Root cause: render-worker and render-worker-light shared the same GPU,
causing contention. Complex TRB renders went from 17s → 36s (2x slower).
Changes:
- Thumbnails back to asset_pipeline queue (not asset_pipeline_light)
- Dispatch routing always uses asset_pipeline (no queue splitting)
- render-worker-light gated behind "multi-gpu" profile — only starts with:
docker compose --profile multi-gpu up -d
- For single-GPU setups: all rendering is sequential on one worker
The dual queue approach is correct for multi-GPU machines where each
worker gets its own GPU. On single-GPU, serial execution is faster
than concurrent GPU contention.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Instead of duplicating output types for every variation (WebP vs PNG,
different resolution), keep one canonical output type and override
specific fields per order line via render_overrides JSONB.
Backend:
- render_overrides JSONB column on OrderLine (DB migration)
- Render task merges overrides with output type settings (format, width,
height, samples, engine, denoiser, transparent_bg, cycles_device)
- POST /orders/{id}/batch-render-overrides endpoint for bulk override
- PatchLineBody accepts render_overrides for per-line patching
Frontend:
- Batch render overrides section on OrderDetail: output format dropdown
(PNG/JPG/WebP) + resolution dropdown (512-4096)
- Clear button to remove overrides
MCP:
- create_order tool: accepts product_ids, output_type, render_overrides,
material_override — enables "render all products as WebP" via Claude
- set_render_overrides tool: batch override on existing orders
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Task 4: Dual render queue
- render-worker: heavy (asset_pipeline, concurrency=1) — HQ 2048x2048, animations
- render-worker-light: light (asset_pipeline_light, concurrency=2) — thumbnails, <=1024
- Thumbnails routed to light queue automatically
- Order line renders routed by resolution at dispatch time
Task 5: GLB caching (skip re-tessellation)
- Before tessellating, check if gltf_geometry MediaAsset exists for the cad_file_id
- If found, copy to expected path — render_blender.py finds it and skips tessellation
- Saves 7-11s per re-render of the same product
Task 6: WebP output format
- New 'webp' option in output_format (OutputType admin)
- Blender renders PNG intermediate, Pillow converts to WebP (quality=90, method=4)
- 50-70% smaller files with no visible quality loss
- Correct MIME type (image/webp) in MediaAsset
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The OptiX cache was mounted at /root/.nv but NVIDIA writes to
/var/tmp/OptixCache_root/optix7cache.db (28MB). Fixed volume mount.
Before: first render after container restart = 181s (OptiX recompilation)
After: first render after container restart = 20s (cached kernels)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Added refetchInterval to the order query that polls every 3 seconds
while render_progress has pending or processing lines. Stops polling
automatically when all renders reach a terminal state (completed/failed/cancelled).
Fixes: render log and backend status not appearing until manual page reload.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Removed from UI (saved to DB but never read by any service):
- Max Concurrent Renders, Stall Timeout, Thumbnail Format, Product Thumbnail Priority
- Render Linear/Angular Deflection (only Scene deflections are used)
- GLB Scale Factor, Smooth Normals, GLB Material Mode, PBR Roughness, PBR Metallic
Fixed Blender status check:
- Old: called is_blender_available() in backend container (Blender not installed there)
- New: dispatches Celery task on asset_pipeline queue → runs in render-worker container
- Returns: available=true, version="Blender 5.0.1", binary path
- Status card moved to System Tools tab with refresh button
Kept active: engine, device, samples, smooth angle, tessellation, scene deflections,
3D viewer zoom limits
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Admin tables (same pattern as OutputTypeTable):
- RenderTemplateTable: 11 cramped columns → expandable form row with grouped fields,
boolean flags consolidated into compact badges, .blend upload in proper section
- PricingTierTable: inline cell editing → expandable form with labeled fields,
shared renderEditFormGrid() for add/edit modes
- GlobalRenderPositionsPanel: tiny rotation inputs → expandable form with w-24 inputs,
proper labels, sensor_width_mm added to edit form
Page polish:
- WorkerManagement: larger scale controls (p-2 rounded-lg), wider number displays (w-12),
proper labels, more prominent Save button
- Billing: status select gets visible dropdown indicator (ChevronDown icon),
hover border to signal interactivity, larger action buttons with borders
- OrderDetail: batch override in proper card with title/description,
per-line override shows compact "+ override" link (expands on click),
active overrides show as amber badge with X to clear
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replaced the 18-column inline cell editing with an expandable detail row:
- Display row always visible (no more disappearing content on edit)
- Edit form opens BELOW in a full-width colSpan row with accent border
- Fields organized in a 4-column grid with proper labels:
Row 1: Name, Renderer, Format, Animation
Row 2: Turntable, Background, Device, Engine
Row 3: Samples, Resolution, Pricing, Workflow
Row 4: Categories, Material Override, Sort Order, Active
Row 5: Denoising settings (6 cols, Blender only)
- Shared renderEditFormGrid() helper for both edit and add modes
- Save/Cancel buttons at bottom of form
- React.Fragment wrappers for dual-row rendering
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The Preferences page had its own inline accent color section that wasn't
updated with the custom picker. Now shows:
- 5 preset swatches + separator + custom swatch with pipette icon
- Color picker + hex input appear when custom is selected
- Larger swatches (w-8 h-8) matching the page's spacious layout
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Added 'custom' accent option with color picker + hex input
