Server: subscription API returns active when smart_mode_free is true.
Admin console: add toggle switch in settings page under feature flags.
Default: true (free for all users during promotion period).
- New grantTrialIfEligible() in subscription.dao.js with SELECT FOR UPDATE
to prevent concurrent bind race condition granting double trials
- confirmBind + mockBind now call the shared function (was duplicated)
- Trial days no longer hardcoded to 7 in binding — respects settings
confirmBind and mockBind now check trial history before granting trial.
Previously only checked for active subscriptions — if trial expired,
rebinding would create a new 7-day trial, allowing infinite free usage.
Now checks if user ever had a trial (any status), skips if so.
- httpsRequest: 15s timeout, prevents hanging on WeChat API outage
- fetchPlatformCertificates: keeps old certs if refresh fails,
only throws if no certs at all (first time failure)
- Replace literal '\n' with real newlines when reading private key from env
- Add PEM header validation to catch format errors early
- File-based key (privateKeyPath) unaffected
- closeStaleOrders: marks payment orders older than 24h as 'closed'
- closeExpiredPending: marks expired binding requests (bind_status=3) as cancelled
- Both run on app startup (SCF cold start), no cron needed
- countPendingByUser already filters to 2h window (previous commit)
CRITICAL fixes:
- C1: Notify amount validation now unconditional (was skippable if amount field missing)
- C2: Sync endpoint validates payer_total against order amount before activation
- C3: Order ID uses crypto.randomBytes(6) instead of Math.random (collision-safe)
HIGH fixes:
- H1: Payment endpoints rate limited to 5/min per IP
- H2: Max 5 pending orders per user, reject new ones until completed/cancelled
- H3: Purchase endpoint returns error (not mock) when wxpay unconfigured in production
Also fixed:
- Notify handler asserts Buffer body, rejects non-Buffer (L3)
- Notify error response is generic, no internal message leak (M1)
- Private key cached in memory after first read (L2)
- Fixed duplicate paymentOrderDao const declaration
1. Move notify route before authMiddleware (WeChat callback has no JWT)
2. Add await to verifyNotifySignature call (was fire-and-forget)
3. Remove dangerous verify fallback — all signature failures now throw
4. Payment sync polls 3x before giving up, never redirects to success
page unless confirmed paid
5. Production guard enforces all WX_MCH_* env vars on startup
Server:
- New lib/wxpay.js: native crypto RSA-SHA256 signing, JSAPI prepay,
AES-256-GCM notify decryption, order query (no npm deps)
- New dao/payment-order.dao.js: createOrder, markPrepay,
markPaidAndActivateSubscription (idempotent + transactional)
- New routes/payment.js: GET order status, POST order sync
- New routes/payment-notify.js: WeChat async callback handler with
signature verification, amount/appid/mchid validation
- Modified subscription/purchase: auto-detects wxpay config, returns
real payment_params or mock fallback
- Schema: payment_orders table with out_trade_no unique key
- app.js: express.raw() for notify path, payment routes mounted
- config.js: wxpay block with 7 env vars
- .env.example: all WeChat Pay fields documented
- .gitignore: certs/, *.pem, *.p12
Miniprogram:
- subscribe-plans doPurchase: calls real purchase API, falls back to
mockPurchase only when server returns mock:true
- Added syncAndRedirect for post-payment order confirmation
- api.js: getPaymentOrder, syncPaymentOrder
- Removed "模拟支付"/"测试环境" from UI text
Miniprogram:
- Add BLE reconnect button on home page when disconnected
- Add loading states to index, profile, subscribe-plans pages
- Add profile editing (avatar + nickname) with COS upload
- Enable pull-to-refresh on history page
- Fix auto-scan self.options → self.data inconsistency
- Add console.error to silent catch blocks
- Gate 'Simple Peripheral' BLE scan behind __DEV__ flag
- Add production config comment to env.js
