Lambda

De Banane Atomic
Révision datée du 3 juin 2024 à 12:43 par Nicolas (discussion | contributions) (→‎Call a lambda from code)
(diff) ← Version précédente | Voir la version actuelle (diff) | Version suivante → (diff)
Aller à la navigationAller à la recherche

Links

Description

Serverless service / Function as a service allowing to run code without having to worry about underlying hardware and OS.
Event driven: the lambda is triggered by an event.
Pay only for what you use: per request and based on the duration of the code execution.

Use cases

  • Data transformation (Kinesis Data Stream as input)
  • File processing (when uploaded to S3 bucket)
  • Website backend microservice
  • Scheduled tasks

Bad use cases

  • Long running processes (timeout after 15mn)
  • Constant workload (no scalability and high cost)
  • Large code base (needed to be loaded at startup)
  • State management (lambda are stateless)

Anti-patterns

  • Monolithic function
    • increase package size
    • hard to enforce least privilege permissions
    • hard to upgrade, maintain and test
  • Recursion
    • endless loop
  • Orchestration
    • avoid complex workflow logic
    • ties lambda with other systems
    • instead consider AWS Step Functions or EventBridge
  • Chaining (synchronously invoke another lambda)
    • instead use EventBridge or QueueService
  • Waiting (synchronously call services or databases)
    • instead use asynchronous calls

Runtime

  • OS
  • Libraries
  • Programming language (.NET, Node.js, Python, Go, Ruby, Java)

Environnement variables

DOTNET_STARTUP_HOOKS ex: path to an assembly to inject logging

Wrapper scripts

Execute the wrapper on top of the runtime and the lambda function.

  • run shell commands and binaries

Use AWS_LAMBDA_EXEC_WRAPPER to point to your wrapper script.

Custom runtime

Provide your custom runtime.

  • unsupported programming language

Handler (entry point)

Method responsible for processing input events.

synchronous execution result returned to the calling app
asynchronous execution result sent to the configured destination otherwise lost

Configuration

Setting Description
Memory from 128 MB (default) to 10 GB
vCPU 1'769 MB = 1 vCPU, 10 GB = 6 vCPU, for single-threaded code allocated more than 1'769 MB is useless. Compare ARM and x86.
timeout from 1s to 15mn, default to 3s
VPC
  • by default no VPC access configured → Amazon VPC → internet access
  • if VPC access is configured → access to VPC resources → no internet connection
    • use an Hyperplane ENI to add internet connection in addition of a VPC access
Ephemeral storage /tmp from 512 MB (default) to 10 GB
Execution role grants permissions to access AWS services
Resource-based policy define what others are allowed to do to the function
Environment variables
  • Key-value pair of strings
  • Stored in the configuration for a Lambda function version
  • Accessible within the Lambda code function
Logging format (text or JSON), level, group

Storage

Directories

Path Description
/var/task Lambda package (function files)
/tmp Temporary storage with write access
/opt Lambda layer (layer files)

Elastic File System (EFS)

  • Persistent
  • Serverless, scalable
  • Support concurrent access
  • Require VPC access

S3

  • Persistent
  • No VPC required
  • Accessed via API

IAM entities

Type Description
User individual with unique credentials
Role entity for temporary access (access to a db)
Policy permission for access to resources (apply to a user or a role to read from a specific db table)
User group logical collection of users for easier management

Synchronous execution

Invocation result

Result content Description
Status code 200 success, 4xx or 5xx error
Executed version
Log result last 4 KB of the execution log in base64 encoding
Function error only if an error occured. Code error, timeout, invalid JSON output
Response body payload

Asynchronous invocation

When a Lambda function is invoked asynchronously:

  • an event is placed in its internal queue
  • it returns a success response (202 Accepted) without additional information to the invoker
  • a separate process reads events from the queue and sends them to the lambda function
  • the result (Invocation Record) and the error are sent to the destinations

Destination

Allow to route asynchronous function results as an execution record to a destination resource.
Available destinations:

Event Source (trigger)

Lambda has builtin functionalities to poll the event source with batching and filtering functionality.
Event source mapping polls and gathers records before it will initialize a Lambda function and send a batch of events to it for processing.

