Categorias
PHP Programação

A bref AWS PHP story – Part 3

We are starting Part 3 of the Series "A bref AWS PHP history". You can check Part 1, where I presented the PHP language as a reliable and good alternative for Serverless applications and Part 2 where we see the usage of CDK features in favour of a faithful CI/CD.

Part 3 is to show the upgrade path to Bref 2 and to achieve more coverage of the AWS resources. We will use DynamoDB, a powerful database for serverless architectures.

Some of those topics seem straightforward to some people, but I would like to avoid guessing that this is known to the audience since I have experienced some PHP developers struggling to put all these together for the first time due to the paradigm change. It should be fun.

Table of contents:

  1. What else are we doing?
  2. Describing more AWS services - Adding a DynamoDB table
  3. Bref upgrade
  4. Testing CDK
  5. PHP and AWS Services
  6. Wrap-up

What else are we doing?

In this section, we'll explore additional functionalities and enhancements to our serverless application. Building upon the foundation laid in Part 2, we'll introduce new features and integrations to further extend the capabilities of our AWS PHP application.

The Part 2 uses the result of the Fibonacci of a provided integer or a random integer from 400 to 1000 (to get a good image and not to overflow integer). This integer is the number of pixels of an image from the bucket and an arbitrary request metadata we are creating. If the image does not exist, the lambda will fetch a random image from the web with that number of pixels, save it and generate the metadata.

The computing complexity is irrelevant because it could be very complex logic or very simple, and the topics we are discussing in this part of the series will use the same design.

The lambda will now search the metadata in a DynamoDB table, saving the metadata when it does not exist. DynamoDB is largely used in Lambda code.

Get the part-3 source-code on GitHub and the diff from part-2.

Describing more AWS services - Adding a DynamoDB Table

DynamoDB plays a crucial role in serverless architectures, offering scalable and high-performance NoSQL database capabilities. In this section, we'll delve into the process of integrating DynamoDB into our AWS CDK stack, expanding our application's data storage and retrieval capabilities.

DynamoDB is a fully managed NoSQL database service provided by AWS, offering seamless integration with other AWS services, automatic scaling, and built-in security features. Its scalability, low latency, and flexible data model make it well-suited for serverless architectures and applications with varying throughput requirements.

    const table = new Table(this, TableName, {
      partitionKey: { name: 'PK', type: AttributeType.STRING },
      sortKey: { name: 'SK', type: AttributeType.STRING },
      removalPolicy: RemovalPolicy.DESTROY,
      tableName: TableName,
    });

Following the same principles for creating other AWS resources, we utilize the AWS CDK to define a DynamoDB table within our stack. Let's dive into the key parameters of the Table constructor:

  • partitionKey: This parameter defines the primary key attribute for the DynamoDB table, used to distribute items across partitions for scalability. In our example, { name: 'PK', type: AttributeType.STRING } specifies a partition key named 'PK' with a string type. The naming convention ('PK') is arbitrary and can be tailored to suit your application's needs.
  • sortKey: For tables requiring a composite primary key (partition key and sort key), the sortKey parameter comes into play. Here, { name: 'SK', type: AttributeType.STRING } defines a sort key named 'SK' with a string type. Like the partition key, the name and type of the sort key can be customized based on your data model.
  • removalPolicy: This parameter determines the behaviour of the DynamoDB table when the CloudFormation stack is deleted. By setting RemovalPolicy.DESTROY, we specify that the table should be deleted (destroyed) along with the stack. Alternatively, you can opt for RemovalPolicy.RETAIN to preserve the table post-stack deletion, which may be useful for retaining data.

By decoupling configuration from implementation, we adhere to SOLID principles, ensuring cleaner and more robust code. This approach fosters flexibility, allowing our code to seamlessly adapt to changes, such as modifications to the table name while maintaining its functionality.