- Pipette icon on the custom swatch (switches to solid when active)
- Color picker appears inline when custom is selected
- Generates hover/light variants automatically from hex (darken/lighten)
- Dark mode accent-light uses rgba for translucency
- Persisted in localStorage, applied before React hydration (no flash)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Widgets scale in with staggered delays (60ms per widget, up to 12)
- Widget hover: lift 3px + accent border glow (dark mode: accent shadow)
- Inner numbers animate up with count-up effect
- Progress bars grow from left with spring curve (800ms delay for content-first feel)
- All wrapped in prefers-reduced-motion guard
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
CSS-layer-only changes for zero-regression visual modernization:
- Typography: Inter font with OpenType features (cv02-cv04, cv11), tighter letter-spacing
- Dark theme: neutral zinc tones (#09090b/#18181b) replacing blue-slate (#0f172a/#1e293b)
- Borders: rgba-based for subtlety instead of solid hex colors
- Cards: rounded-xl, refined shadows with dark mode variant
- Buttons: rounded-lg, subtle lift on hover, smoother transitions (150ms ease-out)
- Badges: rounded-md (softer than full-round), tracking-wide
- Inputs: rounded-lg, ring-offset-0 for tighter focus rings
- Scrollbars: thin 6px custom scrollbars matching theme
- Selection: accent-colored text selection
- Table headers: uppercase tracking-wider for modern data table look
- Utility: fadeIn animation, subtle-pulse for loading states
No component files changed — all styling propagates via CSS variables and Tailwind config.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Added 5 new tools:
- get_product_detail: full product with parts, materials, render history, media assets
- get_render_detail: complete render metadata for an order line (timing, engine, log)
- get_completed_renders: filterable list of completed renders with timing/paths
- get_failed_renders: recent failures with error messages
- get_media_assets: browse media assets by product/type
Total: 17 tools + 2 resources
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Added "Mat Override" column to the review table
- Each line has its own dropdown (per-line takes priority over global)
- Default shows global override if set, otherwise "No override"
- "Clear" option to explicitly remove override on a line when global is set
- Amber background when override is active
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
create_order and add_order_line endpoints were not passing
material_override from the request body to the OrderLine constructor.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Dropdown in Step 3 review to set a single material override for all lines
- Override is passed to each OrderLine.material_override at creation time
- Amber background when override is active
- Library materials loaded on Step 3 entry
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- POST /orders/{id}/batch-material-override endpoint
- Dropdown above the lines table: "Apply to all lines…"
- Options: clear all overrides, or select a library material
- Updates all order lines in one request
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add `material_override` nullable column on OrderLine (DB migration)
- Line override takes priority over OutputType override
- PATCH /orders/{id}/lines/{id} endpoint to update material_override
- Inline dropdown on each order line in the OrderDetail page
- Amber background when override is active
- Same output type, different material per line — no need to create a new output type
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Both camera-orbit and object-rotation modes now parse and apply
the --material-override CLI arg, overriding USD primvar lookups
and material_map before assignment — same pattern as still renders.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The initial implementation only overrode the material_map dict in the task,
but the Blender USD primvar path bypassed it. Now:
- Added --material-override named CLI arg parsed in _blender_args.py
- Both Mode A (factory) and Mode B (template) in _blender_scene_setup.py
override usd_material_lookup and material_map when set
- Passed through full chain: task → step_processor → render_blender → CLI → Blender
- Tested: 175-part bearing rendered with single Steel-Bare material (1/1 materials)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add `material_override` nullable column on OutputType (DB migration)
- When set, ALL product parts get rendered with this single material
- Override applies after alias resolution in render_order_line task
- Admin UI: dropdown in OutputType table to select a library material
- Display: amber badge showing active override material name
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Excel preview detects when a product already has a different STEP file linked
- Excel preview detects intra-Excel conflicts (same product, different CAD model names)
- Product STEP upload warns when replacing an existing file and shows render count
- All warnings are non-blocking (amber badges, toast warnings)
- LEARNINGS.md: all open items resolved
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Per-render-position focal_length_mm/sensor_width_mm (DB → pipeline → Blender)
- FOV-based camera distance with min clamp fix for wide-angle lenses
- Unmapped materials blocking dialog on "Dispatch Renders" with batch alias creation
- Material check endpoint (GET /orders/{id}/check-materials)
- Batch alias endpoint (POST /materials/batch-aliases)
- Quick-map "No alias" badges on Materials page
- Full product hard-delete with storage cleanup (MinIO + disk files + orphaned CadFile)
- Delete button on ProductDetail page with confirmation
- Clickable product names in Media Browser (links to product page)
- Single-line render dispatch/retry (POST /orders/{id}/lines/{id}/dispatch-render)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Root cause 1: BRepBuilderAPI_Transform(shape, trsf, copy=True) destroys
all Poly_Triangulation data. The mm→m scaling was applied before export,
wiping the tessellation from BRepMesh_IncrementalMesh.
