Will Miao
260b21766b
ComfyUI runs a full gc.collect() after every prompt, and its cost scales with the number of GC-tracked objects. With a large library the hydrated model/recipe caches account for ~1M long-lived objects, so each post-prompt collection held the GIL for seconds and froze the whole server. After all scanner init tasks finish, run gc.collect() + gc.freeze() once to move the long-lived objects into the permanent generation, which the collector no longer scans. Matches the workaround the reporter verified manually.
42 lines
1.4 KiB
Python
42 lines
1.4 KiB
Python
"""Helpers for keeping ComfyUI's periodic full GC cheap.
|
|
|
|
ComfyUI runs a full ``gc.collect()`` after every prompt, and the cost of that
|
|
collection scales with the number of GC-tracked objects in the process. LoRA
|
|
Manager keeps large long-lived caches in memory (one dict per model plus hash
|
|
indexes, recipe caches, the i18n trees), so on big libraries every post-prompt
|
|
collection walks hundreds of thousands of objects while holding the GIL,
|
|
freezing the whole server for seconds (issue #1147).
|
|
|
|
Freezing those objects once they are hydrated moves them into the permanent
|
|
generation, which the collector no longer scans.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import gc
|
|
import logging
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
def freeze_long_lived_objects(reason: str) -> None:
|
|
"""Move all currently tracked objects into the permanent generation.
|
|
|
|
Runs a full collection first so only reachable objects are frozen.
|
|
Best-effort: failures are logged, never raised.
|
|
"""
|
|
try:
|
|
collected = gc.collect()
|
|
gc.freeze()
|
|
logger.debug(
|
|
"LoRA Manager: froze GC-tracked objects after %s (%d unreachable collected)",
|
|
reason,
|
|
collected,
|
|
)
|
|
except Exception: # pragma: no cover - defensive guard
|
|
logger.warning(
|
|
"LoRA Manager: failed to freeze GC-tracked objects after %s",
|
|
reason,
|
|
exc_info=True,
|
|
)
|