Skip to content
Grada

Ephemeral PR Previews

When enabled, grada automatically configures your GitHub Actions pipeline to spin up isolated, temporary AWS environments every time a developer opens a Pull Request.

A bot will comment on the PR with a live URL (e.g., http://pr-123-your-app...), allowing your team to test features, UI changes, and API updates before merging to main. When the PR is closed or merged, the environment is automatically destroyed.

🏗️ How it Works

Section titled “🏗️ How it Works”

Under the hood, grada utilizes Terraform Workspaces.

When a PR is opened, Terraform creates a new workspace (e.g., pr-12). It provisions a completely isolated Application Load Balancer and ECS Fargate Task using the exact same infrastructure definitions as your production environment, ensuring 100% parity. On --target lambda projects the same workspace model provisions an isolated function and API Gateway per PR instead, deploys the PR-tagged image to it, and comments the preview API Gateway URL — same open-and-teardown lifecycle, no ALB per PR. Static targets (--target static) skip preview workflows entirely — there is no compute to preview against.

To save time and simplify architecture, PR environments share your production AWS Secrets Manager vault and ECR Image Repository.

⚖️ The Rule of Thumb: Should I enable this?

Section titled “⚖️ The Rule of Thumb: Should I enable this?”

✅ Enable PR Previews if:

  • You are working on a team of 2+ developers and require visual QA or UX sign-off before merging code.
  • You are building a frontend application or full-stack monolith where seeing the live UI is critical.

❌ Do NOT enable PR Previews if:

  • You are a solo developer (you can just test locally).
  • You have a massive volume of PRs (e.g., automated Dependabot updates). Spinning up an AWS Load Balancer takes ~3 minutes, which will slow down rapid automated merges.

💰 AWS Cost Implications

Section titled “💰 AWS Cost Implications”

Because PR previews provision a real Application Load Balancer (ALB) and ECS Fargate compute tasks, they are not free.

  • Compute: You are charged standard AWS Fargate rates per minute while the PR environment is running.
  • Load Balancing: AWS charges $16/month per active Load Balancer. If a PR is open for 2 days, you pay the prorated ALB cost for those 48 hours ($1.00).
  • Addons: Because grada add resources use ${local.app_name}, each open PR preview workspace provisions its own isolated S3 bucket and/or DynamoDB table — accruing usage-based storage, request, and PITR charges until the PR closes and the workspace is destroyed. If you use db:redis, each open PR also runs its own isolated Valkey node at $9.49/mo ($0.013/hr) until the PR closes, while queue:sqs, ai:bedrock, and cron add no fixed per-PR cost (each open PR runs its own copy of the schedule, billed only for Fargate seconds per invocation). Secrets Manager secrets are shared via data source at no extra per-PR secret cost. SES (email:ses) is the exception: the domain identity, DKIM, MAIL FROM, and DNS records are account-wide singletons scoped to the production workspace, so previews neither duplicate them nor delete them on teardown — preview containers simply inherit sending permission through their own task role.
  • Custom domains: domain add resources (ACM certificate, validation, and alias records) are likewise scoped to the production workspace. PR previews serve over their own isolated *.cloudfront.net URL and never claim your production alias, so closing a PR cannot disturb production DNS or TLS.

To keep costs low, ensure your team closes or merges Pull Requests promptly so the teardown.yml workflow can destroy the resources and stop the billing clock!