Error handling

If the invocation errors, then it will try the same invocation again after a minute.
After X fails that exceeds the configured retry attempts (0, 1 or 2), the failed invocation goes to:

  • a dead-letter queue
  • a failure destination

CLI

Ps.svg
# async call (invoke via queue)
aws lambda invoke \
--profile [Profile] \
--function-name [Function] \
--invocation-type Event \
--payload $([Convert]::ToBase64String([Text.Encoding]::UTF8.GetBytes('{ "key": "value" }'))) \
response.json

Optimization

Make the most of warm start

The initialization of the function is run only during the cold start.
During the warm start the environment is thawed and only the function handler is executed. Use function initialization to: (outside of the handler function)

  • load external libraries
  • open db connection
  • load configuration and secrets
  • use the ephemeral storage, such as a local cache for files

SnapStart

Only for Java runtime

Lambda initializes the function at publish rather than during cold-start invocation.
If not invoked, snapshots are deleted after 14 days.
Limitations:

  • cannot be used with provisioned concurrency
  • supports limited runtimes, architecture and ephemeral storage

Ensure to (with post-snapshot hook):

  • reestablish network connections
  • have preloaded data up to date
  • take care of uniqueness (generation of random numbers or GUID), each function starts with a clone of the snapshot

Concurrency

Concurrency type Description
Account concurrency max number of concurrent instances across all functions (AWS → Lambda → Dashboard)
Reserved concurrency upper limit of a function's concurrency
Burst concurrency increase quota, default to 1'000 instances every 10 s
Provisioned concurrency initialize execution env ahead of time. Application Auto Scaling can be used to automatically adjust provisioned concurrency

Setting up your .NET development environment

Ps.svg
# install the VS project template
dotnet new install Amazon.Lambda.Templates

# install the command line tools
dotnet tool install -g Amazon.Lambda.Tools

Install the AWS Toolkit for Visual Studio extension.

AWS Lambda application types

AWS Lambda application type Description
Class library
Executable assembly
AWS Serverless Application ASP.NET application hosted in the AWS environment. It has a handful of additional dependencies that make it interoperate with AWS runtime.

C# class library

You provide Lambda with information about your function's handler in the form of a handler string: ASSEMBLY::TYPE::METHOD

  • ASSEMBLY is the name of the .NET assembly file for your application. If you use the Amazon.Lambda.Tools CLI to build your application and you don't set the assembly name using the AssemblyName property in the .csproj file, then ASSEMBLY is simply the name of your .csproj file.
  • TYPE is the full name of the handler type, which consists of the Namespace and the ClassName.
  • METHOD is the name of the function handler method in your code.

ex: MyProject::MyNamespace.MyClass::MyFunctionHandler

Function.cs
// Assembly attribute to enable the Lambda function's JSON input to be converted into a .NET class.
[assembly: LambdaSerializer(typeof(Amazon.Lambda.Serialization.SystemTextJson.DefaultLambdaJsonSerializer))]

namespace MyProject;

public class Function
{
    public string FunctionHandler(string input, ILambdaContext context)
    {
        return input.ToUpper();
    }
}
aws-lambda-tools-defaults.json
{
  "Information": [ ],
  "profile": "MyAwsProfile",
  "region": "eu-central-1",
  "configuration": "Release",
  "function-architecture": "x86_64",
  "function-runtime": "dotnet8",
  "function-memory-size": 512,
  "function-timeout": 30,
  "function-handler": "MyProject::MyProject.Function::FunctionHandler"
}

On VS, if the extension AWS Toolkit is installed, you have the AWS .NET Mock Lambda Test Tool available which allows you to debug locally.

