apunkt/SurfSense
1
0
Fork 0
mirror of https://github.com/MODSetter/SurfSense.git synced 2026-07-22 23:31:12 +02:00
Code Issues Projects Releases Packages Wiki Activity
SurfSense/surfsense_backend/app/proprietary/platforms/walmart
History
CREDO23 f6970b90e8 fix(walmart): parse category breadcrumb path into breadCrumbs and category
2026-07-19 09:10:03 +02:00
..
__init__.py feat(walmart): add scrape and review orchestration 2026-07-19 08:19:22 +02:00
fetch.py feat(walmart): add fetch with proxy rotation and block detection 2026-07-19 08:19:21 +02:00
next_data.py feat(walmart): add __NEXT_DATA__ extraction 2026-07-19 08:19:21 +02:00
parsers.py fix(walmart): parse category breadcrumb path into breadCrumbs and category 2026-07-19 09:10:03 +02:00
README.md feat(walmart): add scrape and review orchestration 2026-07-19 08:19:22 +02:00
schemas.py feat(walmart): add product and review schemas 2026-07-19 08:19:21 +02:00
scraper.py feat(walmart): add scrape and review orchestration 2026-07-19 08:19:22 +02:00
url_resolver.py feat(walmart): add url resolver 2026-07-19 08:19:21 +02:00

README.md

Walmart Scraper

Two verbs read Walmart's public, anonymous pages: walmart.scrape (products + search/category/browse listings, per-product billing) and walmart.reviews (deep paginated reviews, per-review billing).

Scope

The scraper reads public pages available to anonymous visitors — no login, no account cookies. Data is extracted from the Next.js __NEXT_DATA__ JSON blob embedded in each page (with an __APP_DATA__ fallback), not from the rendered DOM, because Walmart obfuscates CSS classes and A/B-tests layout constantly.

walmart.scrape returns a free sample of on-page reviews (rating distribution, aspects, top reviews) under reviewsSample. walmart.reviews fetches the full review history from the public /reviews/product/{usItemId} page, which robots.txt permits (unlike /search).

Architecture

  • schemas.py defines the stable input, product, review, and error models.
  • url_resolver.py classifies product (/ip/) vs listing (/search, /cp/, /browse/) URLs and extracts the numeric usItemId.
  • next_data.py extracts and navigates the hidden Next.js JSON state.
  • fetch.py owns proxy-aware HTTP access (US-pinned), block detection, and retries.
  • parsers.py contains pure, defensive JSON parsers.
  • scraper.py coordinates discovery, enrichment, pagination, concurrency, limits, and in-stream error items.

Anti-bot

Walmart runs Akamai (edge/TLS) + PerimeterX/HUMAN (behavioral JS). Requests go through US residential proxies with TLS-impersonated headers; blocked responses (body markers, 412/429/503, or the 200-OK CAPTCHA body) rotate to a fresh proxy exit.

Known ceilings and upgrade paths (see fetch.py / scraper.py ponytail: notes): reviews page at 10/page; search capped at Walmart's 25-page limit; session warming (seed _px3/_pxhd on a sticky exit) is the next lever if block rates on the SSR pages climb, and the /orchestra/* GraphQL API is deliberately avoided (rotating persisted-query hashes make it brittle).

Verification

Offline fixtures cover the parsers and both flows:

uv run pytest tests/unit/platforms/walmart/

A manual live check (requires network + residential proxy):

uv run python scripts/e2e_walmart_scraper.py