smartmetrics/ts/smartmetrics.classes.smartmetrics.ts

import * as plugins from './smartmetrics.plugins.js';
import * as interfaces from './smartmetrics.interfaces.js';

export class SmartMetrics {
  public started = false;
  public sourceNameArg: string;
  public logger: plugins.smartlog.Smartlog;
  public registry: plugins.prom.Registry;
  public maxMemoryMB: number;

  // Prometheus gauges for custom metrics
  private cpuPercentageGauge: plugins.prom.Gauge;
  private memoryPercentageGauge: plugins.prom.Gauge;
  private memoryUsageBytesGauge: plugins.prom.Gauge;
  private systemCpuPercentGauge: plugins.prom.Gauge;
  private systemMemUsedPercentGauge: plugins.prom.Gauge;
  private systemMemUsedBytesGauge: plugins.prom.Gauge;
  private systemLoadAvg1Gauge: plugins.prom.Gauge;
  private systemLoadAvg5Gauge: plugins.prom.Gauge;
  private systemLoadAvg15Gauge: plugins.prom.Gauge;

  // HTTP server for Prometheus endpoint
  private prometheusServer?: plugins.http.Server;
  private prometheusPort?: number;

  public setup() {
    this.registry = new plugins.prom.Registry();
    plugins.prom.collectDefaultMetrics(this.registry);

    // Initialize custom gauges
    this.cpuPercentageGauge = new plugins.prom.Gauge({
      name: 'smartmetrics_cpu_percentage',
      help: 'Current CPU usage percentage',
      registers: [this.registry]
    });

    this.memoryPercentageGauge = new plugins.prom.Gauge({
      name: 'smartmetrics_memory_percentage',
      help: 'Current memory usage percentage',
      registers: [this.registry]
    });

    this.memoryUsageBytesGauge = new plugins.prom.Gauge({
      name: 'smartmetrics_memory_usage_bytes',
      help: 'Current memory usage in bytes',
      registers: [this.registry]
    });

    this.systemCpuPercentGauge = new plugins.prom.Gauge({
      name: 'smartmetrics_system_cpu_percent',
      help: 'System-wide CPU usage percentage',
      registers: [this.registry]
    });

    this.systemMemUsedPercentGauge = new plugins.prom.Gauge({
      name: 'smartmetrics_system_memory_used_percent',
      help: 'System-wide memory usage percentage',
      registers: [this.registry]
    });

    this.systemMemUsedBytesGauge = new plugins.prom.Gauge({
      name: 'smartmetrics_system_memory_used_bytes',
      help: 'System-wide memory used in bytes',
      registers: [this.registry]
    });

    this.systemLoadAvg1Gauge = new plugins.prom.Gauge({
      name: 'smartmetrics_system_load_avg_1',
      help: 'System 1-minute load average',
      registers: [this.registry]
    });

    this.systemLoadAvg5Gauge = new plugins.prom.Gauge({
      name: 'smartmetrics_system_load_avg_5',
      help: 'System 5-minute load average',
      registers: [this.registry]
    });

    this.systemLoadAvg15Gauge = new plugins.prom.Gauge({
      name: 'smartmetrics_system_load_avg_15',
      help: 'System 15-minute load average',
      registers: [this.registry]
    });
  }

  constructor(loggerArg: plugins.smartlog.Smartlog, sourceNameArg: string) {
    this.logger = loggerArg;
    this.sourceNameArg = sourceNameArg;
    this.setup();
    this.checkMemoryLimits();
  }

  private checkMemoryLimits() {
    const heapStats = plugins.v8.getHeapStatistics();
    const maxHeapSizeMB = heapStats.heap_size_limit / 1024 / 1024;
    const totalSystemMemoryMB = plugins.os.totalmem() / 1024 / 1024;

    let dockerMemoryLimitMB = totalSystemMemoryMB;

    // Try cgroup v2 first, then fall back to cgroup v1
    try {
      const cgroupV2 = plugins.fs.readFileSync('/sys/fs/cgroup/memory.max', 'utf8').trim();
      if (cgroupV2 !== 'max') {
        dockerMemoryLimitMB = parseInt(cgroupV2, 10) / 1024 / 1024;
      }
    } catch {
      try {
        const cgroupV1 = plugins.fs.readFileSync(
          '/sys/fs/cgroup/memory/memory.limit_in_bytes',
          'utf8'
        ).trim();
        dockerMemoryLimitMB = parseInt(cgroupV1, 10) / 1024 / 1024;
      } catch {
        // Not running in a container — use system memory
      }
    }

    // Pick the most restrictive limit
    this.maxMemoryMB = Math.min(totalSystemMemoryMB, dockerMemoryLimitMB, maxHeapSizeMB);
  }

  public start() {
    const unattendedStart = async () => {
      if (this.started) {
        return;
      }
      this.started = true;
      while (this.started) {
        this.logger.log('info', `sending heartbeat for ${this.sourceNameArg} with metrics`, {
          eventType: 'heartbeat',
          metrics: await this.getMetrics(),
        });
        await plugins.smartdelay.delayFor(20000, null, true);
      }
    };
    unattendedStart();
  }

  public formatBytes(bytes: number, decimals = 2) {
    if (bytes === 0) return '0 Bytes';

    const k = 1024;
    const dm = decimals < 0 ? 0 : decimals;
    const sizes = ['Bytes', 'KB', 'MB', 'GB', 'TB', 'PB', 'EB', 'ZB', 'YB'];

    const i = Math.floor(Math.log(bytes) / Math.log(k));

    return parseFloat((bytes / Math.pow(k, i)).toFixed(dm)) + ' ' + sizes[i];
  }

  public async getMetrics() {
    let pids: number[] = [];
    try {
      pids = await plugins.pidtree.getChildPids(process.pid);
    } catch {
      // pidtree can fail if process tree cannot be read
    }
    const stats = await plugins.pidusage.getPidUsage([process.pid, ...pids]);

