changed methodology (again) to run their code and then scan their LOCAL dict, not the global dict. This will work and be robust I strongly believe

Still too much new code, but we'll slim it down later
This commit is contained in:
John Pollock
2024-12-30 10:12:28 -06:00
parent caee8716b6
commit 14758b1985
+121 -138
View File
@@ -1,107 +1,42 @@
import os
import ast
import time
import copy
import torch
import comfy.model_management
import os
import importlib.util
import logging
from typing import Dict, Type
##############################################################################
# INITIAL SETUP
##############################################################################
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
logging.info("MultiGPU: Initialization started")
def find_node_definition(module_path: str, node_name: str) -> Dict:
"""
Finds a specific node class definition by searching Python files in the given path.
"""
search_dir = os.path.join("custom_nodes", module_path)
if not os.path.exists(search_dir):
logging.info(f"MultiGPU: No custom_nodes directory {module_path}, skipping")
return None
py_files = []
for root, _, files in os.walk(search_dir):
for file in files:
if file.endswith('.py'):
py_files.append(os.path.join(root, file))
if not py_files:
return None
try:
for file_path in py_files:
with open(file_path, 'r', encoding='utf-8') as f:
tree = ast.parse(f.read())
current_device = comfy.model_management.get_torch_device()
logging.info(f"MultiGPU: Initial device {current_device}")
for node in ast.walk(tree):
if isinstance(node, ast.ClassDef) and node.name == node_name:
class_info = {
'input_types': None,
'return_types': None,
'function': None,
'category': None
}
def get_torch_device_patched():
if (
not torch.cuda.is_available()
or comfy.model_management.cpu_state == comfy.model_management.CPUState.CPU
or "cpu" in str(current_device).lower()
):
return torch.device("cpu")
return torch.device(current_device)
for item in node.body:
if isinstance(item, ast.FunctionDef) and item.name == 'INPUT_TYPES':
if any(d.id == 'classmethod' for d in item.decorator_list if isinstance(d, ast.Name)):
for stmt in item.body:
if isinstance(stmt, ast.Return):
try:
class_info['input_types'] = ast.literal_eval(stmt.value)
except:
pass
elif isinstance(item, ast.Assign):
for target in item.targets:
if isinstance(target, ast.Name):
try:
if target.id == 'RETURN_TYPES':
class_info['return_types'] = ast.literal_eval(item.value)
elif target.id == 'FUNCTION':
class_info['function'] = ast.literal_eval(item.value)
elif target.id == 'CATEGORY':
class_info['category'] = ast.literal_eval(item.value)
except:
pass
return class_info
except Exception as e:
logging.error(f"MultiGPU: Error scanning for {node_name}: {str(e)}")
return None
def create_multigpu_node(node_name: str, class_info: Dict) -> Type:
"""Creates a MultiGPU version of the node"""
class MultiGPUNode:
@classmethod
def INPUT_TYPES(cls):
inputs = copy.deepcopy(class_info['input_types'])
if inputs is None:
inputs = {"required": {}, "optional": {}}
elif "required" not in inputs:
inputs["required"] = {}
devices = ["cpu"] + [f"cuda:{i}" for i in range(torch.cuda.device_count())]
inputs["required"]["device"] = (devices,)
return inputs
RETURN_TYPES = class_info['return_types'] if class_info['return_types'] is not None else tuple()
FUNCTION = "override"
CATEGORY = "multigpu"
def override(self, *args, device="cpu", **kwargs):
global current_device
current_device = device
return args if isinstance(args, tuple) else (args,)
MultiGPUNode.__name__ = f"{node_name}MultiGPU"
return MultiGPUNode
comfy.model_management.get_torch_device = get_torch_device_patched
##############################################################################
# OVERRIDE CLASS
##############################################################################
def override_class(cls):
"""Creates a MultiGPU version of a node class that preserves original functionality."""