The implementation code is aware that the name will come from an environment variable and will work with that (yes, if you think that test will be easy to write, you are right):

    const lambdaEnvironment = {
      TableName,
      TableArn: table.tableArn,
      BucketName: brefBucket.bucketName,
    };

Bref Upgrade

Bref, the PHP runtime for AWS Lambda, continually evolves to provide developers with the latest features and optimizations. In this section, we'll discuss the upgrade to Bref 2.0 and explore how it enhances the deployment process and performance of our serverless PHP applications.

In this section, we're upgrading our usage of Bref, a PHP runtime for AWS Lambda, to version 2.0. Bref simplifies the deployment of PHP applications to AWS Lambda, enabling us to run PHP code serverlessly.

The upgrade involves modifying our AWS CDK code to utilize the new features and improvements introduced in Bref 2.0. One notable improvement is the automatic selection of the latest layer of the PHP version, which simplifies the deployment process and ensures that our Lambda functions run on the most up-to-date PHP environment available.

  const getLambda = new PhpFunction(this, ${stackPrefix}${functionName}, {
    handler: 'get.php',
    phpVersion: '8.3',
    runtime: Runtime.PROVIDED_AL2,
    code: packagePhpCode(join(__dirname, ../assets/get), {
      exclude: ['test', 'tests'],
    }),
    functionName,
    environment: lambdaEnvironment,
  });
  • `PhpFunction` Constructor: We're using the `PhpFunction` constructor provided by Bref to define our Lambda function. This constructor allows us to specify parameters such as the handler file, PHP version, runtime, code location, function name, and environment variables.
  • `handler`: Specifies the entry point file for our Lambda function, where the execution starts.
  • `phpVersion`: Defines the PHP version to be used by the Lambda function. In this case, we're using PHP version 8.3.
  • `runtime`: Indicates the Lambda runtime environment. Here, `Runtime.PROVIDED_AL2` signifies the use of the Amazon Linux 2 operating system.
  • `code`: Specifies the location of the PHP code to be deployed to Lambda.
  • `functionName`: Sets the name of the Lambda function.
  • `environment`: Allows us to define environment variables required by the Lambda function, such as database connection strings or configuration settings.

By upgrading to Bref 2.0 and configuring our Lambda function accordingly, we ensure compatibility with the latest enhancements and optimizations provided by Bref, thereby improving the performance and reliability of our serverless PHP applications on AWS Lambda.

Testing CDK

Ensuring the correctness and reliability of our AWS CDK infrastructure is crucial for maintaining a robust serverless architecture. In this section, we'll delve into testing our CDK resources, focusing on the DynamoDB table we added in the previous section.

As described earlier, we utilized the AWS CDK to provision a DynamoDB table within our serverless stack. Now, let's ensure that the table is configured correctly and behaves as expected by writing tests using the CDK's testing framework.

First, let's revisit how we added the DynamoDB table:

const table = new Table(this, TableName, {
  partitionKey: { name: 'PK', type: AttributeType.STRING },
  sortKey: { name: 'SK', type: AttributeType.STRING },
  removalPolicy: RemovalPolicy.DESTROY,
  tableName: TableName,
});

In this code snippet, we define a DynamoDB table with specified attributes such as partition key, sort key, removal policy, and table name. Now, to ensure that this table is created with the correct configuration, we'll write tests using CDK's testing constructs.

Check the following thest:

test('Should have DynamoDB', () => {
  expectCDK(stack).to(
    haveResource(
      'AWS::DynamoDB::Table',
      {
        "DeletionPolicy": "Delete",
        "Properties": {
          "AttributeDefinitions": [
            {
              "AttributeName": "PK",
              "AttributeType": "S",
            },
            {
              "AttributeName": "SK",
              "AttributeType": "S",
            },
          ],
          "KeySchema": [
            {
              "AttributeName": "PK",
              "KeyType": "HASH",
            },
            {
              "AttributeName": "SK",
              "KeyType": "RANGE",
            },
          ],
          "ProvisionedThroughput": {
            "ReadCapacityUnits": 5,
            "WriteCapacityUnits": 5,
          },
          "TableName": "BrefStory-table",
        },
        "Type": "AWS::DynamoDB::Table",
        "UpdateReplacePolicy": "Delete",
      },
      ResourcePart.CompleteDefinition,
    )
  );
});

This test ensures that the DynamoDB table is created with the correct attribute definitions, key schema, provisioned throughput, table name, and other properties specified during its creation. By writing such tests, we validate that our CDK infrastructure is provisioned accurately and functions as intended.

