The output-upload-url 500 was NOT the enum cast — it was:
1. INSERT referenced original_project_id; the exports table column is project_id
2. file_type/create_type literals were lowercase ('video'/'render') but the
export_file_type/export_create_type enums are PascalCase ('Video'/'Render')
Verified the corrected INSERT against the live schema. Now real renders produce
a downloadable export instead of completing with export=nil.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Three bugs surfaced bringing up a real After Effects node (verified: AE 2026
claimed + ran, but produced no usable output):
1. aerender got no -comp/-rqindex → "output argument ignored", nothing rendered.
- Claim now returns comp_name from content.projects.render_aep_comp (e.g. "frfinal")
via new Store.GetTemplateCompName; threaded through ClaimedJob → runner.Job →
aerender args (`-comp <name>`, or `-rqindex 1` fallback when unknown).
2. CreateExportForJob INSERT passed render_quality as a bare param into an enum
column → 500 ("output-upload-url HTTP 500"), so completed renders had no export.
- Cast $8::render.render_quality (+ explicit casts for file_type/create_type enums).
3. flatrender-exports bucket didn't exist → uploads would fail anyway.
- render-svc now MakeBucket(exports, templates) idempotently at startup.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Render — "stuck in Queued" fix:
- Jobs were created Queued and only a Windows AE node could claim them, so in the
dev stack (no node) they queued forever.
- New devworker package: in-process mock worker drives Queued jobs through the steps
with progress + live preview frames → Done. Enabled via RENDER_DEV_WORKER (default
true in compose; set false in prod where real nodes claim jobs).
- db: DevClaimNextQueued (atomic oldest-queued → Preparing) + UpdateJobStepProgress
- Verified live: a stuck job advanced Preparing→Done in ~10s with frontend polling.
Studio — predefined template structure:
- Projects are always copied from a template; structure is fixed. Users customise
existing layers, they don't add new ones.
- New studio-config flag ALLOW_ADD_LAYERS (false): StudioToolbar (add text/image/
video/shape) returns null; SceneEditSidebar "add text layer" button hidden.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Concurrent-render ceiling (a user runs 1 render at a time unless granted more):
- Identity: TokenService emits max_renders claim from User.ParallelRenderingCeiling
- Identity: admin POST /v1/users/{id}/render-slots (AdminService.SetRenderSlotsAsync,
clamped 1..50) — gamification or admin raises a user's ceiling
- render-svc: middleware reads max_renders (default 1); CreateJob rejects with 409
active_render_limit when active jobs >= ceiling
- render-svc: db.CountActiveJobs + ListActiveJobs; GET /v1/renders/active returns
in-flight renders + can_start_new
Full-screen render page (replaces the modal):
- /studio/render/[projectId]: config (resolution/fps) → live preview + progress →
download; resumes this project's in-flight render on mount; blocks when another
render is active; reads ?preset=
- StudioTopBar export menu now navigates to the page; RenderModal deleted (dead)
App-wide minimal progress:
- GlobalRenderProgress pill mounted in the locale layout for authed users; polls
/api/render/active every 4s, shows thumbnail + step + % on every page, click →
the render page; hidden on the render page and when idle
Admin: UserActions gains a "concurrent render slots" control.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Root cause of 'stuck on AE': heavy expression-driven projects take >10min for AE
to open, exceeding the scan timeout → job dies → admin UI stuck 'scanning'.
Fix: extend the stdlib .aep RIFX parser to collect every Utf8 name (ParseNames),
since FIX media placeholders are renamed footage ITEMS (frl_c1m1), not layers, and
text are layer names (frl_c1t1) — both are Utf8 chunks. QuickScan now branches on
?mode= (or auto-detects frl_ names) and scaffolds FIX scenes/elements + frd_*color
slots directly from the binary. Verified on the real final.aep that timed out in AE:
1 scene, 6 elements, 4 colors in 0.5s vs 10-min AE timeout.
Admin 'Quick scan (no AE)' is now the recommended path and passes the project mode.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The disk-column append only hit GetNodeByID (whitespace differed); ListNodes
lacked last_disk_pct/disk_total_gb, so the node list 500'd and rendered empty.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The admin render queue called the user-scoped /v1/renders (so it only showed the
admin's own jobs) and parsed items/total instead of data/meta (→ always empty).
