WildDragonLLC/claudecodeui
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
claudecodeui/claude-data/projects/-home-node/memory/project_wild_dragon.md

7.9 KiB
Raw Blame History

name description type originSessionId
Dragonflight MAM platform Self-hosted media asset management replacing Grass Valley AMPP FramelightX. Now deployed directly on TrueNAS (zampp1/zampp2 retained as remote DeckLink workers). Repo renamed to forge.wilddragon.net/zgaetano/dragonflight on 2026-05-23. project 544a289a-0493-4194-9fbd-112ed250e221

Self-hosted MAM platform under active development. Stack: Node.js/Express, vanilla HTML/CSS/JS frontend, PostgreSQL 16, Redis 7 + BullMQ, S3-compatible (RustFS), FFmpeg, Blackmagic DeckLink SDK, Docker Compose. Repo: forge.wilddragon.net/zgaetano/dragonflight (renamed from wild-dragon on 2026-05-23 — Forgejo handles redirects for the old URL). On-disk checkout on TrueNAS is still at /mnt/NVME/MAM/wild-dragon/ because docker-compose paths reference it; rename later if desired. Remote origin already updated to the new SSH URL.

Why: Replacing Grass Valley AMPP FramelightX with an in-house alternative. Test deployment is the production deployment — the user authorizes pushing directly to main and deploying to zampp1 + zampp2 for continued testing.

How to apply: Treat zampp1/zampp2 as authorized targets for MeshCentral / live deploy. Push to main directly when changes are validated. The repo's deploy/onboard-node.sh is the canonical worker provisioning path.

Cluster topology (as of 2026-05-22)

  • zampp1 — primary; runs mam-api, db, redis, web-ui, worker, capture containers via docker-compose.yml. Postgres (5432) and Redis (6379) are now published on the host for worker-node connectivity.
  • zampp2 — worker ONLY; runs node-agent (host-network, restart: unless-stopped) via docker-compose.worker.yml. Does NOT run the primary stack. Has a DeckLink Duo 2 (4 BNC ports at /dev/blackmagic/io0..io3).
  • zampp2's .env points DATABASE_URL/REDIS_URL at 172.18.91.216 (zampp1). Using .env (not .env.worker); project name is wild-dragon (not the wild-dragon-worker that onboard-node.sh would create).
  • Workers heartbeat to POST /api/v1/cluster/heartbeat; list at GET /api/v1/cluster/.
  • Cluster registry: unique index on cluster_nodes.hostname (migration 007); pickIp() in routes/cluster.js now only flags 172.17.x as docker bridge (NOT the full 172.16/12 block — real LAN 172.18.91.x was being falsely flagged).

Codec selection (recorders)

Per-recorder columns added in migration 008 (recording_* and proxy_* for video bitrate, framerate, audio codec/bitrate/channels, container). services/capture/src/capture-manager.js exports VIDEO_CODECS, AUDIO_CODECS, CONTAINER_FMT, CONTAINER_EXT catalogs. routes/recorders.js RECORDER_FIELDS whitelist passes settings as env vars to the capture sidecar via bootstrapAutoStart().

DeckLink port picker

services/web-ui/public/js/bmd-card.js (window.BMDCards) renders an SVG diagram of the card with click-to-select ports. Models registered: Duo 2, Quad 2, Mini Recorder 4K, Mini Monitor 4K, UltraStudio 4K Mini. Backed by GET /api/v1/cluster/devices/blackmagic which flattens every node's DeckLink capabilities. The recorders.html rewrite that consumes this was lost in a context-compaction event — file on zampp1 still matches the old (pre-tabs, pre-SVG) version.

LAN topology gotcha

The user's LAN is 172.18.91.0/24 — inside RFC1918 and inside Docker's reserved 172.16.0.0/12 range. zampp1=172.18.91.216, zampp2=172.18.91.217. pickIp() in routes/cluster.js was fixed (2026-05-22) to only flag 172.17.x as docker bridge, not all of 172.16/12. The smoke test in deploy/test-cluster.sh already had this fix.

Pending work (as of 2026-05-22)

  1. recorders.html UI rewrite — tabbed codec settings (Video/Audio/Container) for both master & proxy, node selector + BMD card SVG picker for SDI source. Renderer + endpoint are in main; the HTML that wires them up was lost to compaction and never reached Forgejo.
  2. GUI polish pass with flyonui MCP — explicitly the LAST priority.

