Skip to content

docs(handoff): dl-router Discord media β€” three PRs merged and shipped, and the consumer is MEASURED not to be running them - #1236

Open
ZacxDev wants to merge 3 commits into
mainfrom
docs/handoff-dl-router-discord-media
Open

docs(handoff): dl-router Discord media β€” three PRs merged and shipped, and the consumer is MEASURED not to be running them#1236
ZacxDev wants to merge 3 commits into
mainfrom
docs/handoff-dl-router-discord-media

Conversation

@ZacxDev

@ZacxDev ZacxDev commented Sep 2, 2026

Copy link
Copy Markdown
Member

New effort, new doc β€” the two existing dl-router handoffs are closed retrospectives of the July/August build-out with dated slugs, not this work.

Discord is 59% of dl-router's live traffic (20 of 34 routed downloads) and the routing half has been Discord-specific since the first evening β€” but the save side was wrong for it in three ways. All three merged and shipped:

PR squash what
#1110 911af220 save the file that was POSTED, not the proxy thumbnail; a stable ledger key for a signed CDN URL; Discord attachments bypass the yt-dlp branch
#1113 4fdead27 the sibling discord-embed-ext doc β€” crop VERIFIED LIVE, plus three corrections
#1149 67f674f6 media may be an ORDERED LIST of accessors, so one rule covers an anchor-linked image AND a direct video

The part that matters most

πŸ”΄ Deployed β‰  running. After the operator reported reloading, I measured Brave's own record:

dl-router bytes in the repo tree 0.3.2, written 08-31 02:41
last reloaded (Brave's Preferences) 08-30 21:39 β€” five hours before those bytes existed
Brave main process up since 08-28 β€” no restart at all

Brave is still running 0.3.1. That is what the manifest bump exists to detect, and it is rank 1 in the doc with forcing: user.

Open investigation carried forward

Whether preferOriginalUrl is a no-op in production. The defect it fixes is proven (pre-change code demonstrably downloads …?format=webp&width=550), but all 20 live Discord rows are origin-host, so the proxy branch has no live instance behind it β€” and the extension has no logging facility, so the question has no data-driven answer. The doc carries the DOM probe that settles it, and both eliminations are tagged via: measurement with the untested part named explicitly.

Also recorded

The two-writers-one-reader ledger seam and why buildMatchPayload is deliberately not folded in (pinned by three pre-existing tests); why store.source_url_key keeps the query on purpose and Discord is the exception; the hand-copied MAX_MEDIA_ACCESSORS that left both suites green when only one side moved; that a menu test asserting only downloads.length === 0 is vacuous (hit twice in one PR); and the asymmetric deploy order β€” switch first, rule second, or a list rule against the old sidecar takes every routing endpoint to 503.

Docs-only: one file.

πŸ€– Generated with Claude Code

…#1110 save the posted file no

Claude-Session-Id: 604aabe0-8379-4a24-add1-20f32d7011fd
…it is now answered by measureme

Claude-Session-Id: e13f042e-3be9-449d-a597-4b9a7ea7d259
…very round after the first foun

Claude-Session-Id: e13f042e-3be9-449d-a597-4b9a7ea7d259
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant