Extract flame graph for Nodejs directly from an AWS lambda function. Lambda flame is a custom Nodejs runtime that collects and creates a flame graph of your lambda function process.
The first thing is to deploy the Serverless application into your account. You can do it via the AWS console: lambda-flame, or directly into Cloudformation: (make sure you specify the latest version)
lambdaflame:
Type: AWS::Serverless::Application
Properties:
Location:
ApplicationId: arn:aws:serverlessrepo:ap-southeast-1:164102481775:applications/lambda-flame
SemanticVersion: 1.0.0
After the application has been deployed go to the lambda console, click on layers and copy the lambda-flame layer arn;
You can now paste the layer in your SAM/Cloudformation template:
AWSTemplateFormatVersion: "2010-09-09"
Transform: "AWS::Serverless-2016-10-31"
Description: "Lambda flame test template"
Resources:
ExampleFunction:
Type: AWS::Serverless::Function
Properties:
Runtime: provided // Runtime must be set to provided
Timeout: 30
MemorySize: 1024
CodeUri: ./
Handler: handler.handler
Environment:
Variables:
LAMBDA_FLAME_DEST_BUCKET: !Ref FlameBucket
LAMBDA_FLAME_DEBUG: ALL
Policies:
- AWSLambdaBasicExecutionRole
- S3WritePolicy: // Or simply s3:PutObject
BucketName: !Ref FlameBucket
Layers:
- !Sub arn:aws:lambda:ap-southeast-1:{AWS::AccountId}:layer:lambda-flame:13 // Layer arn from the Lambda Flame application
Events:
HTTP:
Type: Api
Properties:
Path: /
Method: Get
FlameBucket:
Type: AWS::S3::Bucket
Properties:
BucketName: lambda-flame-graph-test
Paste the arn in your serverless.yaml
:
service: lambda-flame-test-sls
provider:
name: aws
stackName: lambda-flame-test-sls
region: ap-southeast-1
iamRoleStatements:
- Effect: Allow
Action:
- s3:PutObject
Resource:
- arn:aws:s3:::lambda-flame-graph-test
- arn:aws:s3:::lambda-flame-graph-test/*
package:
individually: true
functions:
exampleFunction:
memorySize: 1024
timeout: 30
handler: handler.handler
runtime: provided
environment:
LAMBDA_FLAME_DEST_BUCKET: lambda-flame-graph-test
LAMBDA_FLAME_DEBUG: ALL
layers:
- !Sub arn:aws:lambda:ap-southeast-1:${AWS::AccountId}:layer:lambda-flame:13
events:
- http:
method: get
path: /
resources:
Resources:
FlameBucket:
Type: AWS::S3::Bucket
Properties:
BucketName: lambda-flame-graph-test
If you are using the webpack plugin, the custom runtime option will not work for versions < 5.4.0
Just add the allowCustomRuntime: true
option to your function:
exampleFunction:
handler: handler.handler
runtime: provided
allowCustomRuntime: true
layers:
- !Sub arn:aws:lambda:ap-southeast-1:${AWS::AccountId}:layer:lambda-flame:13
...
- The
Runtime
property must be set toprovided
- All the result of the v8 profiling and the flamegraph creation will be sent to an S3 bucket of your choice. You must specify a valid S3 Bucket name.
You can use any bucket you want and one bucket for all your functions,
lambda flame will name space all the output files under
s3://<your-bucket-name>/lambda-flame/<function-name>/<epoch timestamp>/
- In order to send the profiling output and flame graph files to your S3 bucket make sure you have the required write permissions
s3:PutObject
- The v8 profile output processing happens outside your handler invocation therefore it won't impact your function profiling. However, is a CPU bound operation, if you want to make it a bit faster just temporarily add more memory
- This tool (for the time being) is not meant to run in a production environment, use it as a debugging purposes only
After the invocation of your function has terminated, you can navigate to your S3 Bucket under s3://<your-bucket-name>/lambda-flame/<function-name>/<epoch timestamp>/
and open the flamegraph.html
You can set all the parameters via environmental variables
Property | Type | Description | Default |
---|---|---|---|
LAMBDA_FLAME_DEST_BUCKET | String |
Valid S3 bucket name required | undefined |
LAMBDA_FLAME_DEBUG | String |
Activate debugging logs, they will appear in Cloudwatch, leave it empty if you don't want them | ALL |
There are a lot of new features I want to add, feel free to contribute, open issues and share your ideas π.
- Provide multiple runtimes other than nodejs (planning to support Go)
- Add more control over via configuration
- Automate builds of Nodejs runtimes
- Possibility to capture a heap dump
- Dramatically improve flame graph visualization and interactivity
This tool (for the time being) is partially using a modified version flamebearer all the credits goes to the author of this library. In the future I am planning to build a custom visualization similar to Clinic flame.