Fix: Remove BRepBuilderAPI_Transform entirely — RWGltf_CafWriter already
handles mm→m conversion and Z-up→Y-up rotation internally.
Root cause 2: RWGltf_CafWriter with MergeFaces=False (the default) fails
to find per-face tessellation from the XCAF component hierarchy, producing
degenerate meshes (~2 vertices per face instead of thousands).
Fix: SetMergeFaces(True) to compose face triangulations into proper
per-shape mesh buffers. Vertex count goes from 1,212 to 46,573.
Also bumps cache key version to v2 to invalidate broken cached GLBs.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Adds early-exit checks in dispatch_order_line_render and
render_order_line_task to prevent rendering when order lines are
cancelled or the parent order is rejected/completed.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
New endpoint POST /admin/settings/regenerate-all-canonical-scenes
queues GLB + USD master export for ALL completed CAD files, replacing
existing assets. Used after pipeline changes that affect tessellation
or normals.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- @timed_step decorator with wall-clock + RSS tracking (pipeline_logger)
- Blender timing laps for sharp edges and material assignment
- MeshRegistry pattern: eliminate 13 scene.traverse() calls across viewers
- Lazy material cloning (clone-on-first-write in both viewers)
- _pipeline_session context manager: 7 create_engine() → 2 in render_thumbnail
- KD-tree spatial pre-filter for sharp edge marking (bbox-based pruning)
- Batch material library append: N bpy.ops.wm.append → single bpy.data.libraries.load
- GMSH single-session batching: compound all solids into one tessellation call
- Validate part-materials save endpoints against parsed_objects (prevents bogus keys)
- ROADMAP updated with completion status
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Extract Base Color, Metallic, Roughness, Transmission, IOR from Blender
asset library materials via catalog_assets.py. Store in catalog JSON and
serve via /api/asset-libraries/pbr-map endpoint. Frontend viewers apply
PBR properties to Three.js MeshStandardMaterial using hex color strings
(avoiding Three.js ColorManagement sRGB/linear issues).
Key fixes:
- RLS bypass for material alias lookup in pbr-map endpoint
- pbrMap empty guard prevents premature grey fallback in viewers
- Cache-Control: no-cache on pbr-map requests to avoid stale data
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Rewrite _traverse_xcaf → _author_xcaf_to_usd that recursively authors
USD prims mirroring the XCAF assembly tree:
- Assembly nodes become UsdGeom.Xform prims with local transforms from
each component label's Location (not composed with parents)
- Leaf shapes get definition-space vertices (face_loc only, no instance
placement) — the USD scene graph composes transforms hierarchically
- Coordinate swap (X,-Z,Y) now authored once as a root Xform on
/Root/Assembly instead of per-vertex transformation
- Sharp/seam edges extracted per-part from definition shape (not global)
This fixes misplaced geometry for sub-assembly parts (e.g. KOMP-EIN
roller cages with -45° Z rotation) that were previously lost by the
flat traversal.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Material prims now use resolved canonical names (e.g.