Admin console:
- Add empty state '暂无数据' to all 5 list views
- Replace plain text plan input with select dropdown
- Add type="date" to record date filters
- Add production config comment
Server:
- CORS origin restricted in production (env CORS_ORIGIN)
- DB pool size configurable via DB_POOL_SIZE env var
- .env.example updated with WeChat Pay + production fields
- RecordView passes user_id as separate filter param, not as keyword
- Admin records endpoint forwards user_id to DAO
- treatmentDao.listAdmin supports exact user_id match filter
- Added visible filter tag with clear button in RecordView
- Always sync treatment end (not just early stop), pass start/end times
- treatment-done uses api.syncTreatment instead of raw http.post
- Move getApp() from module level to onLoad in treatment-done
- Avatar upload only returns URL, no longer updates profile directly
- COS CDN domain configurable via COS_CDN_DOMAIN env var
- Fix operator precedence in request.js token expiry check
- Remove unused result variable from COS putObject
- Replace chooseAvatar button with chooseMedia (more reliable, works on all versions)
- COS avatar URL uses tx.vsai.net.cn instead of generated bucket domain
- Phone shows '已授权' only (no actual number displayed)
- Keep body as Buffer instead of utf8 string for multipart/form-data
- Set content-length header for multer compatibility
- Phone authorization shows '已授权' instead of actual number
- loadProfile() now returns Promise (was missing return)
- Avatar upload via COS instead of storing WeChat temp path
- Add POST /user/avatar endpoint with multer + COS SDK
- Incomplete registration detection: redirect to register if phone is empty
- New register page with chooseAvatar, nickname input (random default),
and required phone number authorization via getPhoneNumber
- Login simplified: removed deprecated wx.getUserProfile, checks
is_new_user flag to redirect new users to registration
- Server login response now includes is_new_user field
- Add expire_time > NOW() filter to findActive() preventing stale subscriptions
- Add express-rate-limit on login endpoints (user: 10/15min, admin: 5/15min)
- Add production guard for default admin credentials
- Fix BLE bindDevice userId encoding (uint32 instead of hexToBytes on numeric)
- Wrap adminCreate in transaction to prevent race condition
- Add settings cache invalidation after admin saves
- Read trial_days from settings instead of hardcoding 7
- Fix double JSON.stringify in commandDao.finish call
- Cancel stale pending bindings before creating new ones
- Reduce token refresh grace period from 3 days to 1 day
- Fix subscribe-success to fetch expiry from server (correct for renewals)
- Add keep-alive name property to DashboardView and SettingsView
- Fix BLE disconnect() to preserve listener registrations across reconnects
Server:
- Add settings-cache with 60s TTL for feature toggle checks
- Enforce maintenance_mode on login, enable_binding on device bind
Admin console:
- Remove dead "发送通知" button from user detail
- Firmware check calls real API and compares versions
Miniprogram:
- Wear-check: dynamic battery/connected from BLE state
- Login: hide non-functional phone auth button
- Treating: show actual selected regions instead of hardcoded "全脸"
- Index: display subscription status with tap to manage
- Auto-scan: show "待检测" instead of "--" for PD data
- Agreements: tap shows "内容建设中" modal
- Subscription purchase now extends expire_time when user has active
subscription, instead of cancelling and replacing
- Admin subscription creation uses same extend logic
- Subscribe page shows "续费" button and extend message for renewals
- Help and contact pages: add pink header, centered icon + text
POST /api/v1/subscription/purchase only creates an order without
activating. Add POST /api/v1/subscription/mock-purchase that does
purchase + verify in one step (non-production only). Miniprogram
subscribe page now calls mock-purchase so subscriptions take effect.
