How to test serverless functions and applications

Testing serverless functions uses traditional test types and techniques, but you must also consider testing serverless applications as a whole. Cloud-based tests will provide the most accurate measure of quality of both your functions and serverless applications.

A serverless application architecture includes managed services that provide critical application functionality through API calls. For this reason, your development cycle should include automated tests that verify functionality when your function and services interact.

If you do not create cloud-based tests, you could encounter issues due to differences between your local environment and the deployed environment. Your continuous integration process should run tests against a suite of resources provisioned in the cloud before promoting your code to the next deployment environment, such as QA, Staging, or Production.

Continue reading this short guide to learn about testing strategies for serverless applications, or visit the Serverless Test Samples repository to dive in with practical examples, specific to your chosen language and runtime.

For serverless testing, you will still write unit, integration and end-to-end tests.

Targeted business outcomes

Testing serverless solutions may require slightly more time to set up tests that verify event-driven interactions between services. Keep the following practical business reasons in mind as you read this guide:

The quality of an application depends on testing a variety of scenarios to verify functionality. Carefully considering the business scenarios and automating those tests to run against cloud services will raise the quality of your application.

Software bugs and configuration problems have the least impact on cost and schedule when caught during an iterative development cycle. If issues remain undetected during development, finding and fixing in production requires more effort by more people.

A well planned serverless testing strategy will increase software quality and improve iteration time by verifying your Lambda functions and applications perform as expected in a cloud environment.

What to test

We recommend adopting a testing strategy that tests managed service behaviors, cloud configuration, security policies, and the integration with your code to improve software quality. Behavior testing, also known as black box testing, verifies a system works as expected without knowing all the internals.

In traditional server-based architecture, teams often define a scope for testing to only include code that runs on the application server. Other components, services, or dependencies are often considered external and out of scope for testing.

Serverless applications often consist of small units of work, such as Lambda functions that retrieve products from a database, or process items from a queue, or resize an image in storage. Each component runs in their own environment. Teams will likely be responsible for many of these small units within a single application.

Some application functionality can be delegated entirely to managed services such as Amazon S3, or created without using any internally developed code. There is no need to test these managed services, but you do need to test the integration with these services.

How to test serverless

You are probably familiar with how to test applications deployed locally: You write tests that run against code running entirely on your desktop operating system, or inside containers. For example, you might invoke a local web service component with a request and then make assertions about the response.

Serverless solutions are built from your function code and cloud-based managed services, such as queues, databases, event buses, and messaging systems. These components are all connected through an event-driven architecture, where messages, called events, flow from one resource to another. These interactions can be synchronous, such as when a web service returns results immediately, or an asynchronous action which completes at a later time, such as placing items in a queue or starting a workflow step. Your testing strategy must include both scenarios and test the interactions between services. For asynchronous interactions, you may to need detect side effects in downstream components that may not be immediately observable.

Replicating an entire cloud environment, including queues, database tables, event buses, security policies, and more, is not practical. You will inevitably encounter issues due to differences between your local environment and your deployed environments in the cloud. The variations between your environments will increase the time to reproduce and fix bugs.

In serverless applications, architecture components commonly exist entirely in the cloud, so testing against code and services in the cloud is necessary to develop features and fix bugs.

Testing techniques

In reality, your testing strategy will likely include a mix of techniques to increase quality of your solutions. You will use quick interactive tests to debug functions in the console, automated unit tests to check isolated business logic, verification of calls to external services with mocks, and occasional testing against emulators that mimic a service.

Testing in the cloud

Testing in the cloud is valuable for all phases of testing, including unit tests, integration tests, and end-to-end tests. When you run tests against cloud-based code that also interacts with cloud-based services, you get the most accurate measure of quality of your code.

A convenient way to run a Lambda function in the cloud is with a test event in the AWS Management Console. A test event is a JSON input to your function. If your function does not require input, the event can be an empty JSON document ({}). The console provides sample events for a variety of service integrations. After creating an event in the console, you can also share it with your team to make testing easier and consistent.

Learn how to debug a sample function in the console.

Test automation samples are available in the Serverless Test Samples repository. The following command line runs an automated Python integration test example:

python -m pytest -s tests/integration -v

Although the test runs locally, it interacts with cloud-based resources. These resources have been deployed using the AWS Serverless Application Model and AWS SAM command line tool. The test code first retrieves the deployed stack outputs, which includes the API endpoint, function ARN, and security role. Next, the test sends a request to the API endpoint, which responds with a list of Amazon S3 buckets. This test runs entirely against cloud-based resources to verify those resources are deployed, secured, and work as expected.