Properties/launchSettings.json
{
  "profiles": {
    "Mock Lambda Test Tool": {
      "commandName": "Executable",
      "commandLineArgs": "--port 5050",
      "workingDirectory": ".\\bin\\$(Configuration)\\net8.0",
      "executablePath": "%USERPROFILE%\\.dotnet\\tools\\dotnet-lambda-test-tool-8.0.exe"
    }
  }
}

C# executable assembly

Using the C# 9's top-level statements feature, you generate an executable assembly which will be run by the Lambda. You provide Lambda only with the name of the executable assembly to run.

Cs.svg
var handler = async (string argument1, ILambdaContext context) => { };

// bootstrap the Lambda runtime and pass it the handler method
await LambdaBootstrapBuilder.Create(handler, new DefaultLambdaJsonSerializer()).Build().RunAsync();

To debug locally you have to create the debug configuration file for you lambda.

Properties\launchSettings.json
{
  "profiles": {
    "Mock Lambda Test Tool": {
      "commandName": "Executable",
      "commandLineArgs": "--port 5050",
      "executablePath": "%USERPROFILE%\\.dotnet\\tools\\dotnet-lambda-test-tool-8.0.exe",
      "workingDirectory": ".\\bin\\$(Configuration)\\net8.0",
      "environmentVariables": {
        "AWS_LAMBDA_RUNTIME_API": "localhost:5050",
        "AWS_PROFILE": "MyProfile",
        "AWS_REGION": "us-east-1"
      }
    }
  }
}

Handle Enums in the input arguments of .NET AWS Lambda Functions

Function.cs
// create a serializer to handle enums
public class LambdaEnumSerializer : DefaultLambdaJsonSerializer
{
    public LambdaEnumSerializer()
        : base(options => options.Converters.Add(new JsonStringEnumConverter())) { }
}

// then use it
[assembly: LambdaSerializer(typeof(LambdaEnumSerializer))]

// OR use a stream as input argument and deserialize it handling enums
public string FunctionHandler(Stream input, ILambdaContext context)
{
    var options = new JsonSerializerOptions
    {
        PropertyNameCaseInsensitive = true,
        Converters = { new JsonStringEnumConverter() }
    };

    JsonSerializer.Deserialize<MyClass>(input, options);
}

Logging

Lambda pushes logs to the CloudWatch Logs service under /aws/lambda/[FunctionName].
Permissions (function's execution role, ex: AWSLambdaBasicExecutionRole policy):

  • create CloudWatch groups
  • create CloudWatch streams
  • write log events
  • Logs are organized into groups (1 group per function)
    • each group contains a collection of streams (1 stream per function instance) YYYY/MM/DD[Function version][Execution env GUID]
Cs.svg
LambdaLogger.Log("Message");

Powertools

Allows Logging, Tracing, Metrics, Parameters.

Logging

Use structured logging.

Cs.svg
// nuget package: AWS.Lambda.Powertools.Logging
public class Function
{
    public Function()
    {
        Logger.UseFormatter(new CustomLogFormatter());
    }

    // When debugging in non-production environments, you can instruct Logger to log the incoming event
    [Logging(LogEvent = true)]
    // define the case of property names (default: snake)
    [Logging(LoggerOutputCase = LoggerOutputCase.CamelCase)]
    // Delete all custom keys. Due to Lambda Execution Context reuse, this means that custom keys can be persisted across invocations.
    [Logging(ClearState = true)]
    public string FunctionHandler(string input, ILambdaContext context)
    {
        Logger.LogInformation($"Input: '{input}'");

        var lookupInfo = new Dictionary<string, object>
        {
            { "LookupId", 10 }
        };  
        Logger.AppendKeys(lookupInfo);

        return input;
    }
}
CustomLogFormatter.cs
public class CustomLogFormatter : ILogFormatter
{
    public object FormatLogEntry(LogEntry logEntry)
    {
        return new
        {
            Message = logEntry.Message,
            Service = logEntry.Service,
            CorrelationIds = new 
            {
                AwsRequestId = logEntry.LambdaContext?.AwsRequestId,
                XRayTraceId = logEntry.XRayTraceId,
                CorrelationId = logEntry.CorrelationId
            },
            LambdaFunction = new
            {
                Name = logEntry.LambdaContext?.FunctionName,
                Arn = logEntry.LambdaContext?.InvokedFunctionArn,
                MemoryLimitInMB = logEntry.LambdaContext?.MemoryLimitInMB,
                Version = logEntry.LambdaContext?.FunctionVersion,
                ColdStart = logEntry.ColdStart,
            },
            Level = logEntry.Level.ToString(),
            Timestamp = logEntry.Timestamp.ToString("o"),
            Logger = new
            {
                Name = logEntry.Name,
                SampleRate = logEntry.SamplingRate
            },
        };
    }
}

