deploy-verify（部署验证）

Deploy & Verify

Deploy Cloudflare Workers to an environment and verify the changes work by inferring what to test from recent git changes.

Workflow

1. Pre-deploy checks
2. Deploy to target environment
3. Infer verification plan from git diff
4. Run verification
5. Report results (pass/flag issues)

Step 1: Pre-Deploy Checks

Before deploying, verify:

1. Types pass — tsc --noEmit or project equivalent
2. Tests pass — npm test / bun test / project equivalent
3. No uncommitted changes — git status should be clean (warn if dirty)
4. Detect environment config — read wrangler.toml / wrangler.jsonc for available environments

# List available environments from wrangler config
grep -E '^\[env\.' wrangler.toml | sed 's/\[env\.\(.*\)\]/\1/'

Step 2: Deploy

# To a named environment (staging, preview, etc.)
wrangler deploy --env <environment>

# To production (default if no --env)
wrangler deploy

If the project has multiple Workers (monorepo), detect which one changed:

• Check git diff --name-only for paths matching Worker directories
• Only deploy the Worker(s) that changed

Step 3: Infer Verification Plan

This is the key step. Look at what changed and determine what to verify:

# What changed since last deploy?
git diff HEAD~1 --name-only
git diff HEAD~1 --stat
git log HEAD~1..HEAD --oneline

Inference rules:

Step 4: Run Verification

Use curl or fetch to test the deployed URL:

# Basic health check
curl -s -o /dev/null -w "%{http_code}" https://<worker-url>/

# Check specific endpoint with response body
curl -s https://<worker-url>/api/endpoint | jq .

# Check response headers
curl -sI https://<worker-url>/api/endpoint

# POST with body
curl -s -X POST https://<worker-url>/api/endpoint \
  -H "Content-Type: application/json" \
  -d '{"key": "value"}'

Also check wrangler logs for errors after hitting endpoints:

# Tail logs (run in background, hit endpoints, then check)
wrangler tail --env <environment> --format json

Step 5: Report

If all checks pass:

Deploy verified:
- Environment: staging
- URL: https://my-worker-staging.example.workers.dev
- Checks passed:
  - GET /api/stories → 200, response shape correct
  - POST /api/generate → 200, returns stream
  - KV read/write → working

If issues found (flag, don't rollback):

Deploy issues found:
- Environment: staging
- URL: https://my-worker-staging.example.workers.dev
- PASS: GET /api/stories → 200
- FAIL: POST /api/generate → 500
  - Error in logs: "Missing AI binding"
  - Likely cause: AI binding not configured in staging env
- Action needed: Check wrangler.toml [env.staging] AI bindings

Do NOT automatically rollback. Flag the issues and let the user decide.

Multi-Environment Patterns

Common setup:

# wrangler.toml
name = "my-worker"

[env.staging]
name = "my-worker-staging"
route = "staging.example.com/*"

[env.production]
name = "my-worker"
route = "example.com/*"

Default flow: Deploy to staging → verify → user promotes to production. Direct to prod: Only when user explicitly asks. Still run verification after.

Secrets

If verification fails with auth/config errors, check that secrets match between environments:

wrangler secret list --env staging
wrangler secret list --env production

Secrets don't copy between environments. A common gotcha after adding a new env.

Troubleshooting

Gotchas

• Always verify the deployed version matches your expected commit. Run wrangler deployments list to confirm your code actually deployed. Cached old versions are a common false trail.
• Don't strip console.log in production builds. You need logs for debugging — if you can't see what's happening, you can't fix what's broken. Keep logging in production, especially during early development.
• Secrets don't copy between environments. Adding a secret to production doesn't add it to staging. After creating a new environment, run wrangler secret list --env <env> to verify ALL required secrets are set.
• Check both .dev.vars AND GitHub Actions secrets. They are independent — a secret in .dev.vars only works locally. GH Actions needs its own secrets configured.
• Deployed URL might not match what you expect. If using custom domains, verify the route pattern in wrangler.toml matches the URL you're testing. Preview deployments use different URLs than production.
• After deploy, wait 30 seconds before testing. Workers edge cache can serve stale responses briefly. If you see old behavior, wait and retry before debugging.