Extends pyisolate process isolation with wheel-based dependency
management, sandbox mode policy, and compatibility fixes validated
against DA3 as the first complex isolated custom node.
- Add sandbox_mode policy (required/disabled) with COMFY_HOST_POLICY_PATH
env override for host security configuration
- Plumb cuda_wheels config and standardize child environment detection
- Add PLY, NPZ, File3D, VIDEO serializers for core save nodes
- Register isolated extension web directories on host side
(sandbox_mode=disabled) for frontend JS widget serving
- Capture ACCEPT_ALL_INPUTS from child node class to prevent
@classproperty trigger on proxy class
- Serialize NodeOutput with ui/expand/block_execution through JSON-RPC
- Cherry-pick V3 python_module metadata fix (e4592190)
- Remove improperly committed cache artifacts
DynamicVRAM's on-demand model loading/offloading conflicted with process isolation in three ways: RPC tensor transport stalls from mid-call GPU offload, race conditions between model lifecycle and active RPC operations, and false positive memory leak detection from changed finalizer patterns.
- Marshal CUDA tensors to CPU before RPC transport for dynamic models
- Add operation state tracking + quiescence waits at workflow boundaries
- Distinguish proxy reference release from actual leaks in cleanup_models_gc
- Fix init order: DynamicVRAM must initialize before isolation proxies
- Add RPC timeouts to prevent indefinite hangs on model unavailability
- Prevent proxy-of-proxy chains from DynamicVRAM model reload cycles
- Add torch.device/torch.dtype serializers for new DynamicVRAM RPC paths
- Guard isolation overhead so non-isolated workflows are unaffected
- Migrate env var to PYISOLATE_CHILD