class NodeOverride(cls):
@classmethod
def INPUT_TYPES(s):
inputs = copy.deepcopy(cls.INPUT_TYPES())
# In case some node forgot "required"
if "required" not in inputs:
inputs["required"] = {}
devices = ["cpu"] + [f"cuda:{i}" for i in range(torch.cuda.device_count())]
inputs["required"]["device"] = (devices,)
return inputs
@@ -109,69 +44,117 @@ def override_class(cls):
CATEGORY = "multigpu"
FUNCTION = "override"
def override(self, *args, device, **kwargs):
def override(self, *args, device="cpu", **kwargs):
global current_device
current_device = device
fn = getattr(super(), cls.FUNCTION)
return fn(*args, **kwargs)
return NodeOverride
def register_core_nodes(target_nodes: list):
"""Register MultiGPU versions of core ComfyUI nodes"""
##############################################################################
# OUR LOCAL MAPPING OF MULTIGPU OVERRIDE NODES
# (No strict reason it has to be before the function defs, but it's typical.)
##############################################################################
NODE_CLASS_MAPPINGS = {}
##############################################################################
# PART 1: CORE NODES
##############################################################################
def register_core_nodes(core_node_names):
"""
Uses ComfyUI's GLOBAL_NODE_CLASS_MAPPINGS to wrap core nodes.
"""
logging.info("MultiGPU: Starting core node registration")
try:
from nodes import NODE_CLASS_MAPPINGS as GLOBAL_NODE_CLASS_MAPPINGS
logging.info("MultiGPU: Processing core nodes")
for node in target_nodes:
if node in GLOBAL_NODE_CLASS_MAPPINGS:
NODE_CLASS_MAPPINGS[f"{node}MultiGPU"] = override_class(GLOBAL_NODE_CLASS_MAPPINGS[node])
logging.info(f"MultiGPU: Registered {node}MultiGPU")
else:
logging.info(f"MultiGPU: Core node {node} not found")
except Exception as e:
logging.error(f"MultiGPU: Error processing core nodes: {str(e)}")
except ImportError as e:
logging.error(f"MultiGPU: Could not import ComfyUI global node mappings: {e}")
return
for node in core_node_names:
if node in GLOBAL_NODE_CLASS_MAPPINGS:
NODE_CLASS_MAPPINGS[f"{node}MultiGPU"] = override_class(GLOBAL_NODE_CLASS_MAPPINGS[node])
logging.info(f"MultiGPU: Registered core node {node}")
else:
logging.info(f"MultiGPU: Core node '{node}' not found in ComfyUI global mappings")
##############################################################################
# PART 2: CUSTOM NODES (LOCAL DICTIONARY APPROACH, with extra debug logs)
##############################################################################
def register_module(module_path: str, target_nodes: list, local_map_name="NODE_CLASS_MAPPINGS"):
"""
1) Load custom_nodes/<module_path>/__init__.py via importlib
2) Grab the dictionary named `local_map_name` (default: 'NODE_CLASS_MAPPINGS') from that module
3) For each node in target_nodes, if found in that local dictionary, wrap it
4) No fallback to ComfyUI's global mappings
"""
base_dir = os.path.join("custom_nodes", module_path)
init_file = os.path.join(base_dir, "__init__.py")
logging.info(f"MultiGPU: Checking custom node module at: {init_file}")
if not os.path.exists(init_file):
logging.info(f"MultiGPU: Module {module_path} not found or missing __init__.py, skipping.")
