Step Functions lets you set a timeout on Task states and the whole execution. By default, a Task state times out after 60 seconds. But an execution can run for a year if no TimeoutSeconds is configured. To a user, the execution would appear as “stuck”. AWS best practices recommend using timeouts to avoid such scenarios [1]. So it’s important to consider what happens when you experience a timeout You can use the Catch clause to handle the States.Timeout error when a Task state times out. You can then perform automated remediation steps. But what happens when the whole execution times out? How c ..read more

Testing is an integral part of software development. Your tests are a living documentation of your system. They inform others how to use your system, but they are so much more than that. One of the most misunderstood parts of Test-Driven Development (TDD) is the “Driven” part of the name. It’s not just about “writing tests before you write the code”. If your tests do not inform and drive your API design, then you’re not really doing TDD. When I say “API”, I mean the general meaning of the term, not the API spec for HTTP APIs, although they are part of it. In the context of a serverless applica ..read more

The myth that “you can’t do WebSockets with serverless” still persists today, even though we have some very good ways to implement WebSockets without needing to manage any servers. Part of the problem is that many still falsely equate “serverless” with Lambda. But serverless is much more than that. To me, it describes any technology that: No need to manage servers. Scale to zero. Usage-based pricing with no minimum charge. With this in mind, API Gateway, AppSync, and IoT Core are all serverless technologies. All three let you implement WebSockets. Momento Topics [1] is another good option if ..read more

DynamoDB announced support for resource-based policies [1] a few days ago. It makes cross-account access to DynamoDB tables easier. You no longer need to assume an IAM role in the target table’s account. I was confused by this update and wondered if it was even a good idea. If you need cross-account access to DynamoDB, then it’s surely a sign you’re breaking service boundaries, right? As I said before [2], a microservice should own its data and shouldn’t share a database with another microservice. In many organizations, microservices run in their own accounts. This provides another layer of in ..read more

I’m a big fan of AWS Step Functions. I use it to orchestrate all sorts of workflows, from payment processing to map-reduce jobs. Why it’s yet another AWS service you need to learn and pay for. And it introduces additional complexities, such as: It’s hard to test [1]. Your business logic is split between configuration and code. New decision points. Such as whether to use Express Workflows or Standard Workflows [2]. So it’s fair to ask “Why should we even bother with Step Functions?” when you can do all the orchestration in code, inside a Lambda function. Let’s break it down. Lambda pros 1. Do ..read more

“Lambdalith” is a monolithic approach to building serverless applications where a single Lambda function serves an entire API, instead of one function per endpoint. It’s an increasingly popular approach. It provides portability between Lambda functions and container applications. You can lift and shift an existing application into Lambda without rewriting it. You can use web frameworks you are already familiar with, and lean on the existing ecosystems of tools, ORMs and middleware. It also makes testing easier, because you can apply familiar testing methodologies. Tools like the AWS Lambda Web ..read more

I thought Lambda needed a specialised runtime. One that works well with its resource-constraint execution environment. I even floated a few ideas in the past but sadly I don’t have the chops to make them happen myself. So I was pleasantly surprised when AWS open-sourced the LLRT runtime for JavaScript [1]! What is LLRT? LLRT, or Low Latency Runtime, is a new and experimental JavaScript runtime for Lambda. It promises 10x faster startup time. Which should significantly help with the dreaded Lambda cold starts. Naturally, I had to test it out for myself and see if the hype was real ..read more

I shared on Linkedin [1] the other day that you should avoid using Cognito subs as the user ID for your system. One of the reasons is that a user’s sub does not carry over when you migrate to a new user pool. Someone responded by asking “Is this type of migration really that common that it necessitates consideration?” It’s a great question, so let’s dive into it. When should you consider a user pool migration? How best to do this migration? When to consider user pool migration Migrating users from one Cognito User Pool to another can be highly disruptive. But sometimes it’s our last resort ..read more

By now, most of us have moved away from using IAM users for CI/CD pipelines. Instead, we’d use dedicated CI/CD roles, one for each pipeline. This forces us to consider who can assume this role. Identity federation is widely supported by 3rd-party providers such as GitHub Actions [1]. So, no more putting IAM credentials in CI/CD tools and worry that they might be compromised in a security breach [2]. However, attackers can still compromise the pipeline through supply chain attacks. For example, by compromising a Docker image we depend on in our CI/CD pipeline. Or by compromising static analysis ..read more

Imagine this… You have followed AWS best practices and set up a dead-letter queue (DLQ) or an OnFailure destination for every async Lambda function. (sidebar: you should prefer Lambda Destination over DLQs, here’s why [1]) A message arrives in your DLQ. You are alerted right away because you have alarms on all of your DLQs. You investigate the problem and determine that it was temporary and the message should be re-processed. But now what? Do you extract the payload and invoke the original function manually? What if there are hundreds of similar messages? This manual approach doesn’t scale wel ..read more