wangbo/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2026-04-15 21:12:30 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity

feat(entrypoint): dynamic CUDA arch detection, first-run override, fix git clone, clarify Sage Attention flags

Browse Source 
Compute TORCH_CUDA_ARCH_LIST from torch.cuda device properties to build
for the exact GPUs present, improving correctness across mixed setups.

Add first-run dependency install gate with a COMFY_AUTO_INSTALL=1
override to re-run installs on later boots without removing the flag.

Use `python -m pip` consistently with `--no-cache-dir` to avoid stale
wheels and reduce container bloat during rebuilds.

Fix git clone commands to standard HTTPS (no Markdown link syntax) and
use shallow fetch/reset against origin/HEAD for speed and reliability.

Clarify Sage Attention flags: set SAGE_ATTENTION_AVAILABLE only when the
module is importable; require FORCE_SAGE_ATTENTION=1 to enable at boot.

Keep readable GPU logs via `nvidia-smi`, while relying on torch for
compile-time arch targeting. Improve logging throughout the flow.
This commit is contained in:
clsferguson 2025-09-26 12:10:28 -06:00 committed by GitHub
parent f2f351d235
commit 13f3f11431
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 110 additions and 118 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

228
entrypoint.sh
View File

@ -10,6 +10,7 @@ CUSTOM_NODES_DIR="$BASE_DIR/custom_nodes"
SAGE_ATTENTION_DIR="$BASE_DIR/.sage_attention"
SAGE_ATTENTION_BUILT_FLAG="$SAGE_ATTENTION_DIR/.built"
PERMISSIONS_SET_FLAG="$BASE_DIR/.permissions_set"
FIRST_RUN_FLAG="$BASE_DIR/.first_run_complete"
# Function to log with timestamp
log() {
@ -24,61 +25,49 @@ import sys
if not torch.cuda.is_available():
print('[ERROR] PyTorch CUDA not available')
sys.exit(1)
device_count = torch.cuda.device_count()
print(f'[TEST] PyTorch CUDA available with {device_count} devices')
for i in range(device_count):
props = torch.cuda.get_device_properties(i)
print(f'[TEST] GPU {i}: {props.name} (Compute {props.major}.{props.minor})')
" 2>/dev/null
}
# Function to detect all GPUs and their generations
# Function to detect all GPUs and their generations (best-effort labels)
detect_gpu_generations() {
local gpu_info=$(nvidia-smi --query-gpu=name --format=csv,noheader,nounits 2>/dev/null || echo "")
local gpu_info
gpu_info=$(nvidia-smi --query-gpu=name --format=csv,noheader,nounits 2>/dev/null || echo "")
local has_rtx20=false
local has_rtx30=false
local has_rtx40=false
local has_rtx50=false
local gpu_count=0
if [ -z "$gpu_info" ]; then
log "No NVIDIA GPUs detected"
return 1
fi
log "Detecting GPU generations:"
while IFS= read -r gpu; do
gpu_count=$((gpu_count + 1))
log " GPU $gpu_count: $gpu"
case "$gpu" in
*"RTX 20"*|*"2060"*|*"2070"*|*"2080"*|*"2090"*)
has_rtx20=true
;;
*"RTX 30"*|*"3060"*|*"3070"*|*"3080"*|*"3090"*)
has_rtx30=true
;;
*"RTX 40"*|*"4060"*|*"4070"*|*"4080"*|*"4090"*)
has_rtx40=true
;;
*"RTX 50"*|*"5060"*|*"5070"*|*"5080"*|*"5090"*)
has_rtx50=true
;;
*"RTX 20"*|*"2060"*|*"2070"*|*"2080"*|*"2090"*) has_rtx20=true ;;
*"RTX 30"*|*"3060"*|*"3070"*|*"3080"*|*"3090"*) has_rtx30=true ;;
*"RTX 40"*|*"4060"*|*"4070"*|*"4080"*|*"4090"*) has_rtx40=true ;;
*"RTX 50"*|*"5060"*|*"5070"*|*"5080"*|*"5090"*) has_rtx50=true ;;
esac
done <<< "$gpu_info"
# Store detection results globally
export DETECTED_RTX20=$has_rtx20
