The second week of AWS re:Invent brought a surprise major announcement from the AWS Amplify team: You can run Docker containers as AWS Amplify APIs now!!!
The main Achilles heel of migrating to serverless architectures is having to relearn and rewrite traditional serverful code. With the new Amplify Serverless Containers, it is now possible to deploy code in stateful Docker containers with the exact same workflow as the rest of Amplify CLI.
# example of the new CLI workflow, see docs for details # Amplify CLI version 4.39+ $ amplify init # if you're setting up new Amplify project $ amplify configure project > Do you want to enable container-based deployments? Yes $ amplify add api Please select from one of the below mentioned services: REST ? Which service would you like to use API Gateway + Lambda ❯ API Gateway + AWS Fargate (Container-based)
For the first time, this opens up the possibility of running Rails, Django, Laravel, or any other traditionally serverful APIs seamlessly alongside a serverless architecture. If it fits in a Docker container, you can run it! This thus unlocks the full capabilities of their respective ecosystems.
Emphasis on seamless integration: this means that frontend and mobile developers don't need to go to 9 months of cloud school (having to learn the details of VPC, subnets, NACLs, IAM policies, SSL generation, continuous deploy, cost optimization, etc) just in order to get a containerized API up and running on AWS.
This, to me, is a big deal for a number of interrelated reasons:
Long Running Code: AWS Lambda functions are limited to 15 minutes. If I need to run longer tasks, like video conversion or batch processing, I would have to jump out of my existing workflow and use a completely different set of tools. Now, I can run Rails + Sidekiq or Flask + Celery with just an industry standard
docker-compose.yml. Because Amplify's Serverless Containers feature uses AWS Fargate, we may in future have the option of vertically scaling system resources as needed, beyond the traditional resource constraints of AWS Lambda.
- Access Entire Ecosystems: Libraries and frameworks that were written for serverful paradigms typically become useless when shifting to serverless/Jamstack architectures, requiring yet another learning curve and search for tools. Now you are free to mix and match them - anything that runs in a Docker container is fair game (and everything runs in a Docker container!)
- Incremental Migration: This also means that it is no longer a heavy lift to move existing apps and codebases to a serverless/Jamstack architecture. As long as you can separate your UI layer from your API layer, you are now able to host your UI on Amplify Console, and your APIs as a REST or GraphQL service with API Gateway + AWS Fargate.
Easy Access to other Amplify categories: because this plugs right into the Amplify API category workflow, this means that your code inside your containers get all the same benefits of Amplify APIs like easy authentication with Amazon Cognito and easy access to Amplify-managed resources like Storage, Functions, GraphQL and more. Because networking is handled for you, you can even talk to other containers - the default
docker-compose.ymlexample in the CLI shows how you can make an Express (JS) app interoperate with a Flask (Python) app!
So, yeah. I think this is awesome - You can run basically anything on AWS Amplify now - meaning that you spend less time wrangling infrastructure and more time writing valuable features. Don't take my word for it, see Ant Stanley (organizer of Serverless Days and overall cloud expert):
Ant Stanley Partial Landscaping #BLMThis is awesome. Easiest way to run a stateful workload in the cloud. Got Wordpress, Django, Express, Joomla, Flask, Laravel etc ... this is the easiest and quickest way to get it running in a scalable way in the cloud. twitter.com/renebrandel/st…17:36 PM - 10 Dec 2020Rene Brandel @renebrandel⭐️NEW: @AWSAmplify now supports containers⭐️ 👉 Deploy APIs & host web apps using @awscloud Fargate 👉 Build & deploy pipeline out-of-the-box 👉 Single & multi-container deployments (incl. networking) Amazing work by @panchorodriguez @menyao @undef_obj https://t.co/YBhqVkqnCS
P.S. for those who are making comparisons with AWS CDK, note that this capability was actually made with CDK - it solves a higher level problem than CDK.