Spaces:
Running
Running
File size: 14,750 Bytes
0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 0e1d4ae 0d34ea8 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 |
"""
GPU monitoring service for Video Model Studio.
Tracks NVIDIA GPU resources like utilization, memory, and temperature.
"""
import os
import time
import logging
from typing import Dict, List, Any, Optional, Tuple
from collections import deque
from datetime import datetime
import numpy as np
import pandas as pd
logger = logging.getLogger(__name__)
logger.setLevel(logging.INFO)
# Optional import of pynvml
try:
import pynvml
PYNVML_AVAILABLE = True
except ImportError:
PYNVML_AVAILABLE = False
logger.info("pynvml not available, GPU monitoring will be limited")
class GPUMonitoringService:
"""Service for monitoring NVIDIA GPU resources"""
def __init__(self, history_minutes: int = 10, sample_interval: int = 5):
"""Initialize the GPU monitoring service
Args:
history_minutes: How many minutes of history to keep
sample_interval: How many seconds between samples
"""
self.history_minutes = history_minutes
self.sample_interval = sample_interval
self.max_samples = (history_minutes * 60) // sample_interval
# Track if the monitoring thread is running
self.is_running = False
self.thread = None
# Check if NVIDIA GPUs are available
self.has_nvidia_gpus = False
self.gpu_count = 0
self.device_info = []
self.history = {}
# Try to initialize NVML
self._initialize_nvml()
# Initialize history data structures if GPUs are available
if self.has_nvidia_gpus:
self._initialize_history()
def _initialize_nvml(self):
"""Initialize NVIDIA Management Library"""
if not PYNVML_AVAILABLE:
logger.info("pynvml module not installed, GPU monitoring disabled")
return
try:
pynvml.nvmlInit()
self.gpu_count = pynvml.nvmlDeviceGetCount()
self.has_nvidia_gpus = self.gpu_count > 0
if self.has_nvidia_gpus:
logger.info(f"Successfully initialized NVML, found {self.gpu_count} GPU(s)")
# Get static information about each GPU
for i in range(self.gpu_count):
self.device_info.append(self._get_device_info(i))
else:
logger.info("No NVIDIA GPUs found")
except Exception as e:
logger.warning(f"Failed to initialize NVML: {str(e)}")
self.has_nvidia_gpus = False
def _initialize_history(self):
"""Initialize data structures for storing metric history"""
for i in range(self.gpu_count):
self.history[i] = {
'timestamps': deque(maxlen=self.max_samples),
'utilization': deque(maxlen=self.max_samples),
'memory_used': deque(maxlen=self.max_samples),
'memory_total': deque(maxlen=self.max_samples),
'memory_percent': deque(maxlen=self.max_samples),
'temperature': deque(maxlen=self.max_samples),
'power_usage': deque(maxlen=self.max_samples),
'power_limit': deque(maxlen=self.max_samples),
}
def _get_device_info(self, device_index: int) -> Dict[str, Any]:
"""Get static information about a GPU device
Args:
device_index: Index of the GPU device
Returns:
Dictionary with device information
"""
if not PYNVML_AVAILABLE or not self.has_nvidia_gpus:
return {"error": "NVIDIA GPUs not available"}
try:
handle = pynvml.nvmlDeviceGetHandleByIndex(device_index)
# Get device name (decode if it's bytes)
name = pynvml.nvmlDeviceGetName(handle)
if isinstance(name, bytes):
name = name.decode('utf-8')
# Get device UUID
uuid = pynvml.nvmlDeviceGetUUID(handle)
if isinstance(uuid, bytes):
uuid = uuid.decode('utf-8')
# Get memory info, compute capability
memory_info = pynvml.nvmlDeviceGetMemoryInfo(handle)
compute_capability = pynvml.nvmlDeviceGetCudaComputeCapability(handle)
# Get power limits if available
try:
power_limit = pynvml.nvmlDeviceGetPowerManagementLimit(handle) / 1000.0 # Convert to watts
except pynvml.NVMLError:
power_limit = None
return {
'index': device_index,
'name': name,
'uuid': uuid,
'memory_total': memory_info.total,
'memory_total_gb': memory_info.total / (1024**3), # Convert to GB
'compute_capability': f"{compute_capability[0]}.{compute_capability[1]}",
'power_limit': power_limit
}
except Exception as e:
logger.error(f"Error getting device info for GPU {device_index}: {str(e)}")
return {"error": str(e), "index": device_index}
def collect_gpu_metrics(self) -> List[Dict[str, Any]]:
"""Collect current GPU metrics for all available GPUs
Returns:
List of dictionaries with current metrics for each GPU
"""
if not PYNVML_AVAILABLE or not self.has_nvidia_gpus:
return []
metrics = []
timestamp = datetime.now()
for i in range(self.gpu_count):
try:
handle = pynvml.nvmlDeviceGetHandleByIndex(i)
# Get utilization rates
utilization = pynvml.nvmlDeviceGetUtilizationRates(handle)
# Get memory information
memory_info = pynvml.nvmlDeviceGetMemoryInfo(handle)
# Get temperature
temperature = pynvml.nvmlDeviceGetTemperature(handle, pynvml.NVML_TEMPERATURE_GPU)
# Get power usage if available
try:
power_usage = pynvml.nvmlDeviceGetPowerUsage(handle) / 1000.0 # Convert to watts
except pynvml.NVMLError:
power_usage = None
# Get process information
processes = []
try:
for proc in pynvml.nvmlDeviceGetComputeRunningProcesses(handle):
try:
process_name = pynvml.nvmlSystemGetProcessName(proc.pid)
if isinstance(process_name, bytes):
process_name = process_name.decode('utf-8')
except pynvml.NVMLError:
process_name = f"Unknown (PID: {proc.pid})"
processes.append({
'pid': proc.pid,
'name': process_name,
