Add Extra Playwright Servers (per-watch selectable self-hosted CDP browser servers)

[geekho-me]

What

Adds an Extra Playwright Servers feature: configure multiple self-hosted Playwright/CDP browser servers in Settings → Captcha/Proxies, and select which one to use per-watch via the existing "Fetch Method" dropdown. It mirrors the existing "Extra Browsers" feature but is dedicated to self-hosted browser servers (e.g. browserless / sockpuppetbrowser).

Why / use case

Run multiple browser servers in different geolocations or configurations (for example, one behind a VPN for a specific country) and pick the right one for an individual watch — useful for geo-specific content, price/restock checks on region-locked stores, or spreading fetch load across servers.

How it works

• New global setting settings.requests.extra_playwright_servers (model/App.py).
• New SingleExtraPlaywrightServer form with ws:///wss:// URL validation, added as a FieldList to the global settings form (forms.py).
• New settings UI section in the Captcha/Proxies tab, styled to match the existing "Extra Browsers" section (settings.html, plus the SCSS/compiled CSS).
• ChangeDetectionStore.extra_playwright_servers exposes configured servers as ("extra_playwright_server_<name>", "<name>") choices, filtering out incomplete entries (store/__init__.py).
• The watch edit form offers each server as a Fetch Method (blueprint/ui/edit.py).
• In processors/base.py, a watch whose fetch_backend starts with extra_playwright_server_ resolves the configured URL and forces the Playwright fetcher (connect_over_cdp) with that connection URL.

Fix: screenshot/capability detection for prefixed fetch backends

While building this I found that get_fetcher_capabilities() (pluggy_interface.py) and Watch.fetcher_supports_screenshots (model/Watch.py) matched the watch's fetch_backend against the fetcher registry verbatim. For the prefixed backends (extra_playwright_server_* and the existing extra_browser_*) nothing matched, so capabilities fell back to all-False. The visible symptom: the preview/history "Current screenshot" tab showed "Screenshot requires a Content Fetcher … that supports screenshots" even though the screenshot had been captured.

Both functions now resolve the prefixes the same way processors/base.py does (extra_playwright_server_* → Playwright fetcher; extra_browser_* → html_webdriver). This also fixes the pre-existing extra_browser_* case.

Tests

• tests/test_extra_playwright_servers.py — Flask-client feature test: saving a server in Settings persists it, the datastore exposes/filters it, it appears as a "Fetch Method" on the edit page, and the ws:///wss:// validator rejects other schemes.
• tests/unit/test_fetcher_capabilities.py — browser-free unit tests for capability resolution of extra_playwright_server_*, extra_browser_*, plain html_requests, and unknown backends, for both get_fetcher_capabilities() and Watch.fetcher_supports_screenshots.

Docs

• docs/extra-playwright-servers.md — configuration, the ws:///wss:// CDP requirement, compatible servers, and a test compose stack.

Compatibility note

changedetection connects with Playwright's connect_over_cdp() (Chrome DevTools Protocol) at the root path, so a compatible server is a CDP gateway that accepts a connection at / — e.g. browserless or sockpuppetbrowser. The upstream microsoft/playwright run-server (Playwright wire protocol) is not compatible with connect_over_cdp and is documented as such.

Notes for reviewers

• A docker-compose.test-playwright.yml helper stack (browserless + sockpuppetbrowser) is included to exercise the feature; happy to drop it if you'd prefer it out of the repo.
• The compiled static/styles/styles.css change is the build output of the SCSS change (it comma-extends the existing extra-browsers rules to the new IDs).
• The translation catalog (messages.pot / .po) was regenerated with python setup.py extract_messages && update_catalog && compile_catalog.

🤖 Generated with Claude Code

[dgtlmoon]

this already exists?

[AmirF194]

This is a nicely scoped feature that reuses the existing custom-connection path rather than adding a parallel one. A selected extra_playwright_server_<name> resolves to the same custom_browser_connection_url that extra_browsers already feeds into playwright.py connect_over_cdp(...), and the only real delta is forcing the Playwright fetcher via the new use_playwright_fetcher flag, which mirrors the existing browser-steps override. Backward compatibility is handled the proven way: the setting defaults to [] in model/App.py and every reader uses .get('extra_playwright_servers', []), matching exactly how extra_browsers was introduced, so existing watches need no migration. The capability fix is a real catch and correctly factored, and it also fixes a pre-existing extra_browser_* false-negative on the screenshot tab.

No blocking concerns from me. The one edge behavior worth knowing: if a server is deleted or renamed after a watch selects it, the filter returns empty and it silently falls back to html_requests (processors/base.py:172-174). That is the same silent downgrade extra_browsers already has, so it is not a regression, just a candidate for a future UX note.

On CI: the red run is environmental, not this feature. basic-tests failed on two different unrelated tests across the interpreters (3.11 on test_favicon_inline_data_uri, 3.14 on test_obfuscations with a transient 500), and this PR changes no favicon, obfuscation, or price code (those strings only appear in regenerated .po catalogs). Different failure per runner plus a transient 500 is the flaky signature, and the run is from late May against what is now a stale base, so a rebase and re-run should go green. Every fetcher-relevant suite (unit including the new ones, playwright, custom-browser, selenium, proxy, upgrade-path) passed.

Optional: the actual call_browser routing (that selecting a server sets the connection URL and forces Playwright) is the one untested surface. Everything around it is covered, but in a 29-file PR that is the behavior most likely to regress silently. A small unit test that mocks the fetcher and asserts use_playwright_fetcher plus the connection URL would close it, which the test-file comment already acknowledges.