aeternalabshq/pullmd-playwright repository overview

⁠PullMD Playwright Sidecar

Companion service for PullMD⁠ — renders JavaScript-heavy pages via headless Chromium so the main pullmd container can fall back to it when static-HTML extraction returns body-soup or thin output.

This image is not useful on its own. Use it together with aeternalabshq/pullmd.

⁠Endpoints

• GET /health → {"ok": true, "playwright": "<version>", "browser": "chromium"}
• POST /render body {"url": "https://…"} → text/html (rendered DOM after JavaScript execution)
• Returns 503 Retry-After: 5 when the concurrency limit (default 4) is saturated, 504 on render timeout (20s hard cap).

⁠Wiring

In your docker-compose.yml, run alongside pullmd on a shared internal network and set PLAYWRIGHT_URL=http://playwright:8002/render on the pullmd service:

services:
  pullmd:
    image: aeternalabshq/pullmd:latest
    environment:
      - PLAYWRIGHT_URL=http://playwright:8002/render
    depends_on:
      - playwright
    networks:
      - pullmd-internal

  playwright:
    image: aeternalabshq/pullmd-playwright:latest
    networks:
      - pullmd-internal

networks:
  pullmd-internal:
    driver: bridge

⁠Identity

The rendered request has to be internally consistent. Chromium builds its Sec-CH-UA header from its own identity, and overriding the User-Agent on the browser context does not touch it — so by default a request goes out claiming Chrome/147 in one header and "HeadlessChrome";v="131" in the next. Bot rules read the second one: holding everything else constant and swapping only that brand token is the difference between 200 and 403 on hosts behind a bot manager.

The sidecar therefore rewrites the Client-Hints metadata to match the User-Agent it sends (brand, version, platform, mobile flag). For a non-Chromium User-Agent — the iPhone profile from mobileUa — it strips the Sec-CH-UA* headers instead, because Safari sends no Client Hints and a fabricated one would be a fresh contradiction. Both are best-effort: a failure here is logged and the render proceeds.

⁠Resource footprint

• Image size: ~3.7 GB (Microsoft Playwright base image bundles Chromium, Firefox, WebKit binaries)
• Idle RAM: ~250 MB (one persistent browser instance)
• Per-render: brief spike, freed when the per-request browser context closes

⁠Wait strategy

page.goto(url, wait_until="domcontentloaded") then wait_for_load_state("networkidle") (5s soft-fail). 20-second hard timeout via asyncio.wait_for.

⁠Tags

• latest — main branch
• 1.1.2, 1.1, sha-<short> — released versions

Multi-arch: linux/amd64, linux/arm64.

⁠Source

github.com/AeternaLabsHQ/pullmd⁠ — see playwright-sidecar/ for the source.

⁠License

AGPL-3.0-or-later⁠.