SCHAEFFLER_010101_Steel_Bare) instead of source object names. When
importing USD in Blender, materials show the correct SCHAEFFLER names
directly.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- export_step_to_usd.py: change stage from Y-up to Z-up, keep (X,-Z,Y)
transform — matches GLB orientation exactly (verified: bounding box match)
- export_glb.py: include canonical_material in resolved_material_assignments
DB field (was being dropped during manifest parsing)
- import_usd.py: use pxr customData read (not string primvars — Blender
ignores those, confirmed by testing)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- export_step_to_usd.py: accept --material_map CLI arg, write
schaeffler:canonicalMaterialName as customData on each Mesh prim,
fix geometry transform (strip shape Location before face exploration,
apply both face_loc and shape_loc sequentially)
- import_usd.py: after Blender USD import, use pxr to read customData
directly from the USD file — builds {part_key: material_name} lookup
(Blender ignores STRING primvars and customData, but pxr reads both)
- _blender_materials.py: add apply_material_library_direct() for exact
dict-based material assignment without name-matching heuristics
- _blender_scene_setup.py: prefer direct USD lookup, fall back to
name-matching for legacy USD files without material metadata
- export_glb.py (generate_usd_master_task): resolve material_map via
material_service.resolve_material_map() and pass to subprocess;
include material hash in cache key for invalidation
- ROADMAP.md: update P5 status, add M5-M7 milestones
Tested: 3/3 parts matched (ans_lfs120), 172/175 parts matched
(F-802007.TR4-D1-H122AG). Previous: 0/25 matched.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The queue handles far more than thumbnails: OCC tessellation, USD master
generation, GLB production, order line renders, and workflow renders.
asset_pipeline better reflects its role as the render-worker's primary queue.
Updated all references in: task decorators, celery_app.py, beat_tasks.py,
docker-compose.yml worker command, worker.py MONITORED_QUEUES, admin.py,
CLAUDE.md, LEARNINGS.md, Dockerfile, helpTexts.ts, test files,
and all .claude/commands/*.md skill files.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Previously the endpoint queued USD generation for ALL 295 completed CadFiles,
including 250 orphan CadFiles not linked to any product. Now filters to only
CadFiles referenced by at least one Product.cad_file_id, reducing the backfill
from ~285 to ~41 tasks.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- beat_tasks.py: import app.models at module level so SQLAlchemy can
resolve relationship("Template") and relationship("User") when domain
models are imported in isolation inside task functions. Fixes all
beat tasks (batch_render_notifications, recover_stuck_cad_files) that
crashed every 60s with mapper initialization error.
- _blender_materials.py: build_mat_map_lower() now adds a slug-normalized
key variant (re.sub([^a-z0-9]+, _, kl)) for each mat_map entry. OCC
part names like 'F-802007_TR4-D1-H122AG' → slug 'f_802007_tr4_d1_h122ag'
now matches USD-imported Blender objects. Existing prefix fallback
(key.startswith(part_key)) catches AF-suffix variants.
- Orders.tsx: kanban drag-to-reject implemented. submitted/processing
cards are draggable (cursor-grab). Rejected column highlights with
red ring on drag-over. Drop opens reject reason modal via createPortal.
Confirm calls rejectOrder() mutation + invalidates orders cache.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Dokumentiert drei neue Learnings aus der GE360-HF Wälzkörper-Positions-Untersuchung:
1. BRepMesh auf Compound: Triangulation in Definition-Space, Face-loc = Instance-Placement
2. IsSame() vs IsPartner() für Assembly-Instanz-Deduplizierung
3. Stale GLB-Cache maskiert Code-Fixes — nach Tessellierungs-Änderungen Cache invalidieren
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
In _extract_mesh(), BRep_Tool.Triangulation_s(face, face_loc) returns a face_loc
that already encodes the instance's full placement transform when a compound shape
is tessellated with BRepMesh_IncrementalMesh. Applying shape_trsf on top doubled
every rotation/translation, causing multiple roller elements to collapse to the same
wrong world position (e.g. Z(-75°)×2 ≡ Z(+105°)×2 mod 360° → identical positions).
Fix: use elif so shape_loc is only applied as a fallback when face_loc is identity.
Adds seam edge extraction (UV seam primvar) and improves _traverse_xcaf doc.
docs: learning erfasst - OCC face_loc double-transform in compound tessellation
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add per-file Redis SET NX EX 1800 locks to generate_gltf_geometry_task
and generate_usd_master_task — concurrent duplicates (e.g. double-click
of bulk action buttons) now log a warning and return immediately instead
of running two expensive OCC tessellation subprocesses on the same file
- Fix eng.dispose() called inside with Session() block in cache-hit path
of both tasks — moved to after the with block exits (Tasks 3+4 from plan)
- Add cad.updated_at = datetime.utcnow() in save_manual_material_overrides
(was missing vs parallel save_part_materials endpoint)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- export_gltf.py: replace single-material fallback (only fired when
len(appended)==1) with a universal sentinel that appends
SCHAEFFLER_059999_FailedMaterial unconditionally and assigns it to
every mesh object not matched by name-based lookup.
Also adds in-memory magenta fallback if library append fails.
Removes 2 temporary [DEBUG] print lines from investigation.
- blender_render.py: add FailedMaterial assignment inside
_apply_material_library() for unmatched parts (was log-only before).
Includes copy-on-write guard (users > 1) matching existing pattern.
Also added alias 'Stahl; Durotect CMT' (semicolon) → Durotect-Blue
to cover STEP files using semicolon separator instead of comma.
