移除Glances并改造内置资源监控页面
This commit is contained in:
liuwei
@@ -4,6 +4,7 @@ from loguru import logger
|
||||
import os
|
||||
import time
|
||||
import subprocess
|
||||
import socket
|
||||
from datetime import datetime
|
||||
import platform
|
||||
import psutil
|
||||
@@ -23,6 +24,16 @@ system_bp = Blueprint('system', __name__)
|
||||
|
||||
# 记录应用启动时间
|
||||
APP_START_TIME = time.time()
|
||||
# 记录最近一次网络计数器采样，用于在资源监控页估算上/下行速率。
|
||||
# 这里故意只做“页面级轻量采样”：
|
||||
# 1. 不起额外守护线程，避免为了展示速率再引入常驻后台任务；
|
||||
# 2. 只有用户刷新/轮询资源页时才计算速率，开销接近于零；
|
||||
# 3. 即便进程重启缓存丢失，也只会让第一次速率显示为 0，不影响整体可用性。
|
||||
NETWORK_IO_SAMPLE = {
|
||||
"timestamp": 0.0,
|
||||
"bytes_sent": 0,
|
||||
"bytes_recv": 0,
|
||||
}
|
||||
|
||||
|
||||
def _system_config_path() -> str:
|
||||
@@ -68,6 +79,178 @@ def _format_bytes_to_mb(value: int) -> float:
|
||||
return round((_safe_float(value, 0.0) / 1024 / 1024), 2)
|
||||
|
||||
|
||||
def _safe_divide(numerator: float, denominator: float, default: float = 0.0) -> float:
|
||||
"""安全除法，避免速率与占比计算时被 0 除打断。"""
|
||||
try:
|
||||
if not denominator:
|
||||
return default
|
||||
return numerator / denominator
|
||||
except Exception:
|
||||
return default
|
||||
|
||||
|
||||
def _primary_disk_path() -> str:
|
||||
"""返回当前系统最稳妥的主盘路径。"""
|
||||
# 资源监控页既要兼容你本地 Windows 开发环境，也要兼容线上 Linux：
|
||||
# 1. 优先用系统根目录，Linux 下是 /；
|
||||
# 2. Windows 下会自动变成当前盘符根路径；
|
||||
# 3. 避免把磁盘路径硬编码成 /，导致本地调试时报错。
|
||||
return os.path.abspath(os.sep)
|
||||
|
||||
|
||||
def _format_datetime_text(timestamp_value: float | int | None) -> str:
|
||||
"""把时间戳格式化为后台页面可直接展示的文本。"""
|
||||
if not timestamp_value:
|
||||
return "-"
|
||||
try:
|
||||
return datetime.fromtimestamp(float(timestamp_value)).strftime("%Y-%m-%d %H:%M:%S")
|
||||
except Exception:
|
||||
return "-"
|
||||
|
||||
|
||||
def _sample_network_speed() -> dict:
|
||||
"""根据两次页面采样估算网络上下行速率。"""
|
||||
counters = psutil.net_io_counters()
|
||||
now = time.time()
|
||||
current_sent = _safe_int(getattr(counters, "bytes_sent", 0))
|
||||
current_recv = _safe_int(getattr(counters, "bytes_recv", 0))
|
||||
last_timestamp = _safe_float(NETWORK_IO_SAMPLE.get("timestamp"))
|
||||
elapsed = max(now - last_timestamp, 0.0)
|
||||
|
||||
upload_speed = 0.0
|
||||
download_speed = 0.0
|
||||
if elapsed > 0 and last_timestamp > 0:
|
||||
upload_speed = _safe_divide(current_sent - _safe_int(NETWORK_IO_SAMPLE.get("bytes_sent")), elapsed, 0.0)
|
||||
download_speed = _safe_divide(current_recv - _safe_int(NETWORK_IO_SAMPLE.get("bytes_recv")), elapsed, 0.0)
|
||||
upload_speed = max(upload_speed, 0.0)
|
||||
download_speed = max(download_speed, 0.0)
|
||||
|
||||
NETWORK_IO_SAMPLE["timestamp"] = now
|
||||
NETWORK_IO_SAMPLE["bytes_sent"] = current_sent
|
||||
NETWORK_IO_SAMPLE["bytes_recv"] = current_recv
|
||||
|
||||
return {
|
||||
"bytes_sent": current_sent,
|
||||
"bytes_recv": current_recv,
|
||||
"upload_speed_bps": round(upload_speed, 2),
|
||||
"download_speed_bps": round(download_speed, 2),
|
||||
}
|
||||
|
||||
|
||||
def _extract_server_runtime_snapshot() -> dict:
|
||||
"""构建资源监控页使用的轻量服务器运行态快照。"""
|
||||
# 这套快照有意只覆盖“日常观察最有价值”的内容：
|
||||
# 1. 主机资源：CPU / 内存 / 磁盘 / 网络；
|
||||
# 2. 应用进程：当前 ABOT 进程是否活着、吃了多少资源；
|
||||
# 3. 基础设施：MySQL / Redis 继续复用现有摘要探测；
|