DeckLink fix (issue #10, closed) — infra layer

  • recorders.js now binds /dev/blackmagic for SDI local sidecars
  • resolveNodeTarget() routes SDI sidecars to the correct remote node-agent
  • node-agent v1.2.0: POST /sidecar/start, DELETE /sidecar/:id, GET /sidecar/:id/status
  • docker-compose.worker.yml: node-agent gets /var/run/docker.sock + LIVE_DIR
  • Sidecar port scheme for remote nodes: 7438 + deviceIndex (device 0 → 7438, device 1 → 7439)
  • All deployed to zampp1+zampp2 on 2026-05-22. FFmpeg now opens the card and reads audio.

DeckLink end-to-end working (commit 30b4def, 2026-05-22)

  • services/capture/patch_decklink.py now applies services/capture/decklink-sdk16.patch (~346 lines) via git apply -R instead of regex stunts
  • The patch is git diff release/7.1 origin/master -- 'libavdevice/decklink_{enc,dec,common}{.cpp,.h}' — upstream's own SDK 16 migration that renames every IDeckLink* interface to its _v14_2_1 versioned form consistently
  • Verified zampp2 SDI1 recorder: 423 frames in 14 sec @ 29.97 fps, ProRes HQ 91 Mbps clean
  • Why the old patch failed: Fix 1 renamed the allocator class to _v14_2_1, but Fixes 2+3 stubbed out the matching SetVideoInputFrameMemoryAllocator call and replaced videoFrame->GetBytes() with a QueryInterface-around-IDeckLinkVideoBuffer dance — mixing two different API generations. Compiled, didn't crash, silently dropped every video frame
  • How to apply: When SDK or FFmpeg revs, regenerate the patch with git diff <old> <new> -- libavdevice/decklink_* against an FFmpeg tree built with --enable-decklink and drop into services/capture/decklink-sdk16.patch
  • Bisection trick: Build the BMD-provided Linux Capture sample from services/capture/sdk/extracted/Blackmagic DeckLink SDK 16.0/Linux/Samples/Capture to ground-truth whether a frame-zero problem is signal vs FFmpeg
  • DeckLink Duo Mini active ports are 0 and 1 (Duo (1) and Duo (2)); ports 2-3 are inactive on this card

zampp1 vs zampp2 layout (as of 2026-05-22)

  • Repos checked out at /opt/wild-dragon on both
  • zampp2 hostname is zampp2; zampp1 hostname is zampp (mam-api sets NODE_HOSTNAME=zampp1 so cluster_nodes.hostname is zampp1)
  • zampp2 runs ONLY node-agent now (primary stack torn down 2026-05-22)
  • DeckLink Duo Mini device paths are /dev/blackmagic/io0..io3 (NOT dv0..dv3); only port io1 (device index 1) shows any signal during testing
  • mam-api is published on port 47432 on zampp2's .env (historical — zampp2 no longer runs mam-api)
  • Git creds installed at /root/.git-credentials on both zampp1 and zampp2

Pushed commits (already on main)

  • 3b4af6e node-agent: prefer host LAN IP, NODE_IP override
  • 0efef0d cluster route: pickIp() + /devices/blackmagic endpoint
  • a39c983 migration 007: dedupe hostnames + unique index
  • 049beb8 migration 008: expanded codec columns
  • 40a66ba worker compose: network_mode: host for node-agent
  • f4a83ee capture-manager: dynamic ffmpeg args
  • 485af25 capture index.js: bootstrap reads codec env vars
  • 4c65753 recorders route: full codec field whitelist
  • 0ebb3cf onboard-node: auto-detect host LAN IP
  • d39f86d web-ui: bmd-card.js
  • 97628bb chore: remove cloudflare rate-limit probe (.touchtest)
  • 8aa3783 deploy: test-cluster.sh + .touchtest cleanup
  • 4a3a672 cluster: stable hostname for mam-api (NODE_HOSTNAME env), jq-based smoke test
  • 486e3c2 fix(decklink): mount /dev/blackmagic in sidecar + remote node routing via node-agent
  • 30b4def fix(decklink): proper SDK 16 patch via git apply -R
  • 00f3f29 feat(cluster): expose db/redis ports for worker-node connectivity
  • 37767f9 fix(cluster): pickIp() only treats 172.17.x as docker bridge

Git creds on zampp1

/root/.git-credentials holds the Forgejo PAT for user zgaetano. credential.helper=store is set in /root/.gitconfig. Future pushes from zampp1 just need HOME=/root exported (the MeshCentral agent's default $HOME is /usr/local/mesh_services/Mesh Dragon/MeshDragon). Rotate the PAT by overwriting that one file.