PHP and AWS Services

Leveraging PHP in a serverless environment opens up new possibilities for interacting with AWS services. In this section, we'll examine how PHP code seamlessly integrates with various AWS services, following best practices for maintaining clean and modular code architecture.

This is the part where we have fewer serverless needs impacting the code, as the PHP code will follow the same logic we might be using to communicate with AWS services on any other platform overall (there are always some specific use cases).

The reuse of the same existing logic is excellent. It leverages the decision to keep using PHP when moving that workload to Serverless, as the bulk of the knowledge and already proven code would remain as-is. We may escape the trap of classifying that PHP code as legacy as if it should be avoided, terminated or halted.

As a side note, a few external layers of our software architecture are touched if a good software architecture was applied before. Therefore, during the implementation of this architectural change, it should be quick to realise how beneficial and time-saving it is to have a well-architectured application with a balanced decision for patterns, principles, and designs to be applied, ultimately giving flexibility to the application and its features.

The handler is simplified now and should accommodate everything to a class in the direction of following SRP, a principle that we are bringing to the code during the code bites:

Applications, domains, infrastructure, etc

Our `PicsumPhotoService` is still orchestrating the business logic. The Single Responsibility Principle and Inversion of Control are applied. We are injecting the specialized services in the constructor:

// readonly class PicsumPhotoService
    public function __construct(
        private HttpClientInterface $httpClient,
        private ImageStorageService $storageService,
        private ImageRepository $repository,
    )
    {
    }

Each specialized service has all its dependencies injected in the constructor as well. We can see the factory instantiation:

    public static function createPicsumPhotoService(): PicsumPhotoService
    {
        return new PicsumPhotoService(
            HttpClient::create(),
            new S3ImageService(
                new S3Client(),
                getenv('BucketName'),
            ),
            new DynamoDbImageRepository(
                new DynamoDbClient(),
                getenv('TableName'),
            ),
        );
    }

The `ImageStorageService` will handle all image operations, connecting to the AWS Service when appropriate and observing business logic details. This is a slim interface:

interface ImageStorageService
{
    public function getImageFromBucket(int $imagePixels): ?array;

    public function saveImage(int $imagePixels, mixed $fetchedImage): void;

    public function createAndPutMetadata(int $imagePixels, array $metadata): PutObjectOutput;
}

Instead of `: PutObjectOutput`, usually we would return a domain object, to not couple the interface with implementation details of using S3 Services, but for simplicity, I did not create a domain object here. It would be preferable though.

The `ImageRepository` will handle all metadata operations. It will save into a repository and observe logic details as well. Following the same principles, this is a slim interface:

interface ImageRepository
{
    public function findImage(int $imagePixels): ImageMetadataItem;

    public function addImageMetadata(ImageMetadataItem $imageMetadataItem): PutItemOutput;
}

The `ImageMetadataItem` is a representation of one of the domain objects we have in our codebase.

readonly class ImageMetadataItem
{
    public function __construct(public int $imagePixels, public array $metadata)
    {
    }

    public function toDynamoDbItem(): array
    {
        return [
            'PK' => new AttributeValue(['S' => 'IMAGE']),
            'SK' => new AttributeValue(['S' => "PIXELS#{$this->imagePixels}"]),
            'pixels' => new AttributeValue(['N' => "{$this->imagePixels}"]),
            'metadata' => new AttributeValue(['S' => json_encode($this->metadata)]),
            ...ConvertToDynamoDb::item($this->metadata),
        ];
    }

    /**
     * @param array $item
     */
    public static function fromDynamoDb(array $item): static
    {
        return new static(
            (int) $item['pixels']->getN(),
            (array) json_decode($item['metadata']->getS()),
        );
    }
}

If you check the implementation details, it operates transparently with all the services, business logic and AWS Services without any high couple with them. There are two utility functions:

  • toDynamoDbItem: to transform the object into a valid DynamoDb Item to be added
  • fromDynamoDb: to perform the opposite operation, transforming a DynamoDb Item into a domain object

The scope of the operation is very clear and does not bring the domain into dependency on those services, as the domain object can be used independently. It does not block any other way of dealing with it, giving the usage with other types of services, such as different databases or APIs. This is very important to the maintainability of the application without sacrificing the ease of readiness as it keeps the context of the utilities in the right place.

