Update flask_server.py

This commit is contained in:
MacRimi
2025-10-07 02:23:55 +02:00
parent 8111d96a20
commit db38571646

View File

@@ -16,6 +16,7 @@ import time
import socket import socket
from datetime import datetime, timedelta from datetime import datetime, timedelta
import re # Added for regex matching import re # Added for regex matching
import select # Added for non-blocking read
app = Flask(__name__) app = Flask(__name__)
CORS(app) # Enable CORS for Next.js frontend CORS(app) # Enable CORS for Next.js frontend
@@ -1583,12 +1584,17 @@ def get_detailed_gpu_info(gpu):
try: try:
check_result = subprocess.run(['which', 'intel_gpu_top'], capture_output=True, timeout=1) check_result = subprocess.run(['which', 'intel_gpu_top'], capture_output=True, timeout=1)
if check_result.returncode != 0: if check_result.returncode != 0:
# Tool not found
detailed_info['has_monitoring_tool'] = False detailed_info['has_monitoring_tool'] = False
print(f"[v0] intel_gpu_top not found for Intel GPU") print(f"[v0] intel_gpu_top not found for Intel GPU")
return detailed_info return detailed_info
else:
print(f"[v0] intel_gpu_top found for Intel GPU") import os
if not os.path.exists('/dev/dri/card0'):
detailed_info['has_monitoring_tool'] = False
print(f"[v0] /dev/dri/card0 not accessible - intel_gpu_top cannot run")
return detailed_info
print(f"[v0] intel_gpu_top found and /dev/dri/card0 accessible")
except Exception as e: except Exception as e:
print(f"[v0] Error checking for intel_gpu_top: {e}") print(f"[v0] Error checking for intel_gpu_top: {e}")
detailed_info['has_monitoring_tool'] = False detailed_info['has_monitoring_tool'] = False
@@ -1596,27 +1602,21 @@ def get_detailed_gpu_info(gpu):
data_retrieved = False data_retrieved = False
try: try:
# Start intel_gpu_top process with JSON output to stdout result = subprocess.run(
process = subprocess.Popen( ['intel_gpu_top', '-J', '-s', '500', '-o', '-'],
['intel_gpu_top', '-J', '-s', '1000', '-o', '-'], capture_output=True,
stdout=subprocess.PIPE, text=True,
stderr=subprocess.PIPE, timeout=1.5 # Reduced from 2 seconds
text=True
) )
# Wait for output with increased timeout output = result.stdout.strip()
try:
stdout, stderr = process.communicate(timeout=15)
if stdout: if output and result.returncode == 0:
output = stdout.strip() print(f"[v0] intel_gpu_top output received ({len(output)} chars)")
print(f"[v0] intel_gpu_top output received: {output[:300]}...") print(f"[v0] First 500 chars: {output[:500]}")
# Try to parse as JSON # Try to parse as JSON
try: try:
import json
# The output may contain multiple JSON objects, get the first one
json_str = output
# Find first complete JSON object # Find first complete JSON object
brace_count = 0 brace_count = 0
json_end = -1 json_end = -1
@@ -1631,8 +1631,8 @@ def get_detailed_gpu_info(gpu):
if json_end > 0: if json_end > 0:
json_str = output[:json_end] json_str = output[:json_end]
json_data = json.loads(json_str) json_data = json.loads(json_str)
print(f"[v0] Parsed JSON successfully with keys: {list(json_data.keys())}")
# Parse frequency data # Parse frequency data
if 'frequency' in json_data: if 'frequency' in json_data:
@@ -1681,12 +1681,11 @@ def get_detailed_gpu_info(gpu):
detailed_info[key] = float(busy_value) detailed_info[key] = float(busy_value)
data_retrieved = True data_retrieved = True
print(f"[v0] Intel GPU parsed JSON data successfully: {detailed_info}") print(f"[v0] Intel GPU data retrieved successfully")
except (json.JSONDecodeError, ValueError) as e: except (json.JSONDecodeError, ValueError) as e:
print(f"[v0] JSON parsing failed: {e}, using text parsing") print(f"[v0] JSON parsing failed: {e}")
# Fallback to text parsing
# Fallback to text parsing if JSON fails
# Parse frequency: "0/ 0 MHz" # Parse frequency: "0/ 0 MHz"
freq_match = re.search(r'(\d+)/\s*(\d+)\s*MHz', output) freq_match = re.search(r'(\d+)/\s*(\d+)\s*MHz', output)
if freq_match: if freq_match:
@@ -1726,18 +1725,8 @@ def get_detailed_gpu_info(gpu):
if irq_match: if irq_match:
detailed_info['irq_rate'] = int(irq_match.group(1)) detailed_info['irq_rate'] = int(irq_match.group(1))
data_retrieved = True data_retrieved = True
else:
except subprocess.TimeoutExpired: print(f"[v0] No output received from intel_gpu_top (return code: {result.returncode})")
print(f"[v0] intel_gpu_top process timed out after 15 seconds, terminating...")
process.terminate()
try:
process.wait(timeout=1)
except subprocess.TimeoutExpired:
process.kill()
process.wait()
detailed_info['has_monitoring_tool'] = False
print(f"[v0] intel_gpu_top timed out - marking tool as unavailable")
return detailed_info
if data_retrieved: if data_retrieved:
detailed_info['has_monitoring_tool'] = True detailed_info['has_monitoring_tool'] = True
@@ -1746,8 +1735,13 @@ def get_detailed_gpu_info(gpu):
detailed_info['has_monitoring_tool'] = False detailed_info['has_monitoring_tool'] = False
print(f"[v0] Intel GPU monitoring failed - no data retrieved") print(f"[v0] Intel GPU monitoring failed - no data retrieved")
except subprocess.TimeoutExpired:
print(f"[v0] intel_gpu_top timed out after 1.5 seconds - marking tool as unavailable")
detailed_info['has_monitoring_tool'] = False
except Exception as e: except Exception as e:
print(f"[v0] Error getting Intel GPU details: {e}") print(f"[v0] Error getting Intel GPU details: {e}")
import traceback
traceback.print_exc()
detailed_info['has_monitoring_tool'] = False detailed_info['has_monitoring_tool'] = False
elif vendor == 'NVIDIA': elif vendor == 'NVIDIA':
@@ -2876,4 +2870,7 @@ if __name__ == '__main__':
print("Server will be accessible on all network interfaces (0.0.0.0:8008)") print("Server will be accessible on all network interfaces (0.0.0.0:8008)")
print("API endpoints available at: /api/system, /api/storage, /api/network, /api/vms, /api/logs, /api/health, /api/hardware") print("API endpoints available at: /api/system, /api/storage, /api/network, /api/vms, /api/logs, /api/health, /api/hardware")
app.run(host='0.0.0.0', port=8008, debug=False)
, /api/logs, /api/health, /api/hardware")
app.run(host='0.0.0.0', port=8008, debug=False) app.run(host='0.0.0.0', port=8008, debug=False)