Deployment Flow

Deployment Flow

This document describes how changes move from a developer commit to a running production deployment, and how drift detection catches out-of-band changes.

Merge Request Pipeline

When a merge request is opened or updated, the pipeline runs a subset of stages:

  1. Validate — format checks, schema validation, tofu validate
  2. Plan — generates execution plans for all affected stacks

The deploy stage is skipped. Plan output is posted as a comment or artifact so reviewers can inspect the exact infrastructure changes before approving the merge.

Main Branch Pipeline

After a merge request is approved and merged into main, the full pipeline runs automatically. There is no manual gate or approval step.

  1. Validate — same checks as the MR pipeline
  2. Build — compiles the SvelteKit app, builds the container image
  3. Plan — generates fresh execution plans from the merged code
  4. Deploy — runs tofu apply with the saved plan artifacts
  5. Verify — executes health checks against deployed services

Plan-Apply Handoff

The plan artifact produced in the Plan stage is passed directly to the Deploy stage. This guarantees that tofu apply executes exactly the changes that were planned, with no opportunity for drift between planning and applying.

Health Checks

The Verify stage confirms that deployed services are healthy:

  • HTTP GET against known health endpoints
  • Kubernetes pod readiness verification
  • Runner registration confirmation (for runner stacks)

A failed health check marks the pipeline as failed and triggers notification.

Drift Detection

A scheduled pipeline runs at a configured interval (typically daily) independent of any code change. It executes:

tofu plan -detailed-exitcode

The -detailed-exitcode flag causes tofu to exit with code 2 if the plan detects changes (drift) between the declared state and the actual cluster state. Exit code 0 means no drift.

When drift is detected, the pipeline raises an alert. The team can then choose to:

  • Apply the existing code to bring the cluster back in line
  • Update the code to reflect intentional manual changes
  • Investigate and revert unauthorized modifications

Flow Diagram

GloriousFlywheel