@@ -0,0 +1,108 @@
1+import { monitorEventLoopDelay, performance } from "node:perf_hooks";
2+3+const EVENT_LOOP_MONITOR_RESOLUTION_MS = 20;
4+const EVENT_LOOP_DELAY_WARN_MS = 1_000;
5+const EVENT_LOOP_UTILIZATION_WARN = 0.95;
6+const CPU_CORE_RATIO_WARN = 0.9;
7+8+type EventLoopDelayMonitor = ReturnType<typeof monitorEventLoopDelay>;
9+type EventLoopUtilization = ReturnType<typeof performance.eventLoopUtilization>;
10+type CpuUsage = ReturnType<typeof process.cpuUsage>;
11+12+export type GatewayEventLoopHealthReason = "event_loop_delay" | "event_loop_utilization" | "cpu";
13+14+export type GatewayEventLoopHealth = {
15+degraded: boolean;
16+reasons: GatewayEventLoopHealthReason[];
17+intervalMs: number;
18+delayP99Ms: number;
19+delayMaxMs: number;
20+utilization: number;
21+cpuCoreRatio: number;
22+};
23+24+export type GatewayEventLoopHealthMonitor = {
25+snapshot: () => GatewayEventLoopHealth | undefined;
26+stop: () => void;
27+};
28+29+function roundMetric(value: number, digits = 3): number {
30+if (!Number.isFinite(value)) {
31+return 0;
32+}
33+const factor = 10 ** digits;
34+return Math.round(value * factor) / factor;
35+}
36+37+function nanosecondsToMilliseconds(value: number): number {
38+return roundMetric(value / 1_000_000, 1);
39+}
40+41+export function createGatewayEventLoopHealthMonitor(): GatewayEventLoopHealthMonitor {
42+let monitor: EventLoopDelayMonitor | null = null;
43+let lastWallAt = Date.now();
44+let lastCpuUsage: CpuUsage | null = process.cpuUsage();
45+let lastEventLoopUtilization: EventLoopUtilization | null = performance.eventLoopUtilization();
46+47+try {
48+monitor = monitorEventLoopDelay({ resolution: EVENT_LOOP_MONITOR_RESOLUTION_MS });
49+monitor.enable();
50+monitor.reset();
51+} catch {
52+monitor = null;
53+}
54+55+return {
56+snapshot: () => {
57+if (!monitor || !lastCpuUsage || !lastEventLoopUtilization || lastWallAt <= 0) {
58+return undefined;
59+}
60+61+const now = Date.now();
62+const intervalMs = Math.max(1, now - lastWallAt);
63+const cpuUsage = process.cpuUsage(lastCpuUsage);
64+const currentEventLoopUtilization = performance.eventLoopUtilization();
65+const utilization = roundMetric(
66+performance.eventLoopUtilization(currentEventLoopUtilization, lastEventLoopUtilization)
67+.utilization,
68+);
69+const delayP99Ms = nanosecondsToMilliseconds(monitor.percentile(99));
70+const delayMaxMs = nanosecondsToMilliseconds(monitor.max);
71+const cpuTotalMs = roundMetric((cpuUsage.user + cpuUsage.system) / 1_000, 1);
72+const cpuCoreRatio = roundMetric(cpuTotalMs / intervalMs);
73+const reasons: GatewayEventLoopHealthReason[] = [];
74+75+if (delayP99Ms >= EVENT_LOOP_DELAY_WARN_MS || delayMaxMs >= EVENT_LOOP_DELAY_WARN_MS) {
76+reasons.push("event_loop_delay");
77+}
78+if (utilization >= EVENT_LOOP_UTILIZATION_WARN) {
79+reasons.push("event_loop_utilization");
80+}
81+if (cpuCoreRatio >= CPU_CORE_RATIO_WARN) {
82+reasons.push("cpu");
83+}
84+85+monitor.reset();
86+lastWallAt = now;
87+lastCpuUsage = process.cpuUsage();
88+lastEventLoopUtilization = currentEventLoopUtilization;
89+90+return {
91+degraded: reasons.length > 0,
92+ reasons,
93+ intervalMs,
94+ delayP99Ms,
95+ delayMaxMs,
96+ utilization,
97+ cpuCoreRatio,
98+};
99+},
100+stop: () => {
101+monitor?.disable();
102+monitor = null;
103+lastWallAt = 0;
104+lastCpuUsage = null;
105+lastEventLoopUtilization = null;
106+},
107+};
108+}
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。