||||
# 4. 不再依赖 glances 进程，部署和运维负担会轻很多。
|
||||
server = current_app.dashboard_server
|
||||
current_process = psutil.Process(os.getpid())
|
||||
virtual_memory = psutil.virtual_memory()
|
||||
swap_memory = psutil.swap_memory()
|
||||
cpu_usage = psutil.cpu_percent(interval=None)
|
||||
process_cpu_usage = current_process.cpu_percent(interval=None)
|
||||
boot_time = psutil.boot_time()
|
||||
network_sample = _sample_network_speed()
|
||||
disk_io = psutil.disk_io_counters()
|
||||
try:
|
||||
load_values = os.getloadavg()
|
||||
except (AttributeError, OSError):
|
||||
load_values = (0.0, 0.0, 0.0)
|
||||
|
||||
disk_items = []
|
||||
seen_mountpoints = set()
|
||||
for partition in psutil.disk_partitions(all=False):
|
||||
mountpoint = str(getattr(partition, "mountpoint", "") or "").strip()
|
||||
if not mountpoint or mountpoint in seen_mountpoints:
|
||||
continue
|
||||
seen_mountpoints.add(mountpoint)
|
||||
try:
|
||||
usage = psutil.disk_usage(mountpoint)
|
||||
except Exception:
|
||||
continue
|
||||
disk_items.append({
|
||||
"device": str(getattr(partition, "device", "") or "").strip() or mountpoint,
|
||||
"mountpoint": mountpoint,
|
||||
"fstype": str(getattr(partition, "fstype", "") or "").strip(),
|
||||
"total_bytes": _safe_int(getattr(usage, "total", 0)),
|
||||
"used_bytes": _safe_int(getattr(usage, "used", 0)),
|
||||
"free_bytes": _safe_int(getattr(usage, "free", 0)),
|
||||
"usage_percent": round(_safe_float(getattr(usage, "percent", 0.0)), 1),
|
||||
})
|
||||
disk_items.sort(key=lambda item: item.get("usage_percent", 0.0), reverse=True)
|
||||
|
||||
primary_disk_usage = psutil.disk_usage(_primary_disk_path())
|
||||
process_memory = current_process.memory_info()
|
||||
try:
|
||||
open_files = len(current_process.open_files())
|
||||
except Exception:
|
||||
open_files = 0
|
||||
|
||||
try:
|
||||
tcp_connections = current_process.connections(kind="inet")
|
||||
established_connections = sum(
|
||||
1 for conn in tcp_connections if str(getattr(conn, "status", "") or "").upper() == "ESTABLISHED"
|
||||
)
|
||||
except Exception:
|
||||
established_connections = 0
|
||||
|
||||
return {
|
||||
"timestamp": datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
|
||||
"server": {
|
||||
"hostname": socket.gethostname(),
|
||||
"os": platform.system(),
|
||||
"os_version": platform.version(),
|
||||
"python_version": platform.python_version(),
|
||||
"boot_time": _format_datetime_text(boot_time),
|
||||
"uptime_seconds": round(max(time.time() - boot_time, 0), 2),
|
||||
},
|
||||
"cpu": {
|
||||
"usage_percent": round(cpu_usage, 1),
|
||||
"logical_count": psutil.cpu_count(logical=True) or 0,
|
||||
"physical_count": psutil.cpu_count(logical=False) or 0,
|
||||
"load_1": round(_safe_float(load_values[0]), 2),
|
||||
"load_5": round(_safe_float(load_values[1]), 2),
|
||||
"load_15": round(_safe_float(load_values[2]), 2),
|
||||
},
|
||||
"memory": {
|
||||
"usage_percent": round(_safe_float(getattr(virtual_memory, "percent", 0.0)), 1),
|
||||
"total_bytes": _safe_int(getattr(virtual_memory, "total", 0)),
|
||||
"used_bytes": _safe_int(getattr(virtual_memory, "used", 0)),
|
||||
"available_bytes": _safe_int(getattr(virtual_memory, "available", 0)),