If you check all PHP code carefully, Bref is such a great abstraction layer that, removing the code from the handler file, any other line of code can be used as a lambda or a web application interchangeably without changing any line of code. This is very powerful, as you can imagine how you can leverage and migrate some of the existing code to lambda by just creating a handler that will trigger your existing code, if the code is well structured.

Wrap-up

It would be simple like that. Check more details in the source code, install it and try it yourself. This project is ready to:

  • Extend lambda function using Bref
  • Upgrade to use Bref 2.0
  • Create a DynamoDB table
  • Test the stack Cloudformation code
  • Separate the PHP logic
  • Have PHP communicating with AWS Services

Links:

Categorias
Tropeçando

Tropeçando 110

Enabling the Optimal Serverless Platform Team — CDK and Team Topologies

Serverless, and related technologies, have enabled teams to move faster, reduce total cost of ownership and overall empowered developers to have greater ownership of the systems they build. However, Serverless is not a silver bullet — there is an organisational side that’s key to unlock the full benefits of Cloud.

Restructuring a Laravel Controller using Services, Events, Jobs, Actions, and more

A simple but nice walk-though about code decoupling.

The Serverless Server

I'm Will Jordan, and I work on SRE at Fly.io. We transmogrify Docker containers into lightweight micro-VMs and run them on our own hardware in racks around the world, so your apps can run close to your users. Check it out—your app can be up and running in minutes. This is a post about how services like ours are structured, and, in particular, what the term "serverless" has come to mean to me.

Keep Cognitive Complexity Low with PHPStan

What is cognitive complexity? It's the amount of information we have to hold in our heads simultaneously to understand the code. The more indents, continue, break, nested foreach, and if/else branches, the harder is code to read.

You can use PHPStan rules to decrease the cognitive complexity of your codebase. This brings matuiry to your application and a more maintainable code.

How to release PHP 8.1 and 7.2 package in the Same Repository

Some steps to release a package in more than one version, to allow compatibility for different PHP runtimes.

Categorias
PHP Programação

A bref AWS PHP story – Part 2

We are starting Part 2 of the Series "A bref AWS PHP history". You can check Part 1, where I presented the PHP language as a reliable and good alternative for Serverless applications.

Part 2 is to show how CDK will describe more AWS resource dependencies; how policies and roles are involved in this process; how to test if they are applied as expected; and how PHP services will use those resources.

Some of those topics seem straightforward to some people, but I would like to avoid guessing that this is known to the audience since I have experienced some PHP developers struggling to put all these together for the first time due to the paradigm change. It should be fun.

Table of contents:

  1. What else are we doing?
  2. Describing more AWS services - Adding an S3 bucket
  3. Services permissions
  4. Testing CDK
  5. PHP and AWS Services
    1. Handlers
    2. Application, Domain, Infrastructure, etc
  6. Wrap-up
  7. P.S.: Stats

What else are we doing?

The Part 1 function was returning a Fibonacci result from an int. Very simple. We will keep it simple for now to focus on putting the PHP code into a lambda and allowing PHP code to interact with AWS Services.

The computing complexity is irrelevant because it could be very complex logic or very simple, and the topics we are discussing in this part of the series will use the same design.

The lambda will now use the result of the Fibonacci of a provided integer or a random integer from 400 to 1000 (to get a good image and not to overflow integer). This integer is the number of pixels of an image from the bucket and an arbitrary request metadata we are creating. If the image does not exist, the lambda will fetch a random image from the web with that number of pixels, save it and generate the metadata.

Get the part-2 source-code on GitHub and the diff from part-1.

