Skip to content

Latest commit

 

History

History
262 lines (197 loc) · 10.1 KB

migrating.md

File metadata and controls

262 lines (197 loc) · 10.1 KB
Raw

Steps to Migrate from Heroku to Control Plane

We recommend following along with this example project.

  1. Clone the Staging Environment
  2. Create the Review App Process
  3. Deploy to Production

Clone the Staging Environment

By cloning the staging environment on Heroku, you can speed up the initial provisioning of the app on Control Plane without compromising your current environment.

Consider migrating just the web dyno first, and get other types of dynos working afterward. You can also move the add-ons to Control Plane later once the app works as expected.

First, create a new Heroku app with all the add-ons, copying the data from the current staging app.

Then, copy project-specific configs to a .controlplane/ directory at the top of your project. cpflow will pick those up depending on which project folder tree it runs. Thus, this automates running several projects with different configs without explicitly switching configs.

Edit the .controlplane/controlplane.yml file as needed. Note that the my-app-staging name used in the examples below is defined in this file. See this example.

Before the initial setup, add the templates for the app to the .controlplane/controlplane.yml file, using the setup_app_templates key, e.g.:

my-app-staging:
  <<: *common
  setup_app_templates:
    - app
    - redis
    - memcached
    - rails
    - sidekiq

Note how the templates correspond to files in the .controlplane/templates/ directory. These files will be used by the cpflow setup-app and cpflow apply-template commands.

Ensure that env vars point to the Heroku add-ons in the template for the app (.controlplane/templates/app.yml). See this example.

After that, create a Dockerfile in .controlplane/Dockerfile for your deployment. See this example.

You should have a folder structure similar to the following:

app_main_folder/
  .controlplane/
    Dockerfile          # Your app's Dockerfile, with some Control Plane changes.
    controlplane.yml
    entrypoint.sh       # App-specific - edit as needed.
    templates/
      app.yml
      memcached.yml
      rails.yml
      redis.yml
      sidekiq.yml

The example .controlplane/ directory already contains these files.

Finally, check the app for any Heroku-specific code and update it, such as the HEROKU_SLUG_COMMIT env var and other env vars beginning with HEROKU_. You should add some logic to check for the Control Plane equivalents - it might be worth adding a CONTROLPLANE env var to act as a feature flag and help run different code for Heroku and Control Plane until the migration is complete.

You might want to review special gems and create a minimum bootable config.

At first, do the deployments from the command line. Then set up CI scripts to trigger the deployment upon merges to master/main.

Use these commands for the initial setup and deployment:

# Provision infrastructure (one-time-only for new apps) using templates.
cpflow setup-app -a my-app-staging

# Build and push image with auto-tagging, e.g., "my-app-staging:1_456".
cpflow build-image -a my-app-staging --commit 456

# Prepare database.
cpflow run -a my-app-staging --image latest -- rails db:prepare

# Deploy latest image.
cpflow deploy-image -a my-app-staging

# Open app in browser.
cpflow open -a my-app-staging

Then for promoting code upgrades:

# Build and push new image with sequential tagging, e.g., "my-app-staging:2".
cpflow build-image -a my-app-staging

# Or build and push new image with sequential tagging and commit SHA, e.g., "my-app-staging:2_ABC".
cpflow build-image -a my-app-staging --commit ABC

# Run database migrations (or other release tasks) with latest image, while app is still running on previous image.
# This is analogous to the release phase.
cpflow run -a my-app-staging --image latest -- rails db:migrate

# Deploy latest image.
cpflow deploy-image -a my-app-staging

Review Special Gems

Make sure to review "special" gems which might be related to Heroku, e.g.:

  • rails_autoscale_agent. It's specific to Heroku, so it must be removed.
  • puma_worker_killer. In general, it's unnecessary on Control Plane, as Kubernetes containers will restart on their own logic and may not restart at all if everything is ok.
  • rack-timeout. It could possibly be replaced with Control Plane's timeout option.