return
def register_module(module_path: str, target_nodes: list):
"""Register MultiGPU versions of custom nodes"""
try:
# Handle custom nodes
logging.info(f"MultiGPU: Processing module {module_path}")
search_dir = os.path.join("custom_nodes", module_path)
if not os.path.exists(search_dir):
logging.info(f"MultiGPU: Module directory {module_path} not found, skipping")
return
# List all Python files in the module once
py_files = []
for root, _, files in os.walk(search_dir):
for file in files:
if file.endswith('.py'):
py_files.append(os.path.basename(file))
if py_files:
logging.info(f"MultiGPU: Searching in {module_path}: {', '.join(sorted(py_files))}")
for node in target_nodes:
class_info = find_node_definition(module_path, node)
if class_info:
NODE_CLASS_MAPPINGS[f"{node}MultiGPU"] = create_multigpu_node(node, class_info)
logging.info(f"MultiGPU: Registered {node}MultiGPU")
logging.info(f"MultiGPU: Found {module_path}, loading local dictionary from {init_file}")
spec = importlib.util.spec_from_file_location(module_path, init_file)
module = importlib.util.module_from_spec(spec)
spec.loader.exec_module(module)
logging.info(f"MultiGPU: Executed {module_path} initialization")
except Exception as e:
logging.error(f"MultiGPU: Error in {module_path}: {str(e)}")
logging.error(f"MultiGPU: Error loading {module_path}: {e}")
return
# Initialize
NODE_CLASS_MAPPINGS = {}
current_device = None
# Grab that module's local dictionary (e.g. NODE_CLASS_MAPPINGS)
local_map = getattr(module, local_map_name, None)
if not local_map:
logging.info(f"MultiGPU: {module_path} has no '{local_map_name}' dictionary, skipping override.")
return
# DEBUG: Show everything this node map provides
all_defined_nodes = list(local_map.keys())
logging.info(f"MultiGPU: {module_path} local dict keys: {all_defined_nodes}")
# Wrap each node we want
for node in target_nodes:
if node in local_map:
mgpu_class = override_class(local_map[node])
NODE_CLASS_MAPPINGS[f"{node}MultiGPU"] = mgpu_class
logging.info(f"MultiGPU: Successfully wrapped {node} from {module_path}")
else:
logging.info(f"MultiGPU: Node '{node}' not found in {module_path}'s local dictionary")
##############################################################################
# EXAMPLE USAGE
##############################################################################
# 1) CORE NODES
register_core_nodes([
"UNETLoader",
"VAELoader",
"CLIPLoader",
"DualCLIPLoader",
"TripleCLIPLoader",
"CheckpointLoaderSimple",
"ControlNetLoader",
])
# 2) CUSTOM NODES
register_module("ComfyUI-GGUF", [
"UnetLoaderGGUF",
"UnetLoaderGGUFAdvanced",
"CLIPLoaderGGUF",
"DualCLIPLoaderGGUF",
"TripleCLIPLoaderGGUF",
])
# Register all modules
register_module("", ["UNETLoader", "VAELoader", "CLIPLoader", "DualCLIPLoader","TripleCLIPLoader", "CheckpointLoaderSimple", "ControlNetLoader"])
register_module("ComfyUI-GGUF", ["UnetLoaderGGUF", "UnetLoaderGGUFAdvanced", "CLIPLoaderGGUF","DualCLIPLoaderGGUF", "TripleCLIPLoaderGGUF"])
register_module("x-flux-comfyui", ["LoadFluxControlNet"])
register_module("ComfyUI-Florence2", ["Florence2ModelLoader", "DownloadAndLoadFlorence2Model"])
register_module("ComfyUI-LTXVideo", ["LTXVLoader"])
register_module("ComfyUI-MMAudio", [ "MMAudioFeatureUtilsLoader", "MMAudioModelLoader", "MMAudioSampler"])
register_module("ComfyUI-MMAudio", ["MMAudioFeatureUtilsLoader", "MMAudioModelLoader", "MMAudioSampler"])
register_module("ComfyUI_bitsandbytes_NF4", ["CheckpointLoaderNF4"])
logging.info("MultiGPU: Registration complete")
logging.info(f"MultiGPU: Registered nodes: {', '.join(sorted(NODE_CLASS_MAPPINGS.keys()))}")
logging.info("MultiGPU: Registration complete.")
logging.info("MultiGPU: Final mappings: " + ", ".join(sorted(NODE_CLASS_MAPPINGS.keys())))