openapi.yaml has been linted in CI but never reachable over HTTP. Add
--enable-api-docs (off by default) to serve it at /openapi.yaml and render
it at /api-docs.
The viewer is Redoc rather than Swagger UI specifically because it has no
request-execution feature. The local server is unauthenticated by default,
so a docs page with "Try it out" would give one-click access to
/api/interrupt, /api/free and DELETE /api/userdata/{file}.
Notes on the implementation:
- Routes are registered on PromptServer's route table, not on the app, so
they land before the web.static('/') catch-all that would shadow them.
This also means they are served at both /openapi.yaml and
/api/openapi.yaml; the docs page references the spec by a relative URL so
it resolves from either mount point.
- The spec path resolves from __file__, since ComfyUI is routinely launched
from other directories.
- Cache headers are set explicitly: the cache_control middleware only
special-cases js/css/images, so a .yaml response would otherwise be
served stale after an edit.
- /docs is left alone; it already serves embedded node help content.
The Redoc bundle comes from a pinned CDN URL, so the page needs outbound
network access. Offline installs still get the spec itself, and the page
degrades to a notice pointing at it.
Claude-Session: https://claude.ai/code/session_01BvUveU9ofyGrSz3QxYeecB
Add a ModelAttentionBackend node to manually select the attention for models in the workflows. Currently supports pytorch attention or comfy kitchen attention.
Add --use-ck-attention to enable comfy kitchen attention as the default attention backend for all models (might break some).
This avoids name collision (circular imports) for external custom nodes,
for which the comfy path is pushed into sys.path so Python's own logging module
is shadowed otherwise.
fixes: #15229
This priority scheme was broken in the case where you have pin
registration exhaustion while loading a VBAR that gets a big evicition.
The weight would stay in the loaded set but inherit the MRU priority
against other workflow models WRT pin registration which leads to async
offload without pinning.
Fix by universally promiting active pin registration above workflow
pins without concern for the weights/weights-loaded split. This diverges
from the actual budgeting where the split still makes sense.
Changes:
Remove sequential scan hint
Prefer NVML pressure on windows
Add async malloc clamp option (unused by comfy so far)
Workaround AMD windows GPU virtual address space leak
The largest change is the NVML pressure, which works around a cuMemGetInfo
drift from actual VRAM in some circumstances.
Windows has proven this logic works for a long time and there are
corner cases where this materialization actual consumes real RAM
on linux.
Its not as bad as the original windows commit charge surge, but
its still a detectable transient leak. So simplify and unify.
Claude
Jukka Seppänen
Barish Ozbay
comfyanonymous
rattus
Alexander Brown
Alex Harper
Alexis Rolland
blepping
endman100
rivadart
Oliver Freyermuth