* fix (MultiGPU): prevent freeze on manual abort when using MultiGPU CFG Split
Problem:
Upon manual abort application hangs indefinitely.
`InterruptProcessingException` inherits from `BaseException` and bypasses MultiGPU's worker error handling block so thread dies silently, leaving the main thread waiting forever for `result_q.get()`
Fix:
Catch `comfy.model_management.InterruptProcessingException` instead of `Exception` so it's caught and passed back via `result_q` to unblock the main thread when manual abort signal fires.
* oops
?source= or ?module_name= or ?pack_id= (param present but blank) would have returned {} because the helper treated the empty string as an exact-match filter. Coalesce to None at the route boundary so a present-but-blank query param behaves the same as the param being absent. The helper's own behaviour is unchanged and locked in by a new assertion.
Amp-Thread-ID: https://ampcode.com/threads/T-019e86fd-b68f-74de-8c91-d2662377424a
Co-authored-by: Amp <amp@ampcode.com>
Two reviewer-requested improvements to GET /node_startup_errors:
1. Emit the pyproject metadata in the same {project: {...}, tool_comfy: {...}}
shape that comfy_config.config_parser.extract_node_configuration already
returns, instead of inventing a flat {pack_id, display_name, ...} bag.
API consumers can now parse the pyproject block straight through the
shared PyProjectConfig pydantic model. Empty / default-valued leaves
are pruned by a small recursive _prune_empty helper so the payload
stays compact, but nesting and field names match the source-of-truth.
2. Add optional source, module_name, and pack_id query parameters
(combined with AND) so a frontend / Manager can ask ?pack_id=foo
instead of grep'ing through the whole grouped response. pack_id
resolves against pyproject.project.name; entries without a parsed
pyproject are naturally excluded from a pack_id query.
The grouping + filtering + module_path stripping moves into
odes.filter_node_startup_errors so the route handler is a one-liner and
the helper is unit-testable without spinning up an aiohttp app.
Tests: 5 new unit tests covering each filter branch, AND-combination, and
empty-result behaviour, plus an updated pyproject-metadata assertion that
checks the nested PyProjectConfig shape, plus a focused test for the
_prune_empty helper.
* mm: re-instantate smart memory for VRAM
* mm: restore non-dynamic smart memory
By popular demand. We aren't quite ready for the deprecation as non
dynamic enabled GPUs and some high-vram custom model loader setups
prefer the old full hands on.
* memory_management: Add direct to read GPU mode
Make destination optional (or make it optionally GPU) and use aimdo
to file_read direct to GPU.
* ops: Remove stream pin buffers and use aimdo reads
This consumed too much RAM and its better to just take the hit on
the CPU syncing back the stream on a short ring buffer. Aimdo
implements this so just rip the stream pin buffer from comfy.
* model_management: all active pin registration movement
Its better to just let the active model load past the pin limit as
pins and let the pins move around. The saves the HDD and SATA
people disk traffic while only costing a few GPU syncs.
* utils: use aimdo file handle
This opens on windows with more favourable flags
* mp: only count the model proper for loaded_ram and vram
Exclude live loras from the numbers to avoid the case where the reported
loaded memory exceeds the size of the model.
This causes me confusion in the Kijai visualizer when it looked fully
loaded but was hitting disk due to this accounding disrepency.
* utils: add bit reverse utility
useful for max scattering something ordered.
* pinned_memory: Implement offload balancing
Use a max scatter alogorithm to prioritize pins of the same size such
that when doing a little bit of offloading it gets scattered, allowing
the prefetcher to more evenly swollow the offload.
* comfy-aimdo 0.4.7
Aimdo 0.4.7 implement VRAM buffer exhaustion predection to avoid
early speculative load of weights that definately wont fix once the
inference gets further in.
* model-prefetch: consolidate pin ensures on the sync point
This could happen mid prefetch block, cause a sync of the entire
block and lose overlap. Get ahead of the problem with a free down
at the natural compute stream sync point.
* mm: Put a 2GB min on the pin ceiling
This is reasonably bad if it starts causing swap pressure, moreso than
during normal ram-cache proceedings. Clamp it.
* add --fast-disk
* Move dataset/text nodes to text category
* Rename category utils into utilities
* Rename category api node into partner
* Move categories conditioning, latent, sampling, model_patches, training, etc. under model category
* Dispatch partner nodes in to 3d, audio, image, text, video categories
* Move PreviewAny node to utilities category
Jukka Seppänen
Quasar of Mikus
Jedrzej Kosinski
person4268
Alexis Rolland
vidigoat
comfyanonymous
comfyanonymous
Alexander Piskun
Daxiong (Lin)
savvadesogle
Terry Jia
rattus
Luke Mino-Altherr
Charles Chan