hectorzhao/lislgosms
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: deduplicate filesystem mounts in monitoring

Browse Source
This commit is contained in:
hectorzhao
2026-08-31 23:39:24 +08:00
parent 12668cee87
commit 1f2dfb5caf
14 changed files with 246 additions and 51 deletions
Download Patch File Download Diff File Expand all files Collapse all files
api/src/infrastructure-monitoring/infrastructure-alert-settings.service.ts
+3 -2
View File
@@ -6,6 +6,7 @@ import { mkdir, readFile, rename, rm, writeFile } from 'node:fs/promises';
import { dirname } from 'node:path';
import { promisify } from 'node:util';
import { PrismaService } from '../prisma/prisma.service';
import { FILESYSTEM_USAGE_PERCENT } from './filesystem-metrics';
import type { InfrastructureAlertSettings, InfrastructureAlertThresholds } from './infrastructure-monitoring.contracts';
const execFileAsync = promisify(execFile);
@@ -13,7 +14,7 @@ const execFileAsync = promisify(execFile);
export const ALERT_THRESHOLD_DEFINITIONS = [
{ key: 'hostCpu', label: '主机 CPU 使用率', unit: '%', min: 1, max: 100, step: 1, warning: 80, critical: 90, expr: '100 - (avg by (instance) (rate(node_cpu_seconds_total{mode="idle"}[5m])) * 100)', names: ['HostCpuUsageWarning', 'HostCpuUsageCritical'], service: 'host', durations: ['10m', '5m'] },
{ key: 'hostMemory', label: '主机内存使用率', unit: '%', min: 1, max: 100, step: 1, warning: 85, critical: 95, expr: '(1 - node_memory_MemAvailable_bytes / node_memory_MemTotal_bytes) * 100', names: ['HostMemoryUsageWarning', 'HostMemoryUsageCritical'], service: 'host', durations: ['10m', '5m'] },
{ key: 'hostDisk', label: '磁盘（所有挂载点）使用率', unit: '%', min: 1, max: 100, step: 1, warning: 80, critical: 90, expr: '(1 - node_filesystem_avail_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"} / node_filesystem_size_bytes{device=~"/dev/.+",fstype!~"tmpfs|devtmpfs|overlay|squashfs|ramfs"}) * 100', names: ['HostRootDiskUsageWarning', 'HostRootDiskUsageCritical'], service: 'host', durations: ['15m', '5m'] },
{ key: 'hostDisk', label: '磁盘（独立文件系统）使用率', unit: '%', min: 1, max: 100, step: 1, warning: 80, critical: 90, expr: FILESYSTEM_USAGE_PERCENT, names: ['HostRootDiskUsageWarning', 'HostRootDiskUsageCritical'], service: 'host', durations: ['15m', '5m'] },
{ key: 'apiError', label: 'API 5xx 错误率', unit: '%', min: 0.1, max: 100, step: 0.1, warning: 1, critical: 5, expr: '100 * sum(rate(cmpp_api_http_requests_total{status=~"5.."}[5m])) / clamp_min(sum(rate(cmpp_api_http_requests_total[5m])), 0.001)', guard: 'sum(increase(cmpp_api_http_requests_total{status=~"5.."}[5m])) >= 5', names: ['CmppApiHttpErrorRateWarning', 'CmppApiHttpErrorRateCritical'], service: 'api', durations: ['5m', '5m'] },
{ key: 'apiLatency', label: 'API P95 响应时间', unit: '秒', min: 0.1, max: 60, step: 0.1, warning: 1, critical: 3, expr: 'histogram_quantile(0.95, sum by (le) (rate(cmpp_api_http_request_duration_seconds_bucket[10m])))', names: ['CmppApiLatencyWarning', 'CmppApiLatencyCritical'], service: 'api', durations: ['10m', '5m'] },
{ key: 'apiEventLoop', label: 'API 事件循环 P99', unit: '秒', min: 0.01, max: 10, step: 0.01, warning: 0.2, critical: 1, expr: 'cmpp_api_nodejs_event_loop_lag_p99_seconds', names: ['CmppApiEventLoopLagWarning', 'CmppApiEventLoopLagCritical'], service: 'api', durations: ['10m', '5m'] },
@@ -113,7 +114,7 @@ export class InfrastructureAlertSettingsService {
// 低样本量与“未配置容量上限”必须继续作为固定保护条件，避免单次错误或除零结果触发伪告警。
const guard = 'guard' in definition ? ` and (${definition.guard})` : '';
const expr = isWarning ? `(${definition.expr} > ${values[0]}) and (${definition.expr} <= ${values[1]})${guard}` : `(${definition.expr} > ${values[1]})${guard}`;
const diskLocation = definition.key === 'hostDisk' ? ' 挂载点：{{ $labels.mountpoint }}；设备：{{ $labels.device }}。' : '';
const diskLocation = definition.key === 'hostDisk' ? ' 设备：{{ $labels.device }}；文件系统：{{ $labels.fstype }}（绑定挂载已合并）。' : '';
lines.push(` - alert: ${definition.names[index]}`, ` expr: ${expr}`, ` for: ${definition.durations[index]}`, ' labels:', ` severity: ${isWarning ? 'warning' : 'critical'}`, ` service: ${definition.service}`, ' annotations:', ` summary: "${definition.label}${isWarning ? '达到警告阈值' : '达到严重阈值'}"`, ` description: "${definition.label}持续超过${values[index]}${definition.unit}。${diskLocation}"`, ' currentValue: "{{ $value }}"', ` threshold: "${values[index]}${definition.unit}"`);
}
}