Verified: 23/25 objects matched correctly, 2 ISO8734 dowel pins
(empty material) receive SCHAEFFLER_059999_FailedMaterial as sentinel.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Bug 1 — Missing parts (mirror/repeated instances):
- id(solid.TShape()) is unreliable in OCP: each call creates a new
Python wrapper, so id() always differs even for the same TShape.
Replaced with IsSame() for correct TShape-pointer deduplication.
- TopExp_Explorer(SOLID) misses free shells/faces in assemblies.
Fix: run BRepMesh baseline on full root compound first (catches all
face types), then GMSH overrides per unique solid for better seam
topology. REVERSED solids keep BRepMesh to avoid inverted Jacobians.
Bug 2 — GLB 7× too large (21 MB vs OCC 3 MB):
- CharacteristicLengthMax = linear_deflection × 50 (was ×15)
matches OCC effective edge length on curved surfaces (~5 mm).
- MinimumCirclePoints = min(12, ...) (was min(20, ...))
- Result: GMSH 91% of OCC file size (target ≤120% ✓)
Verified on rolling bearing STEP: same 4 skipped nodes as OCC,
25 unique GMSH tessellations (IsSame deduplication), no OOM.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Fine production settings (0.03mm/0.05rad) with GMSH → CharacteristicLengthMax
0.45mm → OOM kill on large assemblies even with per-solid iteration.
GMSH quality is algorithmic (conforming seams, no fan triangles) — a denser
mesh provides no extra UV-unwrap benefit. Cap GMSH tessellation at preview
settings (0.1mm/0.1rad) in the production fallback path.
Normal path (GMSH + valid geometry GLB) is unaffected — continues to reuse
_geometry.glb directly without re-tessellating.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Per-solid iteration prevents OOM on multi-part assemblies (25-part bearing:
2.3GB RAM when processing compound → ~100MB per solid with per-solid approach)
- Fix CharacteristicLengthMax multiplier 5× → 15× and cap MinimumCirclePoints
at 20 (prevents 63-pts/circle on angular_deflection=0.1rad → 231MB → 21MB)
- Geometry task timeout 120s → 600s for large assemblies
- Production task: reuse _geometry.glb when GMSH enabled (no re-tessellation)
and cache _production_geom.glb for OCC (mtime vs STEP check)
- Viewer now prefers production GLB when available (shows correct GMSH mesh)
- GMSH OpenMP multithreading (min(cpu_count,16)) for 4.4× speedup
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
GMSH defaults to single-threaded meshing. Setting General.NumThreads,
Mesh.MaxNumThreads1D and Mesh.MaxNumThreads2D to min(cpu_count, 16) enables
parallel Frontal-Delaunay surface meshing across all available cores.
Benchmark on 121-face assembly (32-core host, capped at 16 threads):
Before: 12.7s total (9.8s in gmsh.model.mesh.generate)
After: 2.8s total (1.1s in gmsh.model.mesh.generate)
Cap at 16 threads — benchmark showed 16 threads (1.1s) matches or beats auto
(1.6s), likely due to NUMA/coordination overhead above that threshold.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Introduces GMSH as an alternative to OCC BRepMesh for STEP→GLB tessellation.
GMSH produces conforming meshes that eliminate fan triangles at cylinder seam
edges — a structural limitation of OCC BRepMesh that cannot be fixed via
deflection parameters.
Changes:
- render-worker/Dockerfile: install gmsh>=4.15.0 + libglu1-mesa + libxft2
- export_step_to_gltf.py: --tessellation_engine occ|gmsh CLI arg +
_tessellate_with_gmsh() using BRep→GMSH→Poly_Triangulation write-back
- admin.py: tessellation_engine setting (SETTINGS_DEFAULTS, SettingsOut,
SettingsUpdate, validation)
- export_glb.py: pass tessellation_engine to export_step_to_gltf.py CLI in
both geometry and production GLB tasks
- Admin.tsx: radio button UI for OCC vs GMSH selection
Tested: 121 faces meshed, 0 BRepMesh fallback, 649K triangles on sample part.
Clean seam edges for UV unwrap — GMSH respects B-rep periodic face boundaries.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Merges PLAN.md (Phases A-F), PLAN_REFACTOR.md (Phases 1-8), plan.md (GMSH),
docs/rfcs/0001, and visual-audit-report findings into a single prioritized roadmap.