    // Aggregate normalized CPU (0-100% of total machine) across process tree
    let cpuPercentage = 0;
    for (const stat of Object.values(stats)) {
      if (!stat) continue;
      cpuPercentage += stat.cpuNormalizedPercent;
    }
    let cpuUsageText = `${Math.round(cpuPercentage * 100) / 100} %`;

    let memoryUsageBytes = 0;
    for (const stat of Object.values(stats)) {
      if (!stat) continue;
      memoryUsageBytes += stat.memory;
    }

    let memoryPercentage =
      Math.round((memoryUsageBytes / (this.maxMemoryMB * 1024 * 1024)) * 100 * 100) / 100;
    let memoryUsageText = `${memoryPercentage}% | ${this.formatBytes(
      memoryUsageBytes
    )} / ${this.formatBytes(this.maxMemoryMB * 1024 * 1024)}`;

    // Get system-wide metrics
    const systemUsage = await plugins.sysusage.getSystemUsage();

    // Update Prometheus gauges with current values
    if (this.cpuPercentageGauge) {
      this.cpuPercentageGauge.set(cpuPercentage);
    }
    if (this.memoryPercentageGauge) {
      this.memoryPercentageGauge.set(memoryPercentage);
    }
    if (this.memoryUsageBytesGauge) {
      this.memoryUsageBytesGauge.set(memoryUsageBytes);
    }
    if (this.systemCpuPercentGauge) {
      this.systemCpuPercentGauge.set(systemUsage.cpuPercent);
    }
    if (this.systemMemUsedPercentGauge) {
      this.systemMemUsedPercentGauge.set(systemUsage.memUsedPercent);
    }
    if (this.systemMemUsedBytesGauge) {
      this.systemMemUsedBytesGauge.set(systemUsage.memUsedBytes);
    }
    if (this.systemLoadAvg1Gauge) {
      this.systemLoadAvg1Gauge.set(systemUsage.loadAvg1);
    }
    if (this.systemLoadAvg5Gauge) {
      this.systemLoadAvg5Gauge.set(systemUsage.loadAvg5);
    }
    if (this.systemLoadAvg15Gauge) {
      this.systemLoadAvg15Gauge.set(systemUsage.loadAvg15);
    }

    // Calculate Node.js metrics directly
    const cpuUsage = process.cpuUsage();
    const process_cpu_seconds_total = (cpuUsage.user + cpuUsage.system) / 1000000;

    const heapStats = plugins.v8.getHeapStatistics();
    const nodejs_heap_size_total_bytes = heapStats.total_heap_size;

    const nodejs_active_handles_total = 0;
    const nodejs_active_requests_total = 0;

    const returnMetrics: interfaces.IMetricsSnapshot = {
      process_cpu_seconds_total,
      nodejs_active_handles_total,
      nodejs_active_requests_total,
      nodejs_heap_size_total_bytes,
      cpuPercentage,
      cpuUsageText,
      memoryPercentage,
      memoryUsageBytes,
      memoryUsageText,
      systemCpuPercent: systemUsage.cpuPercent,
      systemMemTotalBytes: systemUsage.memTotalBytes,
      systemMemAvailableBytes: systemUsage.memAvailableBytes,
      systemMemUsedBytes: systemUsage.memUsedBytes,
      systemMemUsedPercent: systemUsage.memUsedPercent,
      systemLoadAvg1: systemUsage.loadAvg1,
      systemLoadAvg5: systemUsage.loadAvg5,
      systemLoadAvg15: systemUsage.loadAvg15,
    };
    return returnMetrics;
  }

  public async getPrometheusFormattedMetrics(): Promise<string> {
    // Update metrics to ensure gauges have latest values
    await this.getMetrics();

    // Return Prometheus text exposition format
    return await this.registry.metrics();
  }

  public enablePrometheusEndpoint(port: number = 9090): void {
    if (this.prometheusServer) {
      this.logger.log('warn', 'Prometheus endpoint is already running');
      return;
    }

    this.prometheusServer = plugins.http.createServer(async (req, res) => {
      if (req.url === '/metrics' && req.method === 'GET') {
        try {
          const metrics = await this.getPrometheusFormattedMetrics();
          res.writeHead(200, { 'Content-Type': 'text/plain; version=0.0.4' });
          res.end(metrics);
        } catch (error) {
          res.writeHead(500, { 'Content-Type': 'text/plain' });
          res.end('Error generating metrics');
          this.logger.log('error', 'Error generating Prometheus metrics', error);
        }
      } else {
        res.writeHead(404, { 'Content-Type': 'text/plain' });
        res.end('Not Found');
      }
    });

    this.prometheusPort = port;
    this.prometheusServer.listen(port, () => {
      this.logger.log('info', `Prometheus metrics endpoint available at http://localhost:${port}/metrics`);
    });
  }

  public disablePrometheusEndpoint(): void {
    if (!this.prometheusServer) {
      return;
    }

    const port = this.prometheusPort;
    this.prometheusServer.close(() => {
      this.logger.log('info', `Prometheus metrics endpoint on port ${port} has been shut down`);
    });

    this.prometheusServer = undefined;
    this.prometheusPort = undefined;
  }

  public stop() {
    this.started = false;
    this.disablePrometheusEndpoint();
  }
}