Skip to content

Server infrastructure

AMY is hosted in AWS cloud. There are two AMY environments, one for testing and one for production. They use separate setups.


AWS services used

Server is hosted on Elastic Compute Cloud (EC2) instance. Database uses separate EC2 instance through Relational Database Service (RDS). Additionally CloudFront is used as CDN for assets (images, CSS, JavaScript, including NPM dependencies).

WWW server

AMY is hosted by Nginx and has 4 roles:

  1. handles TLS,
  2. serves a Django application through gunicorn (working as a proxy),
  3. serves a maintenance page when upstream proxy is not available, for example during upgrade.

Nginx configuration is similar to vhost-like Apache configuration. The web server was installed through Ubuntu repositories and should start automatically with system's boot.

Nginx is configured to check for existence of maintenance_on.html file in /webapps/{domain}. If it's there, then 503 Service Unavailable is served with that HTML file on all requests.


AMY, as a WSGI application, is run by gunicorn on a separate user/group. Gunicorn creates 3 (configurable value) workers for handling the incoming requests. Logs are located in /webapps/{domain}/logs/gunicorn_supervisor.log and rotated.

TLS certificate

Certbot is used to acquire the TLS domain certificate.

Application directory

AMY directory is located on the server under /webapps/{domain}. It contains:

  • DB_backups - directory with old regular DB backups and backups created before every deployment,
  • logs - directory with logs from various components,
  • repo - directory with application source code and virtual environment,
  • maintenance_off.html - file used for entering maintenance mode (when renamed to maintenance_on.html).

Virtual environment

To get into AMY's virtual environment, issue:

$ source /webapps/{domain}/repo/.venv/bin/activate

You may need to do this from root's (superuser) privilege level.

It's also important to activate environment variables used by the project. You can do that with:

$ source /webapps/{domain}/repo/.venv/bin/postactivate


Redis is installed and served from the application server itself. There is no special configuration used for it, and it's not exposed.


The database servers are regularly backed-up by AWS. For more details see database backups.

Additional backups are created before starting any deployment with pg_dump.


AMY is deployed with Ansible scripts. The standard deployment procedure is documented in a file in separate repository containing these scripts.


Testing stage has recently been updated to fix many of the shortcomings listed below in Infrastructure shortcomings and future plans.

AWS services used

There is no Redis instance as it's not needed right now on the testing stage.

Infrastructure as code

Infrastructure is defined in a separate repository as a Cloud Development Kit (CDK) project. It gets compiled into CloudFormation template.


AMY is deployed to the testing stage with CI/CD pipeline on GitHub Actions.

Infrastructure shortcomings and future plans

Note March 2023: this section mostly applies to the production environment. Testing stage has been revamped and most of the issues listed below have been resolved.

A note has been added to every item that has been resolved in testing.

There are couple of issues with current approach to deployment:

  1. Redis is on the same server as the application
  2. Application server runs certbot to refresh TLS certificate
  3. RQ worker and scheduler are on the same server as the application
  4. Application requires a SSH-enabled server and is not immutable
  5. There's no CI/CD pipeline set up - deployment is manual
  6. There's downtime during deployment
  7. It's not scalable
  8. It requires custom deployment scripts (Ansible).
  9. There's multiple log outputs from various services in one server (two or three from AMY, one from rqworker, one from rqscheduler, one from gunicorn, one from a cronjob, two from nginx)

Below are propositions how to resolve each problem individually.

Migrate Redis

Redis should be on a separate machine. AWS offers multiple services for hosted Redis solutions, but they may not be needed. Perhaps a small EC2 sever dedicated to Redis could work, too?

Note March 2023 [testing stage]: does not apply; there is no Redis required.

Replace certbot with AWS services

There is better and also free alternative to certbot. It's Route 53 for keeping domain records, and Certificate Manager for generating free TLS certificates.

Finally, Route 53 also works with Elastic Load Balancing, which should be used to resolve issues like lack of scalability or CI/CD.

Note March 2023 [testing stage]: TLS and domain is managed by AWS (Route 53 service).

Decouple RQ worker and scheduler from application

Unfortunately this is not easy. RQ worker and scheduler are tightly coupled with the application because they use objects from application's memory pickled and stored in Redis database. This means that RQ worker and scheduler require access to the same source code.

It would be the best to completely decouple RQ jobs from AMY source code. Even better if the jobs themselves became JSON configuration files, and code to run them could be run on a Lambda.

These are all pretty big changes to the application, because Redis and RQ are used in automated emails feature.

Note March 2023: this was not achieved, and is required for migrating production to containerized (cloud native) solution.

Containerize the application to make it immutable

AMY should be containerized. It's already possible to build a Docker image with AMY, but it probably should run migrations automatically (it doesn't yet).

Containerization is also required to run AMY in a scalable manner on ECS / EKS.

Note March 2023 [testing stage]: AMY is containerized and runs on ECS.

Build CI/CD pipeline

Probably using GitHub Actions (see docs).

Note March 2023 [testing stage]: CI/CD pipeline is set up using GitHub Actions.

Use ECS for blue/green deployments and load balancing

ECS should be used to deploy AMY. Blue/green deployment strategy should be used, and a load balancer (which helps resolve multiple issues mentioned above).

Note March 2023 [testing stage]: ECS is used for deployment and load balancing.


AMY should be deployed using IaC (Infrastructure as Code) solution, for example. Terraform or Cloud Formation.

Note March 2023 [testing stage]: AMY infrastructure is deployed using CDK and CloudFormation.

Simplified logs

Once AMY resides on a single VM (for example ECS task instance), it should produce one stream of logs; these logs should be stored in CloudWatch.

Note March 2023 [testing stage]: logs are stored in CloudWatch.