10 priorities with dependency graph and 'what to do next' decision options.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add export_extras=True to bpy.ops.export_scene.gltf() call
- Store schaeffler_sharp_angle_deg in scene custom props before export
→ value is embedded in scenes[0].extras in the GLB JSON chunk
→ survives import/export round-trip intact (verified: 30.0 restored)
- Add tools/restore_sharp_marks.py: companion Blender script that reads
the angle from scene.get("schaeffler_sharp_angle_deg") and re-applies
mark_sharp() + mark_seam() on all mesh objects after GLB import
GLB format cannot store per-edge sharp/seam flags natively; the visual
shading is correct via vertex splits. The extras + restore script give
users the ability to reconstruct Edit Mode markers without a second format.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When re-generating a production or geometry GLB, the old approach deleted the
existing MediaAsset record and created a new one with a new UUID. Any page that
had the old download_url (/api/media/{old-id}/download) cached would then get
a 404 when trying to download, because the asset ID no longer existed in the DB.
Fix: update the existing MediaAsset record in-place (same UUID, new storage_key)
so existing download URLs remain valid after regeneration. Create a new record
only if no existing one is found.
Applies to both generate_gltf_geometry_task and generate_gltf_production_task.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The geometry GLB from export_step_to_gltf.py contains a 'custom_normal' attribute
(CORNER, INT16_2D) from OCC tessellation. If left in place, Blender's glTF exporter
re-exports these pre-baked normals unchanged — ignoring shade_smooth_by_angle
processing and our explicit sharp edge marks.
Fix: remove the 'custom_normal' attribute from all imported mesh objects immediately
after GLB import, before applying smooth shading. Also add orphans_purge() before
export to remove palette materials (mat_0/1/2/3) that become users=0 after library
material substitution.
Same custom_normal clearing applied to blender_render.py for thumbnail renders.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Sharp edges:
- OCC→Blender coordinate transform was wrong: Blender(X,Y,Z) = OCC(X×0.001, -Z×0.001, Y×0.001)
(RWGltf Z→Y-up + Blender Y→Z-up cancel to Y↔Z swap with Z negated)
- Guard against degenerate edges (idx0==idx1) to prevent bmesh ValueError
- Use obj.matrix_world @ v.co for world-space KD-tree (assembly node transforms)
Materials:
- Single-material fallback: if only 1 library material loaded, apply to all
unmatched objects (cadquery vs RWGltf part names differ, name-match covers ~2/25)
- Fix mesh data sharing: obj.data.copy() before materials.clear() to avoid
clearing shared data blocks
- Use obj.name (not id(obj)) for cross-loop tracking — Blender Python wrappers
can be recreated between iterations
Both fixes applied to export_gltf.py (production GLB) and blender_render.py (thumbnails).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Render pipeline:
- Replace per-object _apply_smooth() loop with _apply_smooth_batch(): selects
all 175 parts, calls shade_smooth_by_angle() ONCE in C → reduces 16s to ~0.2s
- Remove 175 per-part "assigned material to part" log lines (replace with summary)
- Add TIMING_SUMMARY log line at end of every render showing all step durations
- _lap() helper records split times for: template_load, glb_import, rotation,
smooth_shading, material_assign, pre_render_setup, gpu_render
Frontend role checks:
- Add global_admin + tenant_admin to User role type in auth store
- Add isAdmin() and isPrivileged() helper functions
- Fix Admin.tsx, Layout.tsx, Notifications.tsx, OrderDetail.tsx, ProductDetail.tsx,
CostOverviewWidget.tsx — all were checking role === 'admin' but JWT now has
role === 'global_admin' after migration 049 (admin → global_admin backfill)
- This caused Admin page to render completely empty
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Mount named volume optix-cache:/root/.nv so the OptiX ComputeCache
survives docker compose rebuild. Without this every rebuild wiped the
BVH acceleration structure, causing the first render of any complex
scene (~175 parts) to take 130–150s instead of 22s while OptiX
recompiles kernels and rebuilds the BVH from scratch.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Per-tenant Azure AI config stored in tenants.tenant_config JSONB
- GET/PUT /api/tenants/{id}/ai-config + POST .../test connection
- api_key never returned to frontend (has_api_key: bool pattern)
- azure_ai.py resolves creds from tenant config when ai_enabled=True
- ai_tasks.py loads tenant config and passes it to validate_thumbnail
- Admin GPU Status section: probe button + status badge + last-checked time
- Notifications: _BELL_CHANNELS filter (notification+alert only in bell)
- Tenants.tsx: per-row Azure AI Config modal with URL auto-parse helper
- Remove duplicate in-memory /gpu-probe endpoints (kept DB-backed /probe/gpu)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- gpu_probe.py: Blender script that probes OPTIX/CUDA/HIP/ONEAPI and
exits 1 on no GPU — used at startup + on-demand from Admin UI
- blender_render.py, still_render.py, turntable_render.py: emit
RENDER_DEVICE_USED: engine=CYCLES device=GPU|CPU compute_type=...
