alexei.dolgolyov
0bb52f9ec6
Create structured plan files with 12 phases covering the full implementation: scaffold, store, crypto, Docker/NPM clients, registry poller, webhook, deployer, API layer, SvelteKit frontend, embedding, and hardening.
3.0 KiB
3.0 KiB
Phase 7: Deployer & Health Checker
Status: ⬜ Not Started Parent plan: PLAN.md Domain: backend
Objective
Implement the core deployment orchestrator: pull → start container → configure NPM proxy → health check → success/rollback. Plus multi-instance support and notifications.
Tasks
- Task 1: Define deployer service struct — depends on Docker client, NPM client, store, notifier
- Task 2: Implement deploy flow: pull image → create container → start → connect to network → configure proxy → health check
- Task 3: Implement subdomain generation per convention:
stage-{stage}-{project}for default,stage-{stage}-{project}-{tag}for specific - Task 4: Sanitize tags for DNS (dots → dashes, lowercase, truncate)
- Task 5: Implement health checker — HTTP GET to
http://container:{port}{healthcheck_path}with retries and timeout - Task 6: Implement rollback on health check failure — remove new container, delete NPM proxy host, update instance status
- Task 7: Implement multi-instance support — multiple tags of same project/stage can run simultaneously
- Task 8: Implement max_instances enforcement — remove oldest instance when limit reached
- Task 9: Implement notification webhook — POST to configured URL on deploy success/failure
- Task 10: Create deploy history records in store (status, timestamps, logs)
- Task 11: Implement deploy log streaming — append log entries during deploy for real-time visibility
Files to Modify/Create
internal/deployer/deployer.go— main deploy orchestratorinternal/deployer/subdomain.go— subdomain generation and DNS sanitizationinternal/deployer/rollback.go— rollback logicinternal/health/checker.go— HTTP health checker with retriesinternal/notify/notifier.go— webhook notification sender
Acceptance Criteria
- Full deploy flow works end-to-end (pull → proxy → health check)
- Failed health checks trigger automatic rollback
- Multi-instance: deploying a new tag doesn't stop existing instances
- max_instances removes oldest instance when exceeded
- Notifications fire on success and failure
- Deploy history is recorded with status and timestamps
Notes
- Health check: 3 retries, 5s between retries, 10s timeout per attempt (configurable later)
- Subdomain pattern comes from global settings
- Notifications are fire-and-forget (don't block deploy on notification failure)
- Deploy logs should be structured entries (timestamp + message) for SSE streaming later
- The deployer is the central orchestrator — webhook handler and poller both call into it
Review Checklist
- All tasks completed
- Rollback cleans up ALL resources (container, proxy, instance record)
- No goroutine leaks
- Error handling at every step of the deploy flow
- Subdomain generation produces valid DNS names