export DETECTED_RTX30=$has_rtx30
export DETECTED_RTX40=$has_rtx40
export DETECTED_RTX50=$has_rtx50
export GPU_COUNT=$gpu_count
log "Detection summary: RTX20=$has_rtx20, RTX30=$has_rtx30, RTX40=$has_rtx40, RTX50=$has_rtx50"
# Test PyTorch CUDA compatibility
if test_pytorch_cuda; then
log "PyTorch CUDA compatibility confirmed"
else
@ -89,27 +78,26 @@ detect_gpu_generations() {
# Function to determine optimal Sage Attention strategy for mixed GPUs
determine_sage_strategy() {
local strategy=""
# Mixed generation logic - prioritize compatibility over peak performance
if [ "$DETECTED_RTX20" = "true" ]; then
if [ "$DETECTED_RTX30" = "true" ] || [ "$DETECTED_RTX40" = "true" ] || [ "$DETECTED_RTX50" = "true" ]; then
if [ "${DETECTED_RTX20:-false}" = "true" ]; then
if [ "${DETECTED_RTX30:-false}" = "true" ] || [ "${DETECTED_RTX40:-false}" = "true" ] || [ "${DETECTED_RTX50:-false}" = "true" ]; then
strategy="mixed_with_rtx20"
log "Mixed GPU setup detected with RTX 20 series - using compatibility mode"
else
strategy="rtx20_only"
log "RTX 20 series only detected"
fi
elif [ "$DETECTED_RTX50" = "true" ]; then
elif [ "${DETECTED_RTX50:-false}" = "true" ]; then
strategy="rtx50_capable"
log "RTX 50 series detected - using latest optimizations"
elif [ "$DETECTED_RTX40" = "true" ] || [ "$DETECTED_RTX30" = "true" ]; then
elif [ "${DETECTED_RTX40:-false}" = "true" ] || [ "${DETECTED_RTX30:-false}" = "true" ]; then
strategy="rtx30_40_optimized"
log "RTX 30/40 series detected - using standard optimizations"
else
strategy="fallback"
log "Unknown or unsupported GPU configuration - using fallback"
fi
export SAGE_STRATEGY=$strategy
}
@ -118,22 +106,22 @@ install_triton_version() {
case "$SAGE_STRATEGY" in
"mixed_with_rtx20"|"rtx20_only")
log "Installing Triton 3.2.0 for RTX 20 series compatibility"
python -m pip install --user --force-reinstall "triton==3.2.0" || {
python -m pip install --no-cache-dir --user --force-reinstall "triton==3.2.0" || {
log "WARNING: Failed to install specific Triton version, using default"
python -m pip install --user --force-reinstall triton || true
python -m pip install --no-cache-dir --user --force-reinstall triton || true
}
;;
"rtx50_capable")
log "Installing latest Triton for RTX 50 series"
python -m pip install --user --force-reinstall triton || \
python -m pip install --user --force-reinstall --pre triton || {
log "WARNING: Failed to install latest Triton, using stable"
python -m pip install --user --force-reinstall "triton>=3.2.0" || true
python -m pip install --no-cache-dir --user --force-reinstall triton || \
python -m pip install --no-cache-dir --user --force-reinstall --pre triton || {
log "WARNING: Failed to install latest Triton, using stable >=3.2.0"
python -m pip install --no-cache-dir --user --force-reinstall "triton>=3.2.0" || true
}
;;
*)
log "Installing latest stable Triton"
python -m pip install --user --force-reinstall triton || {
python -m pip install --no-cache-dir --user --force-reinstall triton || {
log "WARNING: Triton installation failed, continuing without"
return 1
}
@ -141,29 +129,35 @@ install_triton_version() {
esac
}
# Function to compute CUDA arch list from torch
compute_cuda_arch_list() {
python - <<'PY' 2>/dev/null
import torch
archs = set()
if torch.cuda.is_available():
for i in range(torch.cuda.device_count()):
p = torch.cuda.get_device_properties(i)
archs.add(f"{p.major}.{p.minor}")