after GPU activation; exit 2 when CYCLES_DEVICE=gpu and CPU fallback
- render_blender.py: parse RENDER_DEVICE_USED token into render_log
(device_used, compute_type, gpu_fallback); handle exit code 2 as
explicit GPU strict-mode failure
- check_version.py: check_gpu() runs gpu_probe.py at container startup;
CYCLES_DEVICE=gpu aborts startup if no GPU found
- docker-compose.yml: CYCLES_DEVICE=${CYCLES_DEVICE:-auto} env var
- gpu_tasks.py: probe_gpu Celery task on thumbnail_rendering queue;
saves result to system_settings.gpu_probe_last_result; beat every 30min
- worker.py: POST /probe/gpu (trigger) + GET /probe/gpu/result (last result)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
All 19 api/*.ts files already used import api from './client' (X-Tenant-ID
guaranteed). Fixed missing type generics and any usage in 10 files:
- worker.ts: args: any[] → unknown[]
- imports.ts, notifications.ts: add response type generics
- renderTemplates.ts: add typed generics on 6 calls
- materials.ts, cad.ts: type previously untyped api calls
- products.ts: ProductCadUploadResponse interface, typed generics
- uploads.ts: StepUploadResponse interface
- billing.ts, orders.ts: <Blob> on download calls
Zero bare axios usage, zero as any in API layer.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Detect OrderLines stuck in render_status='processing' for >30min,
revoke their Celery task, mark failed, update render_job_doc, and
emit a single CHANNEL_ALERT batch notification to admins.
Runs every 5min via beat schedule alongside recover_stuck_cad_files.
Section B (tenant_id in JWT) confirmed already complete from Phase 4.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
asyncpg+alembic keeps a transaction open via begin_transaction(); calling
execution_options(isolation_level=AUTOCOMMIT) on an already-begun connection
raises InvalidRequestError. Replace with explicit COMMIT before ALTER TYPE
and BEGIN before the backfill UPDATE.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- GET /api/workflows/pipeline-steps: returns all StepName enum values
with category (input|processing|rendering|output) + descriptions;
registered before /{workflow_id} to avoid path collision
- frontend/src/api/workflows.ts: getPipelineSteps(), PipelineStep
and PipelineStepsResponse interfaces
- WorkflowEditor: PipelineStepsPanel showing steps grouped by category
with collapsible accordion sections
- ConfigSidepanel: "Pipeline Step" select dropdown binds any node to a
StepName; selected step description shown below dropdown
- Active workflow indicator: green dot next to is_active=true entries
- Improved empty state with descriptive copy
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Phase 5.1 — MATERIAL_PALETTE removal:
- Remove MATERIAL_PALETTE + _material_to_color() from step_processor.py
- build_part_colors() now returns {part→material_name} for Blender resolver
Phase 6 — Notification Center Refactor:
- Migration 051: add channel (activity|notification|alert) to audit_log,
add frequency (immediate|daily|never) to notification_configs
- Three notification channels: activity (per-render), notification (batch
order summaries), alert (admin infrastructure)
- Per-render emit_notification_sync calls demoted to channel=activity
- New emit_batch_render_notification_sync(): single summary notification
when all order lines reach terminal state ("47/50 succeeded, 3 failed")
- Beat task batch_render_notifications every 60s: safety-net for missed
batch notifications after order completion
- GET /notifications: defaults to channel IN (notification, alert);
accepts ?channel=activity for activity feed
- Unread count badge counts only notification+alert channels
- Notifications.tsx: three tabs (Notifications | Activity | Alerts)
- NotificationSettings.tsx: frequency dropdown per event type
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Phase 2.3 — Fix render cancellation (real Celery task ID):
- orders.py cancel endpoints: read celery_task_id from render_job_doc
instead of synthetic "render-{line_id}" which was a no-op
- render_order_line_still_task: creates RenderJobDocument at task start,
stores self.request.id as celery_task_id, writes step-level state
(RESOLVE_STEP_PATH → BLENDER_STILL) back to order_lines.render_job_doc
Phase 3.1 — Remove Pillow overlay dead code:
- blender_render.py: deleted 55-line Pillow post-processing block
(lines 798-851, green bar + model name label)
- transparent_bg=True is always passed; the else branch was unreachable
- Removed mention from script docstring
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Phase 1 of PLAN_REFACTOR.md — all four sub-tasks implemented:
1.1 PipelineLogger (backend/app/core/pipeline_logger.py)
- Structured step_start/step_done/step_error/step_progress API
- Publishes to Python logging AND Redis SSE via log_task_event
- Context manager `pl.step("name")` for auto-timing
1.2 RenderJobDocument (backend/app/domains/rendering/job_document.py)
- Pydantic JSONB schema: state machine + per-step records + timing