Describing more AWS services - Adding an S3 bucket

S3 buckets are simple yet compelling services for multipurpose workloads. It will be added to the series as a basic storage mechanism. The lambda function, now called GetFibonacciImage function, will need some permissions to manage the bucket.

Starting from the bucket definition, CDK give fantastic constructs, and it goes like this:

cdk-stack.ts

    const brefBucket = new Bucket(this, `${stackPrefix}Bucket`, {
      autoDeleteObjects: true,
      removalPolicy: RemovalPolicy.DESTROY,
    });

By default, buckets will not be deleted during a CDK destroy because they need to be empty. So you will have a hanging bucket in your account. I don't want to keep those contents if the lambda no longer exists. Then autoDeleteObjects and removalPolicy options are selected to enable the destruction of the buckets and their contents if I execute a stack destroy.

We want to decouple the configuration from the implementation to have a more SOLID code. That way, we avoid hard-coded configuration, making our code cleaner and more robust. Then, the code is ready to work, no matter the bucket name.

The implementation code is aware that the name will come from an environment variable and will work with that (yes, if you think that test will be easy to write, you are right):

cdk-stack.ts

and

      environment: {
        BUCKET_NAME: brefBucket.bucketName,
      }

Services permissions

There is a Lambda Function and an S3 Bucket. The described use case determines that the lambda needs read and write permissions to the bucket. And nothing more. It is a good practice to give the minimum necessary permission to a resource:

cdk-stack.ts

    brefBucket.grantReadWrite(getLambda);

The result is a list of actions added to the policy recommended by AWS for operations requiring only read and write.

          Action: [
            "s3:GetObject*",
            "s3:GetBucket*",
            "s3:List*",
            "s3:DeleteObject*",
            "s3:PutObject",
            "s3:PutObjectLegalHold",
            "s3:PutObjectRetention",
            "s3:PutObjectTagging",
            "s3:PutObjectVersionTagging",
            "s3:Abort*",
          ],

Testing CDK

Testing is a great feature of CDK, and we can see how tests can verify our changes with npm t:

That there is a function

  const functionName = 'GetFibonacciImage';
  /* ... */
  it('Should have a lambda function to get fibonacci', () => {
    template.hasResourceProperties('AWS::Lambda::Function', {
      Layers: [Cdk.CdkStack.brefLayerFunctionArn],
      FunctionName: functionName,
    });
  });

And if only the permissions the lambda needs were granted:

  it('Should have a policy for S3', () => {
    template.hasResourceProperties('AWS::IAM::Policy', {
      PolicyName: Match.stringLikeRegexp(`^${stackPrefix}${functionName}ServiceRoleDefaultPolicy`),
      PolicyDocument: {
        Statement: [{
          Action: [
            "s3:GetObject*",
            "s3:GetBucket*",
            "s3:List*",
            "s3:DeleteObject*",
            "s3:PutObject",
            "s3:PutObjectLegalHold",
            "s3:PutObjectRetention",
            "s3:PutObjectTagging",
            "s3:PutObjectVersionTagging",
            "s3:Abort*",
          ],
        }],
      },
    });
  });

You may want to check cdk-stack.test.ts to see more details.

PHP and AWS Services

This is the part where we have fewer serverless needs impacting the code, as the PHP code will follow the same logic we might be using to communicate with AWS services on any other platform overall (there are always some specific use cases).

The reuse of the same existing logic is excellent. It leverages the decision to keep using PHP when moving that workload to Serverless, as the bulk of the knowledge and already proven code would remain as-is. We may escape the trap of classifying that PHP code as legacy as if it should be avoided, terminated or hated.

As a side note, a few external layers of our software architecture are touched if a good software architecture was applied before. Therefore, during the implementation of this architectural change, it should be quick to realise how beneficial and time-saving it is to have a well-architectured application with a balanced decision for patterns, principles, and designs to be applied, ultimately giving flexibility to the application and its features.