print(";".join(sorted(archs)))
PY
}
# Function to build Sage Attention with architecture-specific optimizations
build_sage_attention_mixed() {
log "Building Sage Attention for mixed GPU environment..."
# Create sage attention directory
log "Building Sage Attention for current GPU environment..."
mkdir -p "$SAGE_ATTENTION_DIR"
cd "$SAGE_ATTENTION_DIR"
# Set CUDA architecture list based on detected GPUs
local cuda_arch_list=""
[ "$DETECTED_RTX20" = "true" ] && cuda_arch_list="${cuda_arch_list}7.5;"
[ "$DETECTED_RTX30" = "true" ] && cuda_arch_list="${cuda_arch_list}8.6;"
[ "$DETECTED_RTX40" = "true" ] && cuda_arch_list="${cuda_arch_list}8.9;"
[ "$DETECTED_RTX50" = "true" ] && cuda_arch_list="${cuda_arch_list}12.0;"
# Remove trailing semicolon
cuda_arch_list=${cuda_arch_list%;}
# Export for PyTorch build
export TORCH_CUDA_ARCH_LIST="$cuda_arch_list"
log "Set TORCH_CUDA_ARCH_LIST=$TORCH_CUDA_ARCH_LIST"
# Clone or update repository based on strategy
local cuda_arch_list
cuda_arch_list="$(compute_cuda_arch_list || true)"
if [ -n "${cuda_arch_list:-}" ]; then
export TORCH_CUDA_ARCH_LIST="$cuda_arch_list"
log "Set TORCH_CUDA_ARCH_LIST=$TORCH_CUDA_ARCH_LIST"
else
log "Could not infer TORCH_CUDA_ARCH_LIST from torch; proceeding with PyTorch defaults"
fi
case "$SAGE_STRATEGY" in
"mixed_with_rtx20"|"rtx20_only")
log "Cloning Sage Attention v1.0 for RTX 20 series compatibility"
@ -183,7 +177,7 @@ build_sage_attention_mixed() {
if [ -d "SageAttention/.git" ]; then
cd SageAttention
git fetch --depth 1 origin || return 1
git reset --hard origin/main || return 1
git reset --hard origin/HEAD || return 1
else
rm -rf SageAttention
git clone --depth 1 https://github.com/thu-ml/SageAttention.git || return 1
@ -191,11 +185,9 @@ build_sage_attention_mixed() {
fi
;;
esac
# Build with architecture-specific flags using --user installation
log "Building Sage Attention with multi-GPU support..."
if MAX_JOBS=$(nproc) python -m pip install --user --no-build-isolation .; then
# Create strategy-specific built flag
log "Building Sage Attention..."
if MAX_JOBS=$(nproc) python -m pip install --no-cache-dir --user --no-build-isolation .; then
echo "$SAGE_STRATEGY" > "$SAGE_ATTENTION_BUILT_FLAG"
log "Sage Attention built successfully for strategy: $SAGE_STRATEGY"
cd "$BASE_DIR"
@ -210,17 +202,15 @@ build_sage_attention_mixed() {
# Function to check if current build matches detected GPUs
needs_rebuild() {
if [ ! -f "$SAGE_ATTENTION_BUILT_FLAG" ]; then
return 0 # Needs build
return 0
fi
local built_strategy
built_strategy=$(cat "$SAGE_ATTENTION_BUILT_FLAG" 2>/dev/null || echo "unknown")
if [ "$built_strategy" != "$SAGE_STRATEGY" ]; then
log "GPU configuration changed (was: $built_strategy, now: $SAGE_STRATEGY) - rebuild needed"
return 0 # Needs rebuild
return 0
fi
return 1 # No rebuild needed
return 1
}
# Function to check if Sage Attention is working
@ -230,7 +220,6 @@ import sys
try:
import sageattention
print('[TEST] Sage Attention import: SUCCESS')
# Try to get version info
try:
if hasattr(sageattention, '__version__'):
print(f'[TEST] Version: {sageattention.__version__}')
@ -248,42 +237,43 @@ except Exception as e:
# Main GPU detection and Sage Attention setup
setup_sage_attention() {
# DO NOT set SAGE_ATTENTION_AVAILABLE here; respect any user-provided env choice
# Track build status separately for logging/visibility
# Internal tracking and exported availability flag
export SAGE_ATTENTION_BUILT=0