========================= test session starts =========================
      platform darwin -- Python 3.10.10, pytest-7.3.1, pluggy-1.0.0
      -- /Users/t/code/aws/serverless-test-samples/python-test-samples/apigw-lambda/venv/bin/python
      cachedir: .pytest_cache
      rootdir: /Users/t/code/aws/serverless-test-samples/python-test-samples/apigw-lambda
      plugins: mock-3.10.0
      collected 1 item                                                                                                        

      tests/integration/test_api_gateway.py::TestApiGateway::test_api_gateway 

      --> Stack outputs:

        HelloWorldApi
        = https://p7teqs3162.execute-api.us-east-2.amazonaws.com/Prod/hello/
        > API Gateway endpoint URL for Prod stage for Hello World function

        PythonTestDemo
        = arn:aws:lambda:us-east-2:123456789012:function:testing-apigw-lambda-PythonTestDemo-iSij8evaTdxl
        > Hello World Lambda Function ARN

        PythonTestDemoIamRole
        = arn:aws:iam::123456789012:role/testing-apigw-lambda-PythonTestDemoRole-IZELQQ9MG4HQ
        > Implicit IAM Role created for Hello World function

      --> Found API endpoint for "testing-apigw-lambda" stack...
      --> https://p7teqs3162.execute-api.us-east-2.amazonaws.com/Prod/hello/
      API Gateway response:
      amplify-dev-123456789-deployment|myapp-prod-p-loggingbucket-123456|s3-java-bucket-123456789
      PASSED

      ========================= 1 passed in 1.53s ========================= 
      

For cloud-native application development, testing in the cloud provides the following benefits:

Testing in the cloud does have some disadvantages. The most obvious negative of testing in the cloud is that deployments to cloud environments typically take longer than deployments to a local desktop environments.

Fortunately, tools such as AWS Serverless Application Model (AWS SAM) Accelerate, AWS Cloud Development Kit (AWS CDK) watch mode, and SST (3rd party) reduce the latency involved with cloud deployment iterations. These tools can monitor your infrastructure and code and automatically deploy incremental updates into your cloud environment.

Unlike local testing, testing in the cloud requires additional resources which may incur service costs. Creating isolated testing environments may increase the burden on your DevOps teams, especially in organizations with strict controls around accounts and infrastructure. Even so, when working with complex infrastructure scenarios, the cost in developer time to set up and maintain an intricate local environment could be similar (or more costly) than using disposable testing environments created with Infrastructure as Code automation tools.

Testing in the cloud, even with these considerations, is still the best way to guarantee the quality of your serverless solutions.

Testing with mocks

Testing with mocks is a technique where you create replacement objects in your code to simulate the behavior of a cloud service.

For example, you could write a test that uses a mock of the Amazon S3 service that returns a specific response whenever the CreateObject method is called. When a test runs, the mock returns that programmed response without calling Amazon S3, or any other service endpoints.

Mock objects are often generated by a mock framework to reduce development effort. Some mock frameworks are generic and others are designed specifically for AWS SDKs, such as Moto, a Python library for mocking AWS services and resources.

Note that mock objects differ from emulators in that mocks are typically created or configured by a developer as part of the test code, whereas emulators are standalone applications that expose functionality in the same manner as the systems they emulate.

The advantages of using mocks include the following:

Mock testing has these disadvantages:

Testing with emulation

Emulators are typically a locally running application which mimics a production AWS service.

Emulators have APIs that are similar to their cloud counterparts and provide similar return values. They can also simulate state changes that are initiated by API calls. For example, you might use AWS SAM to run a function with AWS SAM local to emulate the Lambda service so that you can quickly invoke a function. See AWS SAM local in the AWS Serverless Application Model Developer Guide for details.

The advantages of test with emulators include the following:

Testing with emulators has these disadvantages:

Best practices

The following sections provide recommendations for successful serverless application testing.

You can find practical examples of tests and test automation in the Serverless Test Samples repository.

Prioritize testing in the cloud

Testing in the cloud provides the most reliable, accurate, and complete test coverage. Performing tests in the context of the cloud will comprehensively test not only business logic but also security policies, service configurations, quotas, and the most up to date API signatures and return values.

Structure your code for testability

Simplify your tests and Lambda functions by separating Lambda-specific code from your core business logic.

Your Lambda function handler should be a slim adapter that takes in event data and passes only the details that matter to your business logic method(s). With this strategy, you can wrap comprehensive tests around your business logic without worrying about Lambda-specific details. Your AWS Lambda functions should not require setting up a complex environment or large amount of dependencies to create and initialize the component under test.

