Files
core-api/src/domains/tools/system/service.py
T
Jeroen SchweitzerandClaude Opus 4.5 075b0ec297
Build and Push / build (release) Successful in 1m44s
feat: add system stats API for dashboard monitoring
- GET /tools/system/stats - Real-time host system statistics
- CPU usage, memory, all mounted disks, network I/O
- GPU/VRAM stats via nvidia-smi (if available)
- Uses psutil for cross-platform host metrics
- Auto-discovers and filters real filesystems

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-03 17:30:45 +01:00

191 lines
6.0 KiB
Python

"""
System stats service for collecting host system metrics
"""
import subprocess
import socket
from datetime import datetime, timezone
import psutil
from src.shared.logging import get_logger
from typing import List
from src.domains.tools.system.schemas import (
SystemStatsResponse,
CpuStats,
MemoryStats,
DiskStats,
NetworkStats,
GpuStats,
)
# Filesystem types to exclude (virtual/system filesystems)
EXCLUDED_FSTYPES = {
"tmpfs", "devtmpfs", "devfs", "squashfs", "overlay",
"aufs", "proc", "sysfs", "cgroup", "cgroup2",
"debugfs", "tracefs", "securityfs", "pstore",
"hugetlbfs", "mqueue", "binfmt_misc", "autofs",
"fuse.lxcfs", "nsfs", "efivarfs",
}
logger = get_logger(__name__)
class SystemStatsService:
"""Service for collecting host system statistics"""
async def get_stats(self) -> SystemStatsResponse:
"""
Collect current system statistics.
Returns:
SystemStatsResponse with CPU, memory, disks, network, and GPU stats
"""
cpu = self._get_cpu_stats()
memory = self._get_memory_stats()
disks = self._get_all_disk_stats()
network = self._get_network_stats()
gpu = self._get_gpu_stats()
return SystemStatsResponse(
cpu=cpu,
memory=memory,
disks=disks,
network=network,
gpu=gpu,
hostname=socket.gethostname(),
queried_at=datetime.now(timezone.utc),
)
def _get_cpu_stats(self) -> CpuStats:
"""Get CPU usage statistics"""
# Get CPU percentage (blocking call with interval for accuracy)
cpu_percent = psutil.cpu_percent(interval=0.1)
cpu_count = psutil.cpu_count()
# Get load averages (Unix only)
try:
load_1, load_5, load_15 = psutil.getloadavg()
except (AttributeError, OSError):
load_1 = load_5 = load_15 = None
return CpuStats(
usage_percent=cpu_percent,
cores=cpu_count or 1,
load_1m=load_1,
load_5m=load_5,
load_15m=load_15,
)
def _get_memory_stats(self) -> MemoryStats:
"""Get memory usage statistics"""
mem = psutil.virtual_memory()
return MemoryStats(
usage_percent=mem.percent,
total_bytes=mem.total,
used_bytes=mem.used,
available_bytes=mem.available,
)
def _get_all_disk_stats(self) -> List[DiskStats]:
"""Get disk usage statistics for all mounted real filesystems"""
disks = []
seen_devices = set()
for partition in psutil.disk_partitions(all=False):
# Skip excluded filesystem types
if partition.fstype.lower() in EXCLUDED_FSTYPES:
continue
# Skip duplicate devices (same device mounted multiple times)
if partition.device in seen_devices:
continue
seen_devices.add(partition.device)
# Skip Docker/container overlays
if partition.mountpoint.startswith("/var/lib/docker"):
continue
try:
usage = psutil.disk_usage(partition.mountpoint)
disks.append(DiskStats(
mount_point=partition.mountpoint,
device=partition.device,
fstype=partition.fstype,
usage_percent=usage.percent,
total_bytes=usage.total,
used_bytes=usage.used,
free_bytes=usage.free,
))
except (PermissionError, OSError) as e:
logger.debug(f"Skipping {partition.mountpoint}: {e}")
continue
# Sort by mount point for consistent ordering
disks.sort(key=lambda d: d.mount_point)
return disks
def _get_network_stats(self) -> NetworkStats:
"""Get network I/O statistics"""
net_io = psutil.net_io_counters()
return NetworkStats(
bytes_sent=net_io.bytes_sent,
bytes_recv=net_io.bytes_recv,
bytes_total=net_io.bytes_sent + net_io.bytes_recv,
)
def _get_gpu_stats(self) -> GpuStats:
"""Get GPU/VRAM statistics using nvidia-smi"""
try:
# Query nvidia-smi for GPU memory info
result = subprocess.run(
[
"nvidia-smi",
"--query-gpu=name,memory.total,memory.used,memory.free",
"--format=csv,noheader,nounits",
],
capture_output=True,
text=True,
timeout=5,
)
if result.returncode != 0:
logger.debug("nvidia-smi not available or failed")
return GpuStats(available=False)
# Parse output: "NVIDIA GeForce RTX 3080, 10240, 2048, 8192"
line = result.stdout.strip().split("\n")[0] # First GPU
parts = [p.strip() for p in line.split(",")]
if len(parts) >= 4:
name = parts[0]
total_mb = int(parts[1])
used_mb = int(parts[2])
free_mb = int(parts[3])
total_bytes = total_mb * 1024 * 1024
used_bytes = used_mb * 1024 * 1024
free_bytes = free_mb * 1024 * 1024
usage_percent = (used_mb / total_mb * 100) if total_mb > 0 else 0
return GpuStats(
available=True,
name=name,
usage_percent=round(usage_percent, 1),
total_bytes=total_bytes,
used_bytes=used_bytes,
free_bytes=free_bytes,
)
except FileNotFoundError:
logger.debug("nvidia-smi not found - no NVIDIA GPU available")
except subprocess.TimeoutExpired:
logger.warning("nvidia-smi timed out")
except Exception as e:
logger.warning(f"Failed to get GPU stats: {e}")
return GpuStats(available=False)