mkubenka/discourse_aws

Discourse on AWS: Elastic Beanstalk + RDS + S3 + SES + Lambda + Terraform

Tooling to deploy Discourse on AWS.

what's where

• Postgres: RDS
• Rails + Redis: Docker container on Elastic Beanstalk Single Instance deployment
• Docker registry: ECR
• Files and backups: S3
• Outgoing email: SES
• Incoming email + bounce handling: SES + S3 + AWS Lambda
• Passwords for RDS and SES: credstash
• SSL: Let's Encrypt + nginx on the EC2 instance managed by Elastic Beanstalk
• Infrastructure management: Terraform

what's different from the official launcher

• Skips db:migrate and assets:precompile on bootstrap
• Uses a custom boot script to:
  • Fetch passwords and export as environment variables
  • Run db:migrate and assets:precompie
  • Execute the original boot script

hardcoded assumptions

• We will have one Elastic Beanstalk application named discourse
• and two Elastic Beanstalk environments: discourse-dev and discourse-prod

directory layout

path/to/your/configs
  discourse_aws/ # this repo
  containers/app.yml           # optional pups template to customize the Docker build
  ebextensions/                # optional Elastic Beanstalk configs
  dev/*.tf                     # mix and match modules from discourse_aws/tfmodules..
  prod/*.tf                    # to build AWS resources according to your needs.

• discourse_aws/
  • bring in this repo in whatever way you like: git submodule, git subtree, or a plain git clone
• ./containers/app.yml
  • this file, if present, will be merged with this repo's ./containers/app.yml
  • useful for configuring plugins to install (see: ./samples/app.yml)
• ./ebextensions/*.config
  • any file in here will be copied to the sourcebundle for Elastic Beanstalk
  • useful for setting up SSH keys on the EC2 instance (see: ./samples/ssl.config)

prerequisites

• install Terraform
• set up credstash
• set up your domain name of choice
  • CNAME -> cname_prefix.your_aws_region.elasticbeanstalk.com
  • MX -> inbound-smtp.your_ses_region.amazonaws.com
• set up SES
  • make a new SES user/password
  • verify your sender domain or email address
  • make a receipt rule set and make it active
  • request production access (this can be done later)

dev setup

• write your own Terraform config by combining modules in ./tfmodules (see: ./samples/main.tf)
  • consider setting certbot_extra_args to --staging first to test SSL setup
• plug in Terraform variables
  • SES username
  • domain name
  • etc
• initial deploy strategy should be AllAtOnce
  • plus whenever you expect certbot to create a new certificate
• terraform apply
• change db password on RDS
• credstash put discourse_db_password.discourse-dev
• credstash put discourse_smtp_password.discourse-dev
• ./build.sh
  • builds and tags a Discourse docker image as vYYYYmmdd-HHMMSS
• do a docker push: exact command will be printed out by build.sh
• ./deploy-dev.sh
  • creates and deploys an application version using the latest docker image on ECR: vYYYYmmdd-HHMMSS-bYYYYmmdd-HHMMSS
• change deploy strategy to Immutable to avoid downtime during deploys

prod setup

same as dev setup except for:

• credstash put discourse_db_password.discourse-prod
• credstash put discourse_smtp_password.discourse-prod
• deploy script is ./deploy-prod.sh
  • deploys the same application version as discourse-dev environment's

community setup

• setup s3 upload (otherwise your uploads will be wiped after the next deploy)
• setup s3 backup (optional)
• set up incoming email (advisable - to meet SES's bounce handling guidelines)
  • generate a master API key under Admin > API, go to the Lambda function for receiving email, and set the API key as DISCOURSE_API_KEY enivonment variable
  • site setup
    • manual polling enabled
    • reply by email address
    • reply by email enabled
  • test email receiver by sending a test email to one of the test email addresses at http://docs.aws.amazon.com/ses/latest/DeveloperGuide/mailbox-simulator.html
• go through the setup wizard

getting ready for let's encrypt production server

• ssh into the instance
• delete staging certs sudo find /etc/letsencrypt -iname "$your_discourse_hostname*" | xargs rm -rf
• also delete on s3
• remove --staging from certbot_extra_args. empty string is okay
• redeploy through Elastic Beanstalk console or another ./deploy-dev.sh / ./deploy-prod.sh

upgrading / adding plugins

• optional: update your ./containers/app.yml
• ./build.sh: builds a docker image locally
• docker push: as instructed at the end of ./build.sh's output
• ./deploy-dev.sh: packages up a new application version and deploys to discourse-dev
• ./deploy-prod.sh: deploys the current application version deployed to discourse-dev

discourse_aws's Issues

Rate limiting is not working correctly

Discourse is running behind Elastic Beanstalk Nginx reverse proxy which cause all requests to have same IP - 172.17.0.1.

Reference discourse/discourse_docker#336