hotel-auto-x-ctrip-collector

name	hotel-auto-x-ctrip-collector
description	Use when working on hotel-auto-x Ctrip collection, including Ctrip review collection, bad reviews across ctrip/qunar/tongcheng/zhixing, review score, full daily collection, realtime Ctrip metrics, Ctrip session/profile problems, and Ctrip data mapping into daily_report.

hotel-auto-x Ctrip Collector

Use this skill for Ctrip collection work in /home/qing/hotel-auto-x/hotel_auto_x.

Safety

• Do not run full all-store collection during busy scheduler windows unless explicitly requested.
• Respect collect_ctrip=False; some stores intentionally do not collect Ctrip.
• Use stealth=False for Ctrip contexts because stealth can interfere with Ctrip micro-frontend/XHR behavior.
• Do not write directly to daily_report or bad_reviews; use process_result().

Project Map

• Main collector: platforms/ctrip.py::CtripCollector
• Realtime collector: platforms/ctrip_realtime.py::collect_ctrip_realtime
• Ad collector: platforms/ad_ctrip.py::collect_ctrip_ad
• Review scheduler: scheduler/engine.py::task_collect_ctrip_review
• Full scheduler: scheduler/engine.py::task_collect_ctrip_full
• Realtime scheduler: scheduler/realtime_engine.py::task_realtime_ctrip
• Ad scheduler: scheduler/ad_engine.py::task_collect_ctrip_ad
• ETL: pipeline/etl.py

Main Data Types

Daily/full collection can fill:

• Ctrip visitor/order/heat/ranking/sold-room metrics.
• Ctrip review score, PSI, reply rate, collect count/rank.
• Bad reviews from Ctrip-family platforms.

Bad reviews may use platform values such as ctrip, qunar, tongcheng, or zhixing in bad_reviews. For blackhole push payloads, keep source as ctrip and put the sub-channel in row platform.

Review-Only Flow

Production entry: CtripCollector().run_review_only(store_id, stat_date).

Scheduler path:

1. Filter active stores with ctrip_account and collect_ctrip=True.
2. Skip circuit-open, expired-session, or locked Profile stores.
3. Run collector with timeout.
4. Call process_result(result).
5. Mark session active/expired.
6. Retry retryable failures once.

Full Flow

Production entry: CtripCollector().run(store_id, stat_date).

The collector captures API responses and has DOM fallback parsing for important quality/overview values. Treat DOM-only results as useful but verify missing fields against captured API changes.

Realtime Flow

Production entry: collect_ctrip_realtime(store_id).

Validate at least one core field:

• ctrip_orders
• ctrip_room_nights
• ctrip_visitor
• ctrip_rank

api_empty/no_data should be investigated by checking login state, hotel identity, request params, and whether the selected store is configured to collect Ctrip.

Ctrip-Family Channel Rule

For weekly OTA data, if a store normally collects Ctrip, then all-zero Ctrip/Qunar/Tongcheng room-night values are suspicious unless the store is explicitly excluded. Do not use PMS to fill OTA room-night facts.

Minimal Diagnostic Pattern

from platforms.ctrip import CtripCollector

result = CtripCollector().run_review_only("MD00001", "2026-07-06")
print(result.success, result.error, result.method, result.data)

Only write with:

from pipeline.etl import process_result
process_result(result)

when the user explicitly wants data repaired or imported.

Failure Meaning

• session_expired: re-login the store Profile.
• anti_bot: stop aggressive retries and inspect browser state.
• api_empty or no_data: inspect endpoint response, selected hotel identity, date range, and account permissions.
• Missing sub-platform room nights: inspect Ctrip data-center channel-tab API rather than PMS.

Reporting

Report store_id, target date, whether collect_ctrip is enabled, phase, method, core fields, sub-platform bad-review counts, and ETL/write status.