export SAGE_ATTENTION_AVAILABLE=0
# Detect GPU generations
if ! detect_gpu_generations; then
log "No GPUs detected, skipping Sage Attention setup"
return 0
fi
# Determine optimal strategy
determine_sage_strategy
# Build/install if needed
if needs_rebuild || ! test_sage_attention; then
log "Building Sage Attention..."
if install_triton_version && build_sage_attention_mixed && test_sage_attention; then
export SAGE_ATTENTION_BUILT=1
log "Sage Attention is built and available; enable by setting SAGE_ATTENTION_AVAILABLE=1 or using --use-sage-attention explicitly"
export SAGE_ATTENTION_AVAILABLE=1
log "Sage Attention is built and importable; enable at boot by setting FORCE_SAGE_ATTENTION=1"
else
export SAGE_ATTENTION_BUILT=0
export SAGE_ATTENTION_AVAILABLE=0
log "WARNING: Sage Attention is not available after build attempt"
fi
else
export SAGE_ATTENTION_BUILT=1
export SAGE_ATTENTION_AVAILABLE=1
log "Sage Attention already built and importable for current GPU configuration"
fi
}
# If running as root, handle permissions and user mapping
if [ "$(id -u)" = "0" ]; then
# Check if permissions are already set
if [ ! -f "$PERMISSIONS_SET_FLAG" ]; then
log "Setting up user permissions..."
# Map group to PGID if it already exists, otherwise remap the named group
if getent group "${PGID}" >/dev/null; then
EXISTING_GRP="$(getent group "${PGID}" | cut -d: -f1)"
usermod -g "${EXISTING_GRP}" "${APP_USER}" || true
@ -291,17 +281,14 @@ if [ "$(id -u)" = "0" ]; then
else
groupmod -o -g "${PGID}" "${APP_GROUP}" || true
fi
# Map user to PUID
usermod -o -u "${PUID}" "${APP_USER}" || true
# Ensure home and app dir exist and are owned
mkdir -p "/home/${APP_USER}"
for d in "$BASE_DIR" "/home/$APP_USER"; do
[ -e "$d" ] && chown -R "${APP_USER}:${APP_GROUP}" "$d" || true
done
# Make Python system install targets writable for the runtime user (only under /usr/local)
readarray -t PY_PATHS < <(python - <<'PY'
import sys, sysconfig, os, datetime
def log(msg):
@ -325,8 +312,7 @@ if d:
log("Finished emitting target directories")
PY
)
# Make directories writable
for d in "${PY_PATHS[@]}"; do
case "$d" in
/usr/local|/usr/local/*)
@ -337,16 +323,14 @@ PY
*) : ;;
esac
done
# Create permissions set flag
touch "$PERMISSIONS_SET_FLAG"
chown "${APP_USER}:${APP_GROUP}" "$PERMISSIONS_SET_FLAG"
log "User permissions configured"
else
log "User permissions already configured, skipping..."
fi
# Re-exec as the runtime user
exec runuser -u "${APP_USER}" -- "$0" "$@"
fi
@ -366,40 +350,52 @@ fi
# User-site PATHs for --user installs (custom nodes)
export PATH="$HOME/.local/bin:$PATH"
pyver="$(python -c 'import sys; print(f"{sys.version_info.major}.{sys.version_info.minor}")')"
pyver="$(python -c 'import sys; print(f\"{sys.version_info.major}.{sys.version_info.minor}\")')"
export PYTHONPATH="$HOME/.local/lib/python${pyver}/site-packages:${PYTHONPATH:-}"
# Auto-install custom node deps
if [ "${COMFY_AUTO_INSTALL:-1}" = "1" ]; then
log "Scanning custom nodes for requirements..."
# Install any requirements*.txt found under custom_nodes (upgrade within constraints)
while IFS= read -r -d '' req; do
log "pip install --user --upgrade -r $req"
pip install --no-cache-dir --user --upgrade --upgrade-strategy only-if-needed -r "$req" || true