'memory_used': proc.usedGpuMemory,
'memory_used_mb': proc.usedGpuMemory / (1024**2) # Convert to MB
})
except pynvml.NVMLError:
# Unable to get process information, continue with empty list
pass
gpu_metrics = {
'index': i,
'timestamp': timestamp,
'utilization_gpu': utilization.gpu,
'utilization_memory': utilization.memory,
'memory_total': memory_info.total,
'memory_used': memory_info.used,
'memory_free': memory_info.free,
'memory_percent': (memory_info.used / memory_info.total) * 100,
'temperature': temperature,
'power_usage': power_usage,
'processes': processes
}
metrics.append(gpu_metrics)
except Exception as e:
logger.error(f"Error collecting metrics for GPU {i}: {str(e)}")
metrics.append({
'index': i,
'error': str(e)
})
return metrics
def update_history(self):
"""Update GPU metrics history"""
if not self.has_nvidia_gpus:
return
current_metrics = self.collect_gpu_metrics()
timestamp = datetime.now()
for gpu_metrics in current_metrics:
if 'error' in gpu_metrics:
continue
idx = gpu_metrics['index']
self.history[idx]['timestamps'].append(timestamp)
self.history[idx]['utilization'].append(gpu_metrics['utilization_gpu'])
self.history[idx]['memory_used'].append(gpu_metrics['memory_used'])
self.history[idx]['memory_total'].append(gpu_metrics['memory_total'])
self.history[idx]['memory_percent'].append(gpu_metrics['memory_percent'])
self.history[idx]['temperature'].append(gpu_metrics['temperature'])
if gpu_metrics['power_usage'] is not None:
self.history[idx]['power_usage'].append(gpu_metrics['power_usage'])
else:
self.history[idx]['power_usage'].append(0)
# Store power limit in history (static but kept for consistency)
info = self.device_info[idx]
if 'power_limit' in info and info['power_limit'] is not None:
self.history[idx]['power_limit'].append(info['power_limit'])
else:
self.history[idx]['power_limit'].append(0)
def start_monitoring(self):
"""Start background thread for collecting GPU metrics"""
if self.is_running:
logger.warning("GPU monitoring thread already running")
return
if not self.has_nvidia_gpus:
logger.info("No NVIDIA GPUs found, not starting monitoring thread")
return
import threading
self.is_running = True
def _monitor_loop():
while self.is_running:
try:
self.update_history()
time.sleep(self.sample_interval)
except Exception as e:
logger.error(f"Error in GPU monitoring thread: {str(e)}", exc_info=True)
time.sleep(self.sample_interval)
self.thread = threading.Thread(target=_monitor_loop, daemon=True)
self.thread.start()
logger.info("GPU monitoring thread started")
def stop_monitoring(self):
"""Stop the GPU monitoring thread"""
if not self.is_running:
return
self.is_running = False
if self.thread:
self.thread.join(timeout=1.0)
logger.info("GPU monitoring thread stopped")
def get_gpu_info(self) -> List[Dict[str, Any]]:
"""Get information about all available GPUs
Returns:
List of dictionaries with GPU information
"""
return self.device_info
def get_current_metrics(self) -> List[Dict[str, Any]]:
"""Get current metrics for all GPUs
Returns:
List of dictionaries with current GPU metrics
"""
return self.collect_gpu_metrics()
def get_utilization_data(self, gpu_index: int) -> pd.DataFrame:
"""Get utilization data as a DataFrame
Args:
gpu_index: Index of the GPU to get data for
Returns:
DataFrame with time, utilization, and temperature
"""
if not self.has_nvidia_gpus or gpu_index not in self.history:
return pd.DataFrame()
history = self.history[gpu_index]
if not history['timestamps']:
return pd.DataFrame()
# Convert to dataframe
return pd.DataFrame({
'time': list(history['timestamps']),
'GPU Utilization (%)': list(history['utilization']),
'Temperature (°C)': list(history['temperature'])
})
def get_memory_data(self, gpu_index: int) -> pd.DataFrame:
"""Get memory data as a DataFrame
Args:
gpu_index: Index of the GPU to get data for
Returns:
DataFrame with time, memory percent, and memory used
"""
if not self.has_nvidia_gpus or gpu_index not in self.history:
return pd.DataFrame()
history = self.history[gpu_index]
if not history['timestamps']:
return pd.DataFrame()
# Convert to dataframe
memory_used_gb = [m / (1024**3) for m in history['memory_used']]
return pd.DataFrame({
'time': list(history['timestamps']),
'Memory Usage (%)': list(history['memory_percent']),
'Memory Used (GB)': memory_used_gb
})
def get_power_data(self, gpu_index: int) -> pd.DataFrame:
"""Get power data as a DataFrame
Args:
gpu_index: Index of the GPU to get data for
Returns:
DataFrame with time and power usage
"""
if not self.has_nvidia_gpus or gpu_index not in self.history:
return pd.DataFrame()
history = self.history[gpu_index]
if not history['timestamps'] or not any(history['power_usage']):
return pd.DataFrame()
power_limit = max(history['power_limit']) if any(history['power_limit']) else None
df = pd.DataFrame({
'time': list(history['timestamps']),
'Power Usage (W)': list(history['power_usage'])
})
if power_limit and power_limit > 0:
df['Power Limit (W)'] = power_limit
return df
def shutdown(self):
"""Clean up resources when shutting down"""
self.stop_monitoring()
# Shutdown NVML if it was initialized
if PYNVML_AVAILABLE and self.has_nvidia_gpus:
try:
pynvml.nvmlShutdown()
logger.info("NVML shutdown complete")
except Exception as e:
logger.error(f"Error during NVML shutdown: {str(e)}") |