feat(isolation): DynamicVRAM compatibility for process isolation

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
This commit is contained in:
John Pollock
2026-03-04 23:48:02 -06:00
parent a0f8784e9f
commit 9250191c65
38 changed files with 94595 additions and 307 deletions

View File

@@ -789,8 +789,6 @@ class GeminiImage2(IO.ComfyNode):
validate_string(prompt, strip_whitespace=True, min_length=1)
if model == "Nano Banana 2 (Gemini 3.1 Flash Image)":
model = "gemini-3.1-flash-image-preview"
if response_modalities == "IMAGE+TEXT":
raise ValueError("IMAGE+TEXT is not currently available for the Nano Banana 2 model.")
parts: list[GeminiPart] = [GeminiPart(text=prompt)]
if images is not None:
@@ -895,7 +893,7 @@ class GeminiNanoBanana2(IO.ComfyNode):
),
IO.Combo.Input(
"response_modalities",
options=["IMAGE"],
options=["IMAGE", "IMAGE+TEXT"],
advanced=True,
),
IO.Combo.Input(
@@ -925,6 +923,7 @@ class GeminiNanoBanana2(IO.ComfyNode):
],
outputs=[
IO.Image.Output(),
IO.String.Output(),
],
hidden=[
IO.Hidden.auth_token_comfy_org,