The handler is simplified now and should accommodate everything to a class in the direction of following SRP, a principle that we are bringing to the code during the code bites:

Handlers

php/handler/get.php

return function ($request, $context) {
    return \BrefStory\Application\ServiceFactory::createGetFibonacciImageHandler()
        ->handle($request, $context)
        ->toApiGatewayFormatV2();
};

To handle the request details, the Fibonacci code now lives in a proper event handler (implements Bref\Event\Handler).

php/src/Event/Handler/GetFibonacciImageHandler.php

    public function handle($event, Context $context): HttpResponse
    {
        $int = (int) (
            $event['queryStringParameters']['int'] ?? random_int(
                self::MIN_PIXELS_FOR_REASONABLE_IMAGE_AND_NOT_BIG_FIBONACCI,
                self::MAX_PIXELS_FOR_REASONABLE_IMAGE_AND_NOT_BIG_FIBONACCI
            )
        );

        $metadata = $this->photoService->getJpegImageFor($int);

        $responseBody = [
            'context' => $context,
            'now' => $this->dateTimeImmutable()->format('Y-m-d H:i:s'),
            'int' => $int,
            'fibonacci' => $this->fibonacci($int),
            'metadata' => $metadata,
        ];

        $response = new JsonResponse($responseBody);

        return new HttpResponse($response->getContent(), $response->headers->all());
    }

We would also like to start testing the PHP code. As the Event Handler might be a new layer (although very similar to widely used controllers), php/tests/unit/Event/Handler/GetFibonacciImageHandlerTest.php test class was created for that. The part-2 will only focus on this test class to avoid overloading with too many changes, but we would usually have test coverage for all the code in the repository.

Applications, domains, infrastructure, etc

Finally, we are inside the layers where we are most used to. To fit our purposes, the Event Handler will depend on and call an Application layer service that will orchestrate all the steps to fetch the image metadata.

php/src/Application/PicsumPhotoService.php#L34-L42

    public function getJpegImageFor(int $imagePixels): array
    {
        try {
            return $this->getImageFromBucket($imagePixels);
        } catch (NoSuchKeyException) {
            // do nothing
        }

        return $this->fetchAndSaveImageToBucket($imagePixels);
    }

The interesting thing to mention about using AWS Services is how simple S3Client is instantiated. There is a factory to create service:

php/src/Application/ServiceFactory.php#L22-L29

    public static function createPicsumPhotoService(): PicsumPhotoService
    {
        return new PicsumPhotoService(
            HttpClient::create(),
            new S3Client(),
            getenv('BUCKET_NAME'),
        );
    }
  • new S3Client is all we need because the environment will use AWS credentials, provided to lambda at execution time, as an assumed role that will carry the policies we defined in the CDK constructs stack, i.e., with read and write permissions to the bucket
  • getenv('BUCKET_NAME'), which is gracefully provided by CDK when creating our bucket with any dynamic name it pleases to

I asked ChatGPT about this class:

The PicsumPhotoService class seems to be following the Single Responsibility Principle (SRP) as it has only one responsibility, which is to provide methods for fetching and saving JPEG images from the Picsum website.

The class has methods to fetch the image from an S3 bucket, and if it's not available, fetches it from the Picsum website, saves it to the S3 bucket, and creates and puts metadata for the image in the S3 bucket.

The class has a clear separation of concerns, where the S3Client and HttpClientInterface are injected through the constructor, and the different functionalities are implemented in separate private methods. Additionally, each method is doing a single task, which makes the code easy to read, test, and maintain.

Therefore, it can be concluded that the PicsumPhotoService class follows SRP.

Wrap-up

It would be simple like that. Check more details in the source code, install it and try it yourself. This project is ready to:

  • Create a lambda function using Bref
  • Create an S3 Bucket with read and write permissions to the lambda
  • Test the stack Cloudformation code
  • Separate the PHP logic
  • Have PHP communicating with AWS Services
  • Start PHP testing

P.S.: Stats

I did not plan to talk widely about stats now, but I think I can share the most two significant measures I had with this simple code so far.