You can use the CONTROLPLANE env var to separate the gems, e.g.:

# Gemfile
group :staging, :production do
	gem "rack-timeout"

  unless ENV.key?("CONTROLPLANE")
	  gem "rails_autoscale_agent"
    gem "puma_worker_killer"
  end
end

Create a Minimum Bootable Config

You can try to create a minimum bootable config to migrate parts of your app gradually. To do that, follow these steps:

  1. Rename the existing application.yml file to some other name (e.g., application.old.yml)
  2. Create a new minimal application.yml file, e.g.:
SECRET_KEY_BASE: "123"
# This should be enabled for `rails s`, not `rails assets:precompile`.
# DATABASE_URL: postgres://localhost:5432/dbname
# RAILS_SERVE_STATIC_FILES: "true"

# You will add whatever env vars are required here later.
  1. Try running RAILS_ENV=production CONTROLPLANE=true rails assets:precompile (theoretically, this should work without any additional env vars)
  2. Fix whatever code needs to be fixed and add missing env vars (the fewer env vars are needed, the cleaner the Dockerfile will be)
  3. Enable DATABASE_URL and RAILS_SERVE_STATIC_FILES env vars
  4. Try running RAILS_ENV=production CONTROLPLANE=true rails s
  5. Fix whatever code needs to be fixed and add required env vars to application.yml
  6. Try running your production entrypoint command, e.g., RAILS_ENV=production RACK_ENV=production CONTROLPLANE=true puma -C config/puma.rb
  7. Fix whatever code needs to be fixed and add required env vars to application.yml

Now you should have a minimal bootable config.

Then you can temporarily set the LOG_LEVEL=debug env var and disable unnecessary services to help with the process, e.g.:

DISABLE_SPRING: "true"
SCOUT_MONITOR: "false"
RACK_TIMEOUT_SERVICE_TIMEOUT: "0"

Create the Review App Process

Add an entry for review apps to the .controlplane/controlplane.yml file. By adding a match_if_app_name_starts_with key with the value true, any app that starts with the entry's name will use this config. Doing this allows you to configure an entry for, e.g., my-app-review, and then create review apps starting with that name (e.g., my-app-review-1234, my-app-review-5678, etc.). Here's an example:

  my-app-review:
    <<: *common
    match_if_app_name_starts_with: true
    setup_app_templates:
      - app
      - redis
      - memcached
      - rails
      - sidekiq

In your CI scripts, you can create a review app using some identifier (e.g., the number of the PR on GitHub).

# On CircleCI, you can use `echo $CIRCLE_PULL_REQUEST | grep -Eo '[0-9]+$'` to extract the number of the PR.
PR_NUM=$(... extract the number of the PR here ...)
echo "export APP_NAME=my-app-review-$PR_NUM" >> $BASH_ENV

# Only create the app if it doesn't exist yet, as we may have multiple triggers for the review app
# (such as when a PR gets updated).
if ! cpflow exists -a ${APP_NAME}; then
  cpflow setup-app -a ${APP_NAME}
  echo "export NEW_APP=true" >> $BASH_ENV
fi

# The `NEW_APP` env var that we exported above can be used to either reset or migrate the database before deploying.
if [ -n "${NEW_APP}" ]; then
  cpflow run -a ${APP_NAME} --image latest -- rails db:reset
else
  cpflow run -a ${APP_NAME} --image latest -- rails db:migrate
fi

Then follow the same steps for the initial deployment or code upgrades.

Database for Review Apps

For the review app resources, these should be handled as env vars in the template for the app (.controlplane/templates/app.yml), .e.g.:

- name: DATABASE_URL
  value: postgres://postgres:[email protected]:5432/APP_GVC

Notice that APP_GVC is the app name, which is used as the database name on RDS, so that each review app gets its own database on the one RDS instance used for all review apps, which would be, e.g., my-app-review-1234.

Redis and Memcached for Review Apps

So long as no persistence is needed for Redis and Memcached, we have templates for workloads that should be sufficient for review apps in the templates/ directory of this repository. Using these templates results in considerable cost savings compared to paying for the resources on Heroku.

- name: MEMCACHE_SERVERS
  value: memcached.APP_GVC.cpln.local
- name: REDIS_URL
  value: redis://redis.APP_GVC.cpln.local:6379

Deploy to Production

Only try deploying to production once staging and review apps are working well.

For simplicity, keep add-ons running on Heroku initially. You could move over the database to RDS first. However, it's a bit simpler to isolate any differences in cost and performance by first moving over your compute to Control Plane.

Ensure that your Control Plane compute is in the AWS region US-EAST-1; otherwise, you'll have noticeable extra latency with your calls to resources. You might also have egress charges from Control Plane.

Use the cpflow promote-app-from-upstream command to promote the staging app to production.