|
| 1 | +/** |
| 2 | + * Health Check HTTP Server |
| 3 | + * |
| 4 | + * Provides /health and /ready endpoints for container orchestration |
| 5 | + * and external monitoring systems. |
| 6 | + */ |
| 7 | + |
| 8 | +import { createServer, IncomingMessage, ServerResponse } from 'http'; |
| 9 | +import { HEALTH_CHECK } from './config.js'; |
| 10 | +import { logger } from './logger.js'; |
| 11 | + |
| 12 | +// ============================================================================ |
| 13 | +// Types |
| 14 | +// ============================================================================ |
| 15 | + |
| 16 | +interface HealthStatus { |
| 17 | + status: 'healthy' | 'degraded' | 'unhealthy'; |
| 18 | + uptime: number; |
| 19 | + memory: { |
| 20 | + heapUsed: number; |
| 21 | + heapTotal: number; |
| 22 | + rss: number; |
| 23 | + }; |
| 24 | + groups: number; |
| 25 | + version: string; |
| 26 | + timestamp: string; |
| 27 | +} |
| 28 | + |
| 29 | +// ============================================================================ |
| 30 | +// State (injected at startup) |
| 31 | +// ============================================================================ |
| 32 | + |
| 33 | +let getGroupCount: () => number = () => 0; |
| 34 | +let getActiveContainers: () => number = () => 0; |
| 35 | + |
| 36 | +export function setHealthCheckDependencies(deps: { |
| 37 | + getGroupCount: () => number; |
| 38 | + getActiveContainers?: () => number; |
| 39 | +}): void { |
| 40 | + getGroupCount = deps.getGroupCount; |
| 41 | + if (deps.getActiveContainers) { |
| 42 | + getActiveContainers = deps.getActiveContainers; |
| 43 | + } |
| 44 | +} |
| 45 | + |
| 46 | +// ============================================================================ |
| 47 | +// Health Status |
| 48 | +// ============================================================================ |
| 49 | + |
| 50 | +function getHealthStatus(): HealthStatus { |
| 51 | + const memUsage = process.memoryUsage(); |
| 52 | + |
| 53 | + return { |
| 54 | + status: 'healthy', |
| 55 | + uptime: process.uptime(), |
| 56 | + memory: { |
| 57 | + heapUsed: Math.round(memUsage.heapUsed / 1024 / 1024), |
| 58 | + heapTotal: Math.round(memUsage.heapTotal / 1024 / 1024), |
| 59 | + rss: Math.round(memUsage.rss / 1024 / 1024), |
| 60 | + }, |
| 61 | + groups: getGroupCount(), |
| 62 | + version: process.env.npm_package_version || '1.0.0', |
| 63 | + timestamp: new Date().toISOString(), |
| 64 | + }; |
| 65 | +} |
| 66 | + |
| 67 | +// ============================================================================ |
| 68 | +// HTTP Server |
| 69 | +// ============================================================================ |
| 70 | + |
| 71 | +function handleRequest(req: IncomingMessage, res: ServerResponse): void { |
| 72 | + const url = req.url || '/'; |
| 73 | + |
| 74 | + // CORS headers for browser access |
| 75 | + res.setHeader('Access-Control-Allow-Origin', '*'); |
| 76 | + res.setHeader('Content-Type', 'application/json'); |
| 77 | + |
| 78 | + if (url === '/health' || url === '/healthz') { |
| 79 | + const health = getHealthStatus(); |
| 80 | + res.statusCode = health.status === 'healthy' ? 200 : 503; |
| 81 | + res.end(JSON.stringify(health, null, 2)); |
| 82 | + } else if (url === '/ready' || url === '/readyz') { |
| 83 | + // Readiness check - are we ready to accept traffic? |
| 84 | + const ready = getGroupCount() > 0; |
| 85 | + res.statusCode = ready ? 200 : 503; |
| 86 | + res.end(JSON.stringify({ |
| 87 | + ready, |
| 88 | + groups: getGroupCount(), |
| 89 | + activeContainers: getActiveContainers(), |
| 90 | + })); |
| 91 | + } else if (url === '/metrics') { |
| 92 | + // Prometheus-style metrics |
| 93 | + const health = getHealthStatus(); |
| 94 | + res.setHeader('Content-Type', 'text/plain'); |
| 95 | + res.end(`# HELP nanogemclaw_uptime_seconds Application uptime |
| 96 | +# TYPE nanogemclaw_uptime_seconds gauge |
| 97 | +nanogemclaw_uptime_seconds ${health.uptime.toFixed(0)} |
| 98 | +
|
| 99 | +# HELP nanogemclaw_memory_heap_bytes Heap memory used |
| 100 | +# TYPE nanogemclaw_memory_heap_bytes gauge |
| 101 | +nanogemclaw_memory_heap_bytes ${health.memory.heapUsed * 1024 * 1024} |
| 102 | +
|
| 103 | +# HELP nanogemclaw_groups_total Number of registered groups |
| 104 | +# TYPE nanogemclaw_groups_total gauge |
| 105 | +nanogemclaw_groups_total ${health.groups} |
| 106 | +`); |
| 107 | + } else { |
| 108 | + res.statusCode = 404; |
| 109 | + res.end(JSON.stringify({ error: 'Not found' })); |
| 110 | + } |
| 111 | +} |
| 112 | + |
| 113 | +let server: ReturnType<typeof createServer> | null = null; |
| 114 | + |
| 115 | +export function startHealthCheckServer(): void { |
| 116 | + if (!HEALTH_CHECK.ENABLED) { |
| 117 | + logger.info('Health check server disabled'); |
| 118 | + return; |
| 119 | + } |
| 120 | + |
| 121 | + server = createServer(handleRequest); |
| 122 | + |
| 123 | + server.listen(HEALTH_CHECK.PORT, () => { |
| 124 | + logger.info( |
| 125 | + { port: HEALTH_CHECK.PORT }, |
| 126 | + 'Health check server started', |
| 127 | + ); |
| 128 | + }); |
| 129 | + |
| 130 | + server.on('error', (err) => { |
| 131 | + logger.error({ err }, 'Health check server error'); |
| 132 | + }); |
| 133 | +} |
| 134 | + |
| 135 | +export function stopHealthCheckServer(): Promise<void> { |
| 136 | + return new Promise((resolve) => { |
| 137 | + if (server) { |
| 138 | + server.close(() => { |
| 139 | + logger.info('Health check server stopped'); |
| 140 | + resolve(); |
| 141 | + }); |
| 142 | + } else { |
| 143 | + resolve(); |
| 144 | + } |
| 145 | + }); |
| 146 | +} |
0 commit comments