- begin_step/finish_step/fail_step/skip_step helpers
- Migration 048: adds render_job_doc JSONB column to order_lines
- OrderLine model updated with render_job_doc field
1.3 TenantContextMiddleware (backend/app/core/middleware.py)
- Decodes JWT, stores tenant_id + role in request.state
- get_db updated to auto-apply RLS SET LOCAL from request.state
- Registered in main.py (runs before every request)
- JWT now embeds tenant_id claim via create_access_token()
- Login endpoint passes tenant_id to token creation
1.4 ProcessStep Registry (backend/app/core/process_steps.py)
- StepName StrEnum with all 20 pipeline step names
- Single source of truth for log prefixes, DB records, UI labels
Also adds db_utils.py with set_tenant_sync() + get_sync_session()
for use inside Celery tasks (bypass-safe RLS helper).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- GPU: fix Cycles device activation order — set compute_device_type
BEFORE engine init, re-set AFTER open_mainfile wipes preferences
- GPU: remove _mark_sharp_and_seams edit-mode loop (redundant with
Blender 5.0 shade_smooth_by_angle), saves ~200s/render on 175 parts
- Material: fix _AFN suffix mismatch — build AF-stripped mat_map keys
and add prefix fallback in _apply_material_library (blender_render.py)
- Material: production GLB now uses get_material_library_path() which
checks active AssetLibrary instead of empty legacy system setting
- Admin: RenderTemplateTable multi-select output types (M2M frontend)
- Admin: MaterialLibraryPanel replaced with link to Asset Libraries
- UX: move Toaster to top-left to avoid dispatch button overlap
- SQLAlchemy: add .unique() to all RenderTemplate M2M collection queries
- Logging: flush=True on all Blender progress prints, stdout reconfigure
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
OCC's RWGltf_CafWriter appends _AF0/_AF1 assembly-instance suffixes to
mesh object names when a part appears multiple times in an assembly.
The material matching in export_gltf.py only stripped Blender's .001
suffix, leaving 24/175 GLB objects without materials.
Fix: strip _AFN suffixes via while loop (handles nested _AF0_AF1),
add prefix fallback (longest key wins) as last resort before no-match.
Also improve build_materials_from_excel Jaccard matching:
- Strip _AFN and numeric hash suffixes (-21227) before tokenizing
- Add prefix-based fallback (step 3) before position fallback (step 4)
- Raise threshold 0.3 → 0.35 for better precision
- Guard prefix matches to len >= 5 to prevent trivial false positives
Result: Material substitution: 175/175 mesh objects assigned (was 151/175)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
4 root causes fixed:
1. export_colors=False was removed in Blender 4.x — caused every Blender
export to fail (exit 1) and always fall back to trimesh. Remove it.
Blender now runs the full pipeline: materials + sharp edges.
2. GlbModel cloned ref never reset on url change — key={glbBlobUrl} forces
React to remount GlbModel on each new blob URL, resetting the ref so
fresh geometry is always loaded.
3. glbBlobUrl not cleared before re-fetch — setGlbBlobUrl(null) added at
start of downloadUrl effect so spinner shows instead of stale mesh.
4. staleTime: 30_000 delayed picking up new MediaAsset after generation.
Changed to staleTime: 0 so invalidation always triggers immediate refetch.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
InlineCadViewer: STL-derived GLBs have non-indexed geometry (unique vertex
per triangle face). computeVertexNormals() on non-indexed geometry produces
per-face normals (faceted shading). Fix: mergeVertices() first to create
shared/indexed geometry, then computeVertexNormals() averages across
adjacent faces → smooth shading. Indexed Blender GLBs are unaffected.
generate_gltf_geometry_task: asyncio.run() inside a Celery worker that
already runs asyncpg causes 'Future attached to a different loop'. Replace
async _store() with sync SQLAlchemy session (matching the rest of the task).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Linked materials are external references — Blender's GLTF exporter cannot
traverse their node trees to extract Principled BSDF PBR values (metallic,
roughness, base color, normal maps). Appended materials are local copies
that the exporter can fully traverse.
Changes:
- asset_library.py: add link=True parameter (default unchanged for renders)
- export_gltf.py: call apply_asset_library_materials with link=False
- export_gltf.py: add export_materials='EXPORT' + export_image_format='AUTO'
to embed textures and ensure full PBR data in the GLB
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- bpy.ops.import_mesh.stl → bpy.ops.wm.stl_import (removed in Blender 4.0)
- mesh.use_auto_smooth = True → bpy.ops.object.shade_smooth_by_angle()
(use_auto_smooth removed in Blender 4.1)
- Apply smooth shading to all objects unconditionally after scale, so
GLB is smooth-shaded even when no sharp edge midpoints are present
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