- Add GET /api/v1/subscription/plans public endpoint for pricing
- Subscription API now returns trial_used field
- Subscribe-plans page shows trial card, greyed out with "已使用" tag
when trial has been used
- Trial activation calls dedicated trial API, not purchase
- Prices fetched from server settings, fallback to defaults
- Add GET /api/v1/subscription/plans public endpoint (no auth needed)
that reads prices from system_settings table
- Subscribe-plans page now loads prices from server on show
- Falls back to hardcoded defaults if API fails
- Add api.getPlans() to miniprogram API module
- utils.js: force UTC+8 output to match MySQL timezone on SCF
- admin subscriptions: return stats (plan counts + revenue) in response
- admin devices: implement keyword search filter
- device-detail: fix binding history field names (bind_time/unbind_time/bind_status)
- user-detail: fix treatment time field (start_time not started_at)
- dashboard: show actual mode per treatment, use sub_stats from dashboard API
Miniprogram:
- Profile "我的设备": show unbind option when device bound, scan when not
- Add device status indicator (已绑定/未绑定) to profile menu
Admin console:
- Login: add @confirm to inputs so Enter key submits the form
- Record detail: wire up 详情 link with modal showing full record info
- Dashboard: add subscription stats row (月卡/年卡/试用/收入)
- Subscription: add 取消 action for active subscriptions
- Format: fix -8h timezone display for UTC ISO date strings
Server:
- POST /api/v1/admin/subscriptions/cancel: cancel active subscriptions
- Dashboard API: include sub_stats (plan counts + monthly revenue)
- IP extraction: add X-Forwarded-For fallback for logging
The bind + confirm two-step flow has timezone issues between Node.js
(UTC on SCF) and MySQL (timezone +08:00) that cause bind_token to
fail validation. Add POST /api/v1/device/mock-bind that does bind +
confirm in one atomic step, skipping the token entirely. Only
available when NODE_ENV != production.
Update ble-connect and index page mock buttons to use mock-bind.
SCF runs in UTC but MySQL connection uses timezone +08:00, causing
bind_expires computed in JS to mismatch NOW() in SQL queries.
Use DATE_ADD(NOW(), INTERVAL ...) directly in SQL to guarantee
consistent timezone for both storage and comparison.
- nav-back: add text-align:left to override page-header's center
- ble-connect mock: call bind API first to get fresh token before
confirm, instead of relying on potentially stale URL params
- utils.js: use local time instead of UTC for toMysqlDate, fixing
timezone mismatch with MySQL NOW() that caused bind tokens to
appear immediately expired
Server:
- POST /api/v1/admin/password: admin password change with bcrypt migration
- POST /api/v1/subscription/trial: user trial activation, one per user
- POST /api/v1/admin/devices/batch: bulk device import (up to 500)
- Add trial plan (7 days, free) to PLANS constant
Admin console:
- Settings page: add password change form with validation
- Device page: add batch import modal with textarea input
Miniprogram:
- Treating page: add back button with stop-treatment confirmation
- Index page: add mock device bind button (dev mode only)
- Login bcrypt migration: write password_salt = '' instead of NULL,
which violated the NOT NULL constraint and caused 500 errors
- Schema: widen password_hash from CHAR(64) to VARCHAR(100) for bcrypt
compatibility, add DEFAULT '' to password_salt
- init-db: add ALTER TABLE migrations for existing databases
- app.js: enrich error logging with method, path, SQL error details
Server: add filter/search/pagination to admin list endpoints, enrich
user/device queries with JOINs and subqueries, prevent duplicate active
subscriptions on creation.
Admin console: fix record page TypeError on numeric record_id, correct
mode comparison (integer vs string), fix device detail field names and
command opcode, remove hardcoded login credentials, wire up dead buttons
(unbind, view logs, export), fix user/subscription field mappings.
Miniprogram: fix subscription status string/number mismatches across
index/treatment-setup/profile pages, fix device name field reference,
fix treatment-done null device_id by capturing at onLoad.
- bcrypt password hashing with auto-migration from SHA-256
- BLE command retry (3 attempts, 500ms delay, skip on disconnect)
- BLE auto-reconnect with service re-discovery on disconnect
- Treatment page disconnect/reconnect event handling
- Token refresh endpoint with 3-day grace period
- Client-side token auto-refresh when <24h remaining
- Single treatment record detail API with ownership check
Guoguo