Files
flyer-crawler.projectium.com/docs/adr/0062-lightweight-version-sync-workflow.md
2026-02-19 00:36:02 +05:00

204 lines
6.1 KiB
Markdown

# ADR-0062: Lightweight Version Sync Workflow
**Status:** Accepted
**Date:** 2026-02-18
**Decision Makers:** Development Team
**Related:** ADR-061 (PM2 Process Isolation Safeguards)
## Context
After successful production deployments, the version number in `package.json` is bumped and pushed back to the `main` branch. This triggered the full test deployment workflow (`deploy-to-test.yml`), which includes:
- `npm ci` (dependency installation)
- TypeScript type-checking
- Prettier formatting
- ESLint linting
- Unit tests (~150 tests)
- Integration tests (~50 tests)
- React build with source maps
- Full rsync deployment
- PM2 process restart
**Problem:** Running the complete 5-7 minute test suite just to update a 10KB `package.json` file was wasteful:
| Resource | Waste |
| -------------- | ------------------------------------- |
| **Duration** | 5-7 minutes (vs 30 seconds needed) |
| **CPU** | ~95% unnecessary (full test suite) |
| **File I/O** | 99.8% unnecessary (only need 1 file) |
| **CI Queue** | Blocked other workflows unnecessarily |
| **Time/month** | ~20 minutes wasted on version syncs |
The code being tested had already passed the full test suite when originally pushed to `main`. Re-running tests for a version number change provided no additional value.
## Decision
Implement a lightweight **version-sync-only workflow** that:
1. **Triggers automatically** after successful production deployments (via `workflow_run`)
2. **Updates only** the `package.json` version in the test deployment directory
3. **Restarts PM2** with `--update-env` to refresh version metadata
4. **Completes in ~30 seconds** instead of 5-7 minutes
### Architecture
```yaml
# .gitea/workflows/sync-test-version.yml
on:
workflow_run:
workflows: ['Deploy to Production']
types: [completed]
branches: [main]
jobs:
sync-version:
if: ${{ gitea.event.workflow_run.conclusion == 'success' }}
steps:
- Checkout latest main
- Update test package.json version
- PM2 restart with --update-env
- Verify version in PM2 metadata
```
**Key Points:**
- `deploy-to-test.yml` remains **unchanged** (runs normally for code changes)
- `deploy-to-prod.yml` remains **unchanged** (no explicit trigger needed)
- `workflow_run` automatically triggers after production deployment
- Non-blocking: production success doesn't depend on version sync
## Consequences
### Positive
**90% faster** version synchronization (30 sec vs 5-7 min)
**95% less CPU** usage (no test suite execution)
**99.8% less file I/O** (only package.json updated)
**Cleaner separation** of concerns (version sync vs full deployment)
**No workflow file pollution** with conditionals
**Saves ~20 minutes/month** of CI time
### Negative
⚠️ Test environment version could briefly lag behind production (30 second window)
⚠️ Additional workflow file to maintain
⚠️ PM2 version metadata relies on `--update-env` working correctly
### Neutral
- Full test deployment still runs for actual code changes (as designed)
- Version numbers remain synchronized (just via different workflow)
- Same end state, different path
## Alternatives Considered
### 1. Add Conditionals to `deploy-to-test.yml`
**Approach:** Detect version bump commits and skip most steps
**Rejected because:**
- Pollutes workflow with 20+ `if:` conditionals
- Makes workflow complex and brittle
- Still queues a workflow run (even if it exits early)
- Harder to maintain and debug
### 2. Manual Production Deployments
**Approach:** Make production `workflow_dispatch` only (manual trigger)
**Rejected because:**
- Removes automation benefits
- Requires human intervention for every production deploy
- Doesn't align with CI/CD best practices
- Slows down deployment velocity
### 3. Don't Sync Test Versions
**Approach:** Accept that test version lags behind production
**Rejected because:**
- Version numbers become meaningless in test
- Harder to correlate test issues with production releases
- Loses visibility into what's deployed where
### 4. Release Tag-Based Production Deployments
**Approach:** Only deploy production on release tags, not on every push
**Rejected because:**
- Changes current deployment cadence
- Adds manual release step overhead
- Doesn't solve the fundamental problem (version sync still needed)
## Implementation
### Files Created
- `.gitea/workflows/sync-test-version.yml` - Lightweight version sync workflow
### Files Modified
- None (clean separation, no modifications to existing workflows)
### Configuration
No additional secrets or environment variables required. Uses existing PM2 configuration and test deployment paths.
## Verification
After implementation:
1. **Production deployment** completes and bumps version
2. **Version sync workflow** triggers automatically within seconds
3. **Test PM2 processes** restart with updated version metadata
4. **PM2 list** shows matching versions across environments
```bash
# Verify version sync worked
pm2 jlist | jq '.[] | select(.name | endsWith("-test")) | {name, version: .pm2_env.version}'
```
Expected output:
```json
{
"name": "flyer-crawler-api-test",
"version": "0.16.2"
}
{
"name": "flyer-crawler-worker-test",
"version": "0.16.2"
}
```
## Monitoring
Track workflow execution times:
- Before: `deploy-to-test.yml` duration after prod deploys (~5-7 min)
- After: `sync-test-version.yml` duration (~30 sec)
## Rollback Plan
If version sync fails or causes issues:
1. Remove `sync-test-version.yml`
2. Previous behavior automatically resumes (full test deployment on version bumps)
3. No data loss or configuration changes needed
## References
- Inspired by similar optimization in `stock-alert` project (commit `021f9c8`)
- Related: ADR-061 for PM2 process isolation and deployment safety
- Workflow pattern: GitHub Actions `workflow_run` trigger documentation
## Notes
This optimization follows the principle: **"Don't test what you've already tested."**
The code was validated when originally pushed to `main`. Version number changes are metadata updates, not code changes. Treating them differently is an architectural improvement, not a shortcut.