Generally speaking, you should write a handler that extracts and validates data from the incoming event and context objects, then sends that input to methods that perform your business logic.

Accelerate development feedback loops

There are tools and techniques to accelerate development feedback loops. For example, AWS SAM Accelerate and AWS CDK watch mode both decrease the time required to update cloud environments.

The samples in the GitHub Serverless Test Samples repository explore some of these techniques.

We also recommend that you create and test cloud resources as early as possible during development—not only after a check-in to source control. This practice enables quicker exploration and experimentation when developing solutions. In addition, automating deployment from a development machine helps you discover cloud configuration problems more quickly and reduces wasted effort for updates and code review processes.

Focus on integration tests

When building applications with Lambda, testing components together is a best practice.

Tests that run against two or more architectural components are called integration tests. The goal of integration tests is to understand not only how your code will execute across components, but how the environment hosting your code will behave. End-to-end tests are special types of integration tests that verify behaviors across an entire application.

To build integration tests, deploy your application to a cloud environment. This can be done from a local environment or through a CI/CD pipeline. Then, write tests to exercise the system under test (SUT) and validate expected behavior.

For example, the system under test could be an application that uses API Gateway, Lambda and DynamoDB. A test could make a synthetic HTTP call to an API Gateway endpoint and validate that the response included the expected payload. This test validates that the AWS Lambda code is correct, and that each service is correctly configured to handle the request, including the IAM permissions between them. Further, you could design the test to write records of various sizes to verify your service quotas, such as max record size in DynamoDB, are set up correctly.

Create isolated test environments

Testing in the cloud typically requires isolated developer environments, so that tests, data, and events do not overlap.

One approach is to provide each developer a dedicated AWS account. This will avoid conflicts with resource naming that can occur when multiple developers working in a shared code base, attempt to deploy resources or invoke an API.

Automated test processes should create uniquely named resources for each stack. For example, you can set up scripts or TOML configuration files so that AWS SAM CLI sam deploy or sam sync commands will automatically specify a stack with a unique prefix.

In some cases, developers share an AWS account. This may be due to having resources in your stack that are expensive to operate, or to provision and configure. For example, a database may be shared to make it easier to set up and seed the data properly

If developers share an account, you should set boundaries to identify ownership and eliminate overlap. One way to do this is by prefixing stack names with developer user IDs. Another popular approach is to set up stacks based on code branches. With branch boundaries, environments are isolated, but developers can still share resources, such as a relational database. This approach is a best practice when developers work on more than one branch at a time.

Testing in the cloud is valuable for all phases of testing, including unit tests, integration tests, and end-to-end tests. Maintaining proper isolation is essential; but you still want your QA environment to resemble your production environment as closely as possible. For this reason, teams add change control processes for QA environments.

For pre-production and production environments, boundaries are typically drawn at the account level to insulate workloads from noisy neighbor problems and implement least privilege security controls to protect sensitive data. Workloads have quotas. You don't want your testing to consume quotas allocated for production (noisy neighbor) or have access to customer data. Load testing is another activity you should isolate from your production stack.

In all cases, environments should be configured with alerts and controls to avoid unnecessary spending. For example, you can limit the type, tier, or size of resources that can be created, and set up email alerts when estimated costs exceed a given threshold.

Use mocks for isolated business logic

Mock frameworks are a valuable tool for writing fast unit tests. They are especially beneficial when tests cover complex internal business logic, such as mathematical or financial calculations or simulations. Look for unit tests that have a large number of test cases or input variations, where those inputs do not change the pattern or the content of calls to other cloud services.

Code that is covered by unit tests with mocks should also be covered by testing in the cloud. This is recommended because a developer laptop or build machine environment could be configured differently than a production environment in the cloud. For example, your Lambda functions could use more memory or time than allocated when run with certain input parameters. Or your code might include environment variables that are not configured in the same way (or at all), and the differences could cause the code to behave differently or fail.

The benefit of mocks is less for integration tests, because the level of effort to implement the necessary mocks increases with the number of connection points. End-to-end testing should not use mocks, because these tests generally deal with states and complex logic that cannot be easily simulated with mock frameworks.

Lastly, avoid using mocked cloud services to validate the proper implementation of service calls. Instead, make cloud service calls in the cloud to validate behavior, configuration, and functional implementation.

Use emulators sparingly

Emulators can be convenient for some use cases, for example, for a development team with limited, unreliable, or slow internet access. But, in most circumstances, choose to use emulators sparingly.