[Update 22/03/23] Using https://k6.io/

1 - With a brand new stack and a cold lambda:

scenarios: (100.00%) 1 scenario, 200 max VUs, 2m30s max duration (incl. graceful stop):
           * default: 200 looping VUs for 2m0s (gracefulStop: 30s)

     data_received..................: 49 MB  409 kB/s
     data_sent......................: 7.8 MB 65 kB/s
     http_req_blocked...............: avg=2.36ms   min=671ns    med=2.27µs   max=581.87ms p(90)=4.18µs   p(95)=7µs
     http_req_connecting............: avg=712.63µs min=0s       med=0s       max=193.34ms p(90)=0s       p(95)=0s
     http_req_duration..............: avg=531.51ms min=204.46ms med=485.24ms max=3.81s    p(90)=517.98ms p(95)=534.3ms
       { expected_response:true }...: avg=513.6ms  min=204.46ms med=485.07ms max=3.67s    p(90)=516.62ms p(95)=531.5ms
     http_req_failed................: 0.60%  ✓ 272        ✗ 44761
     http_req_receiving.............: avg=123.76µs min=13.77µs  med=44.04µs  max=16.78ms  p(90)=71.27µs  p(95)=85.71µs
     http_req_sending...............: avg=14.79µs  min=4.27µs   med=12.43µs  max=402.74µs p(90)=23.97µs  p(95)=31.4µs
     http_req_tls_handshaking.......: avg=1.37ms   min=0s       med=0s       max=330.58ms p(90)=0s       p(95)=0s
     http_req_waiting...............: avg=531.37ms min=204.36ms med=485.11ms max=3.81s    p(90)=517.77ms p(95)=534.13ms
     http_reqs......................: 45033  373.683517/s
     iteration_duration.............: avg=533.96ms min=204.55ms med=485.34ms max=4.37s    p(90)=518.07ms p(95)=534.4ms
     iterations.....................: 45033  373.683517/s
     vus............................: 200    min=200      max=200
     vus_max........................: 200    min=200      max=200

running (2m00.5s), 000/200 VUs, 45033 complete and 0 interrupted iterations

2 - After the first initial execution, cold lambda and all available images already saved to the bucket, where we got ~3K more requests being served for the same time

scenarios: (100.00%) 1 scenario, 200 max VUs, 2m30s max duration (incl. graceful stop):
           * default: 200 looping VUs for 2m0s (gracefulStop: 30s)

     data_received..................: 53 MB  442 kB/s
     data_sent......................: 8.4 MB 70 kB/s
     http_req_blocked...............: avg=2.26ms   min=631ns    med=2.24µs   max=612.22ms p(90)=4.04µs   p(95)=6.47µs
     http_req_connecting............: avg=663.23µs min=0s       med=0s       max=215.19ms p(90)=0s       p(95)=0s
     http_req_duration..............: avg=490.8ms  min=199.95ms med=484.02ms max=3.17s    p(90)=514.86ms p(95)=527ms
       { expected_response:true }...: avg=490.53ms min=199.95ms med=484.02ms max=2.4s     p(90)=514.85ms p(95)=526.99ms
     http_req_failed................: 0.01%  ✓ 5         ✗ 48754
     http_req_receiving.............: avg=108.86µs min=12.44µs  med=42.68µs  max=17.62ms  p(90)=69.23µs  p(95)=81.87µs
     http_req_sending...............: avg=14.42µs  min=3.9µs    med=12.14µs  max=786.01µs p(90)=23.03µs  p(95)=30.35µs
     http_req_tls_handshaking.......: avg=1.27ms   min=0s       med=0s       max=332.34ms p(90)=0s       p(95)=0s
     http_req_waiting...............: avg=490.68ms min=199.9ms  med=483.91ms max=3.17s    p(90)=514.75ms p(95)=526.89ms
     http_reqs......................: 48759  404.56812/s
     iteration_duration.............: avg=493.16ms min=200.05ms med=484.11ms max=3.17s    p(90)=514.96ms p(95)=527.1ms
     iterations.....................: 48759  404.56812/s
     vus............................: 200    min=200     max=200
     vus_max........................: 200    min=200     max=200

running (2m00.5s), 000/200 VUs, 48759 complete and 0 interrupted iterations