Files
LinuxAwesomeBot/monitor.py
T

222 lines
6.9 KiB
Python

"""Linux 系统监控:功耗(RAPL/电池)、CPU、内存、温度、磁盘、网络、进程。"""
import os
import time
from pathlib import Path
import psutil
# ---------- 功耗 ----------
def _read_energy_uj(path: Path) -> int | None:
try:
return int(path.read_text().strip())
except (OSError, ValueError):
return None
def get_power() -> dict:
"""返回功耗信息,单位瓦(W)。优先电池,其次 Intel RAPL。"""
result = {"source": None, "total_watts": None, "domains": {}, "battery": None}
# 1. 电池功率(笔记本放电/充电)
for p in Path("/sys/class/power_supply").glob("BAT*/power_now"):
try:
# power_now 单位微瓦
watts = int(p.read_text().strip()) / 1_000_000
result["battery"] = round(watts, 2)
result["source"] = "battery"
result["total_watts"] = round(watts, 2)
except (OSError, ValueError):
pass
# 2. Intel RAPL energy 差分
rapl_dir = Path("/sys/class/powercap")
domains_now = {}
for i, d in enumerate(sorted(rapl_dir.glob("intel-rapl:*"))):
name_file = d / "name"
energy_file = d / "energy_uj"
if not energy_file.exists():
continue
domain = name_file.read_text().strip() if name_file.exists() else f"rapl{i}"
energy = _read_energy_uj(energy_file)
if energy is not None:
domains_now[domain] = (energy, time.monotonic())
if domains_now:
# 用上一次读数差分计算瓦数
prev = getattr(get_power, "_prev_rapl", None)
get_power._prev_rapl = domains_now
if prev:
for domain, (energy, ts) in domains_now.items():
if domain in prev:
p_energy, p_ts = prev[domain]
dt = ts - p_ts
if dt > 0:
delta = energy - p_energy
if delta < 0: # 计数器回绕
max_uj = 1 << 32
delta += max_uj
watts = delta / 1_000_000 / dt
result["domains"][domain] = round(watts, 2)
if result["domains"]:
result["source"] = result["source"] or "rapl"
if result["total_watts"] is None:
# 累加所有域(通常 dom0 已包含子域,取最大 dom0)
result["total_watts"] = max(result["domains"].values())
return result
# ---------- 其他指标 ----------
def get_cpu() -> dict:
load1, load5, load15 = _safe_getloadavg()
return {
"percent": psutil.cpu_percent(interval=None),
"per_core": psutil.cpu_percent(interval=None, percpu=True),
"load": [load1, load5, load15],
"cores": psutil.cpu_count(logical=True),
}
def _safe_getloadavg():
try:
return psutil.getloadavg()
except (AttributeError, OSError): # Windows
return (0.0, 0.0, 0.0)
def get_memory() -> dict:
vm = psutil.virtual_memory()
sw = psutil.swap_memory()
return {
"used_gb": round(vm.used / 1024**3, 2),
"total_gb": round(vm.total / 1024**3, 2),
"percent": vm.percent,
"swap_percent": sw.percent,
}
def get_temps() -> dict:
try:
sensors = psutil.sensors_temperatures()
except (AttributeError, OSError):
return {}
out = {}
for name, entries in sensors.items():
out[name] = [
{"label": e.label or f"temp{i}", "current": e.current}
for i, e in enumerate(entries)
if e.current is not None
]
return out
def get_disk() -> dict:
out = []
for part in psutil.disk_partitions():
if "loop" in part.device or not part.mountpoint:
continue
try:
u = psutil.disk_usage(part.mountpoint)
out.append({
"device": part.device,
"mount": part.mountpoint,
"used_gb": round(u.used / 1024**3, 1),
"total_gb": round(u.total / 1024**3, 1),
"percent": u.percent,
})
except (PermissionError, OSError):
continue
return out
_net_prev = {"ts": None, "io": None}
def get_net(interval: float = 0.0) -> dict:
"""网络速率;interval>0 时阻塞采样该秒数,否则用上次调用差分。"""
global _net_prev
if interval > 0:
io0 = psutil.net_io_counters()
time.sleep(interval)
io1 = psutil.net_io_counters()
dt = interval
else:
now = time.monotonic()
io1 = psutil.net_io_counters()
if _net_prev["ts"] is None:
_net_prev = {"ts": now, "io": io1}
return {"up_mbps": None, "down_mbps": None}
dt = now - _net_prev["ts"]
io0 = _net_prev["io"]
_net_prev = {"ts": now, "io": io1}
if dt <= 0:
return {"up_mbps": None, "down_mbps": None}
return {
"up_mbps": round((io1.bytes_sent - io0.bytes_sent) * 8 / dt / 1e6, 2),
"down_mbps": round((io1.bytes_recv - io0.bytes_recv) * 8 / dt / 1e6, 2),
}
_primed = False
_prime_ts = 0.0
def get_top_processes(n: int = 10) -> list:
"""按 CPU 占用排序的 TOP N 进程。首次(或距上次超过60秒)需先基准采样。"""
global _primed, _prime_ts
if not _primed or time.monotonic() - _prime_ts > 60:
for p in psutil.process_iter(["pid"]):
try:
p.cpu_percent(interval=None)
except (psutil.NoSuchProcess, psutil.AccessDenied):
pass
time.sleep(0.5)
_primed = True
_prime_ts = time.monotonic()
procs = []
for p in psutil.process_iter(["pid", "name"]):
try:
if p.info["pid"] in (0, 4): # Windows 的 Idle/System 不计入排行
continue
cpu = p.cpu_percent(interval=None)
mem = p.memory_info().rss / 1024**2
procs.append({
"pid": p.info["pid"],
"name": (p.info["name"] or "?")[:24],
"cpu": round(cpu, 1),
"mem_mb": round(mem, 1),
})
except (psutil.NoSuchProcess, psutil.AccessDenied, psutil.ZombieProcess):
continue
_prime_ts = time.monotonic()
procs.sort(key=lambda x: x["cpu"], reverse=True)
return procs[:n]
def get_uptime() -> float:
try:
with open("/proc/uptime") as f:
return float(f.read().split()[0])
except OSError:
return time.time() - psutil.boot_time()
def get_status(top_n: int | None = None) -> dict:
"""系统总览。top_n 为 None 时不枚举进程(Windows 上较慢)。"""
data = {
"power": get_power(),
"cpu": get_cpu(),
"memory": get_memory(),
"temps": get_temps(),
"disk": get_disk(),
"net": get_net(),
"top": [],
"uptime_s": get_uptime(),
}
if top_n is not None:
data["top"] = get_top_processes(top_n)
return data