done < <(find "$CUSTOM_NODES_DIR" -maxdepth 3 -type f \( -iname 'requirements.txt' -o -iname 'requirements-*.txt' -o -path '*/requirements/*.txt' \) -print0)
# For pyproject.toml-based nodes, EXCLUDE ComfyUI-Manager (it's not meant to be wheel-built)
while IFS= read -r -d '' pjt; do
d="$(dirname "$pjt")"
log "pip install --user . in $d"
(cd "$d" && pip install --no-cache-dir --user .) || true
done < <(find "$CUSTOM_NODES_DIR" -maxdepth 2 -type f -iname 'pyproject.toml' -not -path '*/ComfyUI-Manager/*' -print0)
pip check || true
# First-run detection for custom node deps (with override)
RUN_NODE_INSTALL=0
if [ ! -f "$FIRST_RUN_FLAG" ]; then
RUN_NODE_INSTALL=1
log "First run detected: installing custom node dependencies"
elif [ "${COMFY_AUTO_INSTALL:-0}" = "1" ]; then
RUN_NODE_INSTALL=1
log "COMFY_AUTO_INSTALL=1: forcing custom node dependency install"
else
log "Not first run and COMFY_AUTO_INSTALL!=1: skipping custom node dependency install"
fi
# Build ComfyUI command with Sage Attention flag only if user explicitly enabled it via env
if [ "$RUN_NODE_INSTALL" = "1" ]; then
log "Scanning custom nodes for requirements..."
while IFS= read -r -d '' req; do
log "python -m pip install --user --upgrade -r $req"
python -m pip install --no-cache-dir --user --upgrade --upgrade-strategy only-if-needed -r "$req" || true
done < <(find "$CUSTOM_NODES_DIR" -maxdepth 3 -type f \( -iname 'requirements.txt' -o -iname 'requirements-*.txt' -o -path '*/requirements/*.txt' \) -print0)
while IFS= read -r -d '' pjt; do
d="$(dirname "$pjt")"
log "python -m pip install --user . in $d"
(cd "$d" && python -m pip install --no-cache-dir --user .) || true
done < <(find "$CUSTOM_NODES_DIR" -maxdepth 2 -type f -iname 'pyproject.toml' -not -path '*/ComfyUI-Manager/*' -print0)
python -m pip check || true
# Mark first run complete (or keep flag if already set)
touch "$FIRST_RUN_FLAG" || true
fi
# Build ComfyUI command with Sage Attention flag only if forced
COMFYUI_ARGS=""
if [ "${SAGE_ATTENTION_AVAILABLE:-0}" = "1" ]; then
if [ "${FORCE_SAGE_ATTENTION:-0}" = "1" ]; then
if test_sage_attention; then
COMFYUI_ARGS="--use-sage-attention"
log "Starting ComfyUI with Sage Attention enabled by environment (SAGE_ATTENTION_AVAILABLE=1)"
log "Starting ComfyUI with Sage Attention forced by environment (FORCE_SAGE_ATTENTION=1)"
else
log "WARNING: SAGE_ATTENTION_AVAILABLE=1 but Sage Attention import failed; starting without"
log "WARNING: FORCE_SAGE_ATTENTION=1 but Sage Attention import failed; starting without"
fi
else
if [ "${SAGE_ATTENTION_BUILT:-0}" = "1" ]; then
log "Sage Attention is built and available; set SAGE_ATTENTION_AVAILABLE=1 to enable it"
if [ "${SAGE_ATTENTION_AVAILABLE:-0}" = "1" ]; then
log "Sage Attention is built and available; set FORCE_SAGE_ATTENTION=1 to enable it on boot"
else
log "Sage Attention not available; starting without it"
fi
@ -409,16 +405,12 @@ cd "$BASE_DIR"
# Handle both direct execution and passed arguments
if [ $# -eq 0 ]; then
# No arguments passed, use default
exec python main.py --listen 0.0.0.0 $COMFYUI_ARGS
else
# Arguments were passed, check if it's the default command
if [ "$1" = "python" ] && [ "${2:-}" = "main.py" ]; then
# Default python command, add our args
shift 2 # Remove 'python main.py'
if [ "${1:-}" = "python" ] && [ "${2:-}" = "main.py" ]; then
shift 2
exec python main.py $COMFYUI_ARGS "$@"
else
# Custom command, pass through as-is
exec "$@"
fi
fi