|
||||
"swap_usage_percent": round(_safe_float(getattr(swap_memory, "percent", 0.0)), 1),
|
||||
"swap_total_bytes": _safe_int(getattr(swap_memory, "total", 0)),
|
||||
"swap_used_bytes": _safe_int(getattr(swap_memory, "used", 0)),
|
||||
},
|
||||
"disk": {
|
||||
"primary_usage_percent": round(_safe_float(getattr(primary_disk_usage, "percent", 0.0)), 1),
|
||||
"primary_total_bytes": _safe_int(getattr(primary_disk_usage, "total", 0)),
|
||||
"primary_used_bytes": _safe_int(getattr(primary_disk_usage, "used", 0)),
|
||||
"io_read_bytes": _safe_int(getattr(disk_io, "read_bytes", 0)) if disk_io else 0,
|
||||
"io_write_bytes": _safe_int(getattr(disk_io, "write_bytes", 0)) if disk_io else 0,
|
||||
"items": disk_items[:8],
|
||||
},
|
||||
"network": {
|
||||
**network_sample,
|
||||
"established_connections": established_connections,
|
||||
},
|
||||
"process": {
|
||||
"pid": current_process.pid,
|
||||
"cpu_percent": round(process_cpu_usage, 1),
|
||||
"memory_percent": round(current_process.memory_percent(), 2),
|
||||
"memory_rss_bytes": _safe_int(getattr(process_memory, "rss", 0)),
|
||||
"thread_count": current_process.num_threads(),
|
||||
"open_files": open_files,
|
||||
"create_time": _format_datetime_text(current_process.create_time()),
|
||||
"uptime_seconds": round(max(time.time() - current_process.create_time(), 0), 2),
|
||||
},
|
||||
"infrastructure": {
|
||||
"mysql": _extract_mysql_runtime_snapshot(server.db_manager),
|
||||
"redis": _extract_redis_runtime_snapshot(server.db_manager),
|
||||
},
|
||||
}
|
||||
|
||||
|
||||
def _extract_mysql_runtime_snapshot(db_manager) -> dict:
|
||||
"""采集 MySQL 运行态摘要。
|
||||
|
||||
@@ -761,17 +944,11 @@ def api_docs():
|
||||
@system_bp.route('/system_status')
|
||||
@login_required
|
||||
def system_status():
|
||||
src = request.args.get('src')
|
||||
if not src:
|
||||
try:
|
||||
server = current_app.dashboard_server
|
||||
glances = getattr(server.robot, "config").glances if hasattr(server.robot, "config") else {}
|
||||
host = glances.get("host", "127.0.0.1")
|
||||
port = glances.get("port", 61208)
|
||||
src = f"http://{host}:{port}/"
|
||||
except Exception:
|
||||
src = "http://127.0.0.1:61208/"
|
||||
return render_template('system_status.html', src_url=src)
|
||||
# 资源监控页改为项目内置轻量面板：
|
||||
# 1. 不再依赖 glances 独立进程；
|
||||
# 2. 页面只消费当前服务自身的 API；
|
||||
# 3. 线上部署时不用额外开放 61208 之类的端口。
|
||||
return render_template('system_status.html')
|
||||
|
||||
|
||||
@system_bp.route('/system_llm')
|
||||
@@ -811,6 +988,20 @@ def api_system_info():
|
||||
return jsonify({"success": False, "error": str(e)}), 500
|
||||
|
||||
|
||||
@system_bp.route('/api/system_status_overview')
|
||||
@login_required
|
||||
def api_system_status_overview():
|
||||
"""返回资源监控页使用的轻量服务器快照。"""
|
||||
try:
|
||||
return jsonify({
|
||||
"success": True,
|
||||
"data": _extract_server_runtime_snapshot(),
|
||||
})
|
||||
except Exception as e:
|
||||
logger.error(f"获取资源监控快照失败: {e}")
|
||||
return jsonify({"success": False, "error": str(e)}), 500
|
||||
|
||||
|
||||
@system_bp.route('/api/system_health_summary')
|
||||
@login_required
|
||||
def api_system_health_summary():
|
||||
|
||||
Reference in New Issue
Block a user