the service. It runs after:package:initialize.
  • warmup:cleanupTempDir:cleanup: This is where the warmers' temp folders are removed. It runs after:package:createDeploymentArtifacts.
  • warmup:prewarm:start: This is where the warmers are invoked. It runs after:deploy:deploy or when running the command serverless warmup prewarm.
  • warmup:prewarm:end: This is after the warmers are invoked.

Usage

Packaging

WarmUp supports

serverless package

By default, each warmer function is packaged individually and it uses a folder named .warmup/<function_name> to serve as temporary folder during the packaging process. This folder is deleted at the end of the packaging process unless the cleanFolder option is set to false.

If you are doing your own package artifact you can set the cleanFolder option to false and include the .warmup folder in your custom artifact.

Deployment

WarmUp adds package the warmers and add them to your services automatically when you run

serverless deploy

After the deployment, any warmer with prewarm: true is automatically invoked to warm up your functions without delay.

Prewarming

Apart from prewarming automatically after each deployment. You can invokes a warmer after a sucessful deployment to warm up functions using:

serverless warmup prewarm -warmers <warmer_name>

The warmers flag takes a comma-separated list of warmer names. If it's nor provided, all warmers with prewarm set to true are invoked.

Migrations

v4.X to v5.X

Support multiple warmer

Previous versions of the plugin only support a single warmer which limited use cases like having different concurrentcies in different time periods. From v5, multiple warmers are supported. The warmup field in the custom section or the function section, takes an object where each key represent the name of the warmer and the value the configuration which is exactly as it used to be except for the changes listed below.

custom:
  warmup:
    enabled: true
    events:
      - schedule: rate(5 minutes)

have to be named, for example, to default:

custom:
  warmup:
    default:
      enabled: true
      events:
        - schedule: rate(5 minutes)

Change the default temporary folder to .warmup

Previous versions of the plugin named the temporary folder to create the warmer handler _warmup. It has been renamed to .warmup to better align with the serverless framework and other plugins' behaviours.

Remembe to add .warmup to your git ignore.

Default to Unqualified alias

Previous versions of the plugin used the $LATEST alias as default alias to warm up if no alias was provided. From v5, the unqualified alias is the default. You can still use the $LATEST alias by setting it using the SERVERLESS_ALIAS environment variable.

custom:
  warmup:
    default:
      environment:
        SERVERLESS_ALIAS: $LATEST

Automatically exclude package level includes

Previous versions of the plugin exclude everything in the service folder and include the .warmup folder. This caused that any files that you include to the service level were also included in the plugin specially if you include ancestor folders (like ../**) From v5, all service level include are automatically excluded from the plugin. You still override this behaviour using the package option.

Removed shorthand

Previous versions of the plugin support replacing the configuration by a boolean, a string representing a stage or an array of strings representing a lsit of stages. From v5, this is not supported anymore. The enabled option is equivalent.

custom:
  warmup: 'prod'

is the same as

custom:
  warmup:
    default: # Name of the warmer, see above
      enabed: 'prod'

Removed legacy options

The following legacy options have been completely removed:

  • default Has been renamed to enabled
  • schedule schedule: rate(5 minutes) is equivalent to events: - schedule: rate(5 minutes).
  • source Has been renamed to payload
  • sourceRaw Has been renamed to payloadRaw

Automatically creates a role for the lambda

If no role is provided in the custom.warmup config, a default role with minimal permissions is created for each warmer. See "Permissions" section

Support Tracing

If tracing is enabled at the provider level or at the warmer config level, the X-Ray client is automatically installed and X-Ray tracing is enabled.

Add a 1 second connect timeout to the AWS SDK

See the "Networking" section for more details.

Cost

You can check the Lambda pricing and CloudWatch pricing or can use the AWS Lambda Pricing Calculator to estimate the monthly cost

Example

If you have a single warmer and want to warm 10 functions, each with memorySize = 1024 and duration = 10, using the default settings (and we ignore the free tier):

  • WarmUp: runs 8640 times per month = $0.18
  • 10 warm lambdas: each invoked 8640 times per month = $14.4
  • Total = $14.58

CloudWatch costs are not in this example because they are very low.

Contribute

Help us making this plugin better and future-proof.

  • Clone the code
  • Install the dependencies with npm install
  • Create a feature branch git checkout -b new_feature
  • Add your code and add tests if you implement a new feature
  • Validate your changes npm run lint and npm test (or npm run test-with-coverage)

License

This software is released under the MIT license. See the license file for more details.

Acknowledgements

Thanks to Fidel who initially developed this plugin.

view on Github

Latest commit b2f54ec on Sep 24, 2017

New to serverless?

To get started, pop open your terminal & run:

npm install serverless -g