Priority of log level settings in Powertools for AWS Lambda:

  1. AWS_LAMBDA_LOG_LEVEL environment variable
  2. Setting the log level in code using [Logging(LogLevel = )]
  3. POWERTOOLS_LOG_LEVEL environment variable

Parameters

Fetch values from Secrets Manager, Parameter Store, DynamoDB and App Config

Cs.svg
public class Function
{
    public string FunctionHandler(string input, ILambdaContext context)
    {
        // Get Secrets Provider instance (Secrets Manager)
        ISecretsProvider secretsProvider = ParametersManager.SecretsProvider;

        // Retrieve a single secret
        string? value = await secretsProvider.GetAsync("/my/secret");
    }
}

Deployment

Ps.svg
dotnet lambda deploy-function [AssemblyName] --profile [Profile]

Permissions

WatchLogs

EventBridge

Call a lambda from code

Cs.svg
// nuget: AWSSDK.Lambda
var jsonSerializerOptions = new JsonSerializerOptions
{
    PropertyNamingPolicy = JsonNamingPolicy.CamelCase,
    Converters = { new JsonStringEnumConverter() }
};

var amazonLambdaConfig = new AmazonLambdaConfig
{
    Profile = new Profile("MyProfile"),
    RegionEndpoint = RegionEndpoint.USEast1
};
var amazonLambdaClient = new AmazonLambdaClient(amazonLambdaConfig);

var request = new InvokeRequest
{
    FunctionName = "MyFunction",
    Payload = JsonSerializer.Serialize(myObject, jsonSerializerOptions),
    LogType = LogType.Tail,
    InvocationType = InvocationType.Event // async call (invoke with queue)
    InvocationType = InvocationType.RequestResponse // sync call (invoke directly and wait for the response)
};
var response = await amazonLambdaClient.InvokeAsync(request);

if(response.HttpStatusCode == System.Net.HttpStatusCode.OK)
{
    var payload = Encoding.ASCII.GetString(response.Payload.ToArray()); // to debug only
    var result = JsonSerializer.Deserialize<MyClass>(response.Payload, jsonSerializerOptions);
}

Templates

Ps.svg
# list your installed Amazon Lambda Templates
dotnet new list lambda

# install Amazon Lambda Templates
dotnet new install Amazon.Lambda.Templates
Template Description
Lambda ASP.NET Core Web API Web API
Lambda ASP.NET Core Web Application with Razor Pages Razor MVC Web Application
Lambda Serverless with Powertools for AWS Lambda (.NET) Powertools is a developer toolkit to implement Serverless best practices and increase developer velocity

Lambda ASP.NET Core Web API

Use a CloudFormation stack and an S3 to upload the application (zip) and the CloudFormation template file.

  • CloudFormation stack
  • IAM Role (Lambda Role)
  • IAM Role (Lambda Role)
  • Lambda Function (Web API)
  • VPC
  • API Gateway (REST)

Errors

Could not find the specified handler assembly with the file name LambdaTest

The Lambda on AWS has wrongly set the handler to LambdaTest.
AWS - Lambda - Functions - select your function - Code tab - Runtime settings - Edit - change the Handler

Your function doesn't have permission to write to Amazon CloudWatch Logs

  1. AWS → IAM → Roles → [the roles used by your lambda] → Persissions policies → Add permission → Create inline policy
  2. Service = CloudWatch Logs
  3. All CloudWatch Logs actions (logs:*)
  4. All Resources