- render-svc: GET /v1/admin-renders (admin) → ListAllJobs across users, optional
?status= filter; gateway-wired
- admin renders page now fetches /v1/admin-renders and reads data/meta correctly
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- render-svc: admin-scoped store (ListAllExports / GetExportByIDAny /
SoftDeleteExportAny) + GET/DELETE/download-url under /v1/admin-exports
(admin-gated; separate prefix so it routes to render, not identity's /admin)
- gateway: /v1/admin-exports/* → render
- admin /admin/exports: paginated table of every rendered export with thumbnail,
type/quality, size, duration, dimensions, produce + expiry dates; download
(presigned URL) and delete
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Push a font once → every node installs it → admin sees per-node status.
- render-svc: font_requests + node_fonts tables (mig 25); admin GET/POST/DELETE
/v1/node-fonts (with per-node status matrix); internal (HMAC) GET pending +
POST status for node-agents
- node-agent: fontSyncLoop polls pending fonts every 60s, downloads, installs
(Windows Fonts dir + registry / macOS / linux fc-cache), reports Installed/Failed
- gateway: /v1/node-fonts/* → render
- admin /admin/node-fonts: upload a .ttf/.otf → install on all nodes; per-node
Installed/Pending/Failed badges + counts + delete
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Business rule: each user has a daily render limit. Admin-stop refunds the used
charge (not the user's fault); a user's own cancel does not.
- identity: ConsumeRenderChargeAsync / RefundRenderChargeAsync on DailyRemainRenderCount
with lazy daily reset (mig 24: daily_renders_reset_at). Convention: max=0 ⇒ UNLIMITED,
so existing 0/0 users keep rendering until an admin sets a real limit.
- identity InternalController (service-token): POST /v1/internal/render-charge/{consume,refund}
- render-svc: identityclient + on Create consume (block 429 when limit reached, fail-open
on identity outage); on admin Stop refund the job owner; user /cancel unchanged
- compose: IDENTITY_URL for render-svc, ServiceToken for identity-svc
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The render-queue cancel button used the owner-scoped /cancel (WHERE user_id=…),
so an admin couldn't stop another user's job. Added:
- render-svc: POST /v1/renders/:job_id/stop (admin-gated) → store.StopJob cancels
any in-progress job regardless of owner and frees the assigned node
- admin: render-queue button now "توقف" → /api/admin/renders/{id}/stop (with confirm)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- gateway proxy: trim trailing slash before forwarding upstream. gin's
RedirectTrailingSlash adds /nodes → /nodes/ while render-svc redirects
/nodes/ → /nodes, forming an infinite redirect loop (admin pages 500'd)
- accept is_admin as bool OR string "true" in render/file/notification/gateway
auth middleware (identity emits it as a string) — admin endpoints were 403'ing
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
render-svc:
- db.UpdateJobPreview(): writes base64 PNG to render_jobs.image_preview_b64
(only on active jobs; Done/Failed/Cancelled rows ignored)
- POST /v1/internal/render/jobs/:job_id/preview — node agent endpoint
- Route registered under /v1/internal (nodeAuth)
node-agent:
- runner.PreviewFn callback type alongside ProgressFn
- runner.preview.go: GeneratePreviewB64(percent, quality, resolution)
— pure stdlib (image/png + encoding/base64), no external deps
— 320×180 dark frame with animated progress bar + colored indicator dots
- mock render: pushes a preview frame at every step (5→95%)
- real AE render: pushes a preview frame every 30s
- client.UpdatePreview(): POST /v1/internal/render/jobs/:job_id/preview
- main.go: onPreview callback wires client.UpdatePreview() into runner.Run()
frontend:
- render-jobs.ts: RenderJobRow.preview_b64 field; read from progress endpoint
- status/route.ts: previewB64 included in JSON response
- RenderModal: aspect-ratio preview pane during polling — shows spinner until
first frame arrives, then live-updates with each poll (every 3s);
step label overlaid as badge bottom-right
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add full V2 architecture: identity, content, studio (.NET 10) and file,
render, notification, gateway (Go) services with vendored deps, plus DB
migrations, event/API contracts, and an init-db script.
Wire the Next.js frontend to the gateway: server-side JWT auth routes
(login/register/refresh/logout/me), gateway fetch helper, and session/
cookie/jwt helpers under src/lib.
Containerize the stack via docker-compose.v2.yml and per-service
Dockerfiles. Base images resolve through a Nexus mirror (Docker Hub) and
MCR directly; npm/NuGet pull from Nexus groups. Self-host fonts via
next/font/local to avoid Google Fonts (geo-blocked).
Add CI workflow and ignore .env.v2, *.stackdump, and .NET bin/obj.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
soroush.asadi