0 Fork 0

mirror of https://github.com/tiennm99/store-scraper-bot.git synced 2026-10-11 03:13:50 +00:00

Files

T

tiennm99 797f040574 feat: Cloudflare Workers code port (deploy pending)

Refactors source to be Worker-shaped. No live deploy yet — sister deploy plan
runs Atlas provisioning + smoke later.

- wrangler.toml with nodejs_compat_v2, daily UTC 0 cron (= 7am Asia/Ho_Chi_Minh)
- package.json: drop node-telegram-bot-api, node-cron, dotenv, pino,
  pino-pretty; add wrangler devDep; bump to 0.2.0
- src/bot/telegram-api.js: raw fetch wrapper for Telegram Bot API
- src/bot/dispatch.js: per-message dispatcher extracted from polling loop
- src/repository/mongodb.js: memoized MongoClient per warm isolate, typed
  MongoUnavailable error, fast-fail timeouts
- src/repository/store.js: factory binding env once
- All 4 repositories converted to factory shape
- src/api/{apple,google}-scraper.js: take store instead of importing repos
- src/index.js: Worker entry exporting { fetch, scheduled }; webhook validates
  X-Telegram-Bot-Api-Secret-Token; ack-then-waitUntil pattern
- src/scheduler/scheduler.js: trimmed; runDailyCheck only (no node-cron)
- src/config.js, src/logger.js: env-driven, console.log JSON output
- scripts/register-webhook.js: setWebhook + setMyCommands; --dry-run supported
- scripts/check-secret-leaks.js: lint blocks console.log(env.<SECRET>)
- plans/260426-2015-cloudflare-worker-code-port: this code port plan
- plans/260426-2327-cloudflare-deploy-and-smoke: sister deploy plan

Validated via node --check on all 32 source files; lint clean. Real deploy
gates (bundle size, cold-start CPU) run in deploy plan.

2026-04-26 23:36:39 +07:00

3.2 KiB

Raw Blame History

Deploy Phase 02 — First Deploy + Hard Gates

Overview

Priority: P0 — gates can abort the entire plan.
Status: pending
Description: Bundle-size dry-run, real deploy, mongo-ping CPU gate, auto-pause behavior test.

Implementation Steps

Bundle-size hard gate:
```
npx wrangler deploy --dry-run --outdir=.tmp-deploy
du -sh .tmp-deploy
```
- PASS: ≤ 2.7 MiB. Continue.
- FAIL: > 2.7 MiB. Stop. Pivot to Upstash (separate plan; out of scope here).
First real deploy:
```
npx wrangler deploy
```
Capture the URL printed (e.g. https://js-store-scraper-bot.<account>.workers.dev). Update .env.deploy with this WORKER_URL.

Add temporary /__mongo-ping route for CPU gate (revert before commit at end of phase):

Edit src/index.js fetch handler — top of method, before path routing:

if (request.method === 'GET' && new URL(request.url).pathname === '/__mongo-ping') {
  const t = Date.now();
  const { db } = await getMongo(env);
  await db.command({ ping: 1 });
  return new Response(JSON.stringify({ wall_ms: Date.now() - t }), {
    headers: { 'Content-Type': 'application/json' },
  });
}

npx wrangler deploy.

Cold-start CPU gate (operator):
- Run 5 cold cycles, 10+ min apart:
```
curl https://js-store-scraper-bot.<account>.workers.dev/__mongo-ping
```
- Open CF dashboard → Workers → js-store-scraper-bot → Logs / Metrics.
- Note CPU time per invocation (separate from wall_ms).
- PASS: cold CPU < 40ms (10ms safety under 50ms cap).
- FAIL: ≥ 40ms. Decide: upgrade to Workers Paid ($5/mo, 50ms baseline → 5min ceiling), or pivot to Upstash.
Auto-pause behavior test:
- Atlas UI → cluster → Pause.
- Wait ~30s for Atlas to actually pause.
- curl .../__mongo-ping → should return 5xx within ~5s, NOT hang.
- Atlas UI → resume cluster.
- Verify: error was catchable (logged via wrangler tail); not a hang.

Cleanup: revert /__mongo-ping route in src/index.js. Re-deploy:

git checkout src/index.js   # revert the temp ping route
npx wrangler deploy

Record baseline in docs/using-mongodb.md (created in Phase 04): cold-ping P95 wall_ms.

Todo List

Bundle ≤ 2.7 MiB
First deploy success; URL captured
.env.deploy WORKER_URL set
Temporary /__mongo-ping deployed
Cold CPU < 40ms across 5 cold cycles
Auto-pause yields catchable error
Temp route reverted + redeployed
Baseline P95 captured for docs

Success Criteria

wrangler deploy clean.
All hard gates pass.
Atlas dashboard "Current Connections" stays low (≤ 5) during ping smoke.

Risk Assessment

Risk	Likelihood	Impact	Mitigation
Bundle exceeds 2.7 MiB	M	Plan-aborting	Pivot path documented in plan.md
Cold CPU > 50ms	M	Plan-aborting on Free	Escalate to Paid; or pivot
Atlas connection refused (`0.0.0.0/0` not yet propagated)	L	Medium	Wait 1-2 min; retry
Forgot to revert `/__mongo-ping`	L	Low	Step 6 explicit; lint adds no help here

Next Steps

Blocks: Phase 03.
Unblocks: Phase 03.

3.2 KiB Raw Blame History