By avoiding emulators, you will be able to build and innovate with the latest service features and up to date APIs. You will not be stuck waiting on vendor releases to achieve feature parity. You will reduce your upfront and ongoing expenses for purchasing and configuration on multiple development systems and build machines. Moreover, you will avoid the problem that many cloud services simply do not have emulators available. A testing strategy that depends on emulation will make it impossible to use those services (leading to potentially more expensive workarounds) or produce code and configurations that aren’t well tested.

When you do use emulation for testing, you must still test in the cloud to verify configuration and to test interactions with cloud services that can only be simulated or mocked in an emulated environment.

Challenges testing locally

When you use emulators and mocked calls to test on your local desktop you might experience testing inconsistencies as your code progresses from environment to environment in your CI/CD pipeline. Unit tests to validate your application’s business logic on your desktop may not accurately test critical aspects of the cloud services.

The following examples provide cases to watch out for when testing locally with mocks and emulators:

Example: Lambda function creates an S3 bucket

If a Lambda function’s logic depends on creating an S3 bucket, a complete test should confirm that Amazon S3 was called and the bucket was successfully created.

The mock and emulation setups will test what the Lambda function will do if it calls Amazon S3; however, those tests will not verify that the Lambda function, as configured, is capable of successfully creating the Amazon S3 bucket. You must make sure the role assigned to the function has an attached security policy that allows the function to perform the s3:CreateBucket action. If not, the function will likely fail when deployed to a cloud environment.

Example: Lambda function processes messages from an Amazon SQS queue

If an Amazon SQS queue is the source of a Lambda function, a complete test should verify that the Lambda function is successfully invoked when a message is put in a queue.

Emulation testing and mock testing are generally set up to run the Lambda function code directly, and to simulate the Amazon SQS integration by passing a JSON event payload (or a deserialized object) as the function handler’s input.

Local testing that simulates the Amazon SQS integration will test what the Lambda function will do when it’s called by Amazon SQS with a given payload, but the test will not verify that Amazon SQS will successfully invoke the Lambda function when it is deployed to a cloud environment.

Some examples of configuration problems you might encounter with Amazon SQS and Lambda include the following:

As these examples show, tests that cover business logic but not the configurations between cloud services are likely to provide unreliable results.

FAQ

I have a Lambda function that performs calculations and returns a result without calling any other services. Do I really need to test it in the cloud?

Yes. Lambda functions have configuration parameters that could change the outcome of the test. All Lambda function code has a dependency on timeout and memory settings, which could cause the function to fail if those settings are not set properly. Lambda policies also enable standard output logging to Amazon CloudWatch. Even if your code does not call CloudWatch directly, permission is needed to enable logging. This required permission cannot be accurately mocked or emulated.

How can testing in the cloud help with unit testing? If it’s in the cloud and connects to other resources, isn’t that an integration test?

We define unit tests as tests that operate on architectural components in isolation, but this does not prevent tests from including components that may call other services or use some network communication.

Many serverless applications have architectural components that can be tested in isolation, even in the cloud. One example is a Lambda function that takes input, processes the data, and sends a message to an Amazon SQS queue. A unit test of this function would likely test whether input values result in certain values being present in the queued message.

Consider a test that is written by using the Arrange, Act, Assert pattern:

A mock testing approach would involve mocking the queue with an in-process mock object, and creating an in-process instance of the class or module that contains the Lambda function code. During the Assert phase, the queued message would be retrieved from the mocked object.

In a cloud-based approach, the test would create an Amazon SQS queue for the purposes of the test, and would deploy the Lambda function with environment variables that are configured to use the isolated Amazon SQS queue as the output destination. After running the Lambda function, the test would retrieve the message from the Amazon SQS queue.

The cloud-based test would run the same code, assert the same behavior, and validate the application’s functional correctness. However, it would have the added advantage of being able to validate the settings of the Lambda function: the IAM role, IAM policies, and the function’s timeout and memory settings.

Next steps and resources

Use the following resources to learn more and explore practical examples of testing.

Sample implementations

The Serverless Test Samples repository on GitHub contains concrete examples of tests that follow the patterns and best practices described in this guide. The repository contains sample code and guided walkthroughs of the mock, emulation, and cloud testing processes described in previous sections. Use this repository to get up to speed on the latest serverless testing guidance from AWS.

Further reading

Visit Serverless Land to access the latest blogs, videos, and training for AWS serverless technologies.

The following AWS blog posts are also recommended reading:

Tools