cloudron-box/src/metrics.js

'use strict';

exports = module.exports = {
    getSystem,
    getSystemStream,

    getContainer,

    sendToGraphite
};

const apps = require('./apps.js'),
    assert = require('assert'),
    BoxError = require('./boxerror.js'),
    constants = require('./constants.js'),
    debug = require('debug')('box:metrics'),
    docker = require('./docker.js'),
    fs = require('fs'),
    net = require('net'),
    os = require('os'),
    { Readable } = require('stream'),
    safe = require('safetydance'),
    services = require('./services.js'),
    superagent = require('./superagent.js');

async function getDockerMetrics() {
    const allAddons = [ 'turn', 'mail', 'mongodb', 'mysql', 'postgresql' ];

    const containerNames = allAddons;
    for (const app of await apps.list()) {
        if (app.containerId) containerNames.push(app.containerId); // containerId can be null if app is installing
        if (app.manifest.addons?.redis && app.enableRedis) containerNames.push(`redis-${app.id}`);
    }

    const metrics = {};
    for (const containerName of containerNames) {
        const stats = await docker.getStats(containerName);
        if (Object.keys(stats.memory_stats).length === 0) continue; // the container is stopped. better not to inspect and check State since a race is possible

        const networkRead = stats.networks ? stats.networks.eth0.rx_bytes : 0; // in host mode (turn), networks is missing
        const networkWrite = stats.networks ? stats.networks.eth0.tx_bytes : 0; // in host mode (turn), networks is missing

        const memUsed = stats.memory_stats.usage;
        const memMax = stats.memory_stats.limit;

        const blkioStats = stats.blkio_stats.io_service_bytes_recursive;
        const blockRead = blkioStats.filter(entry => entry.op === 'read').reduce((sum, entry) => sum + entry.value, 0);
        const blockWrite = blkioStats.filter(entry => entry.op === 'write').reduce((sum, entry) => sum + entry.value, 0);

        const cpuUsage = stats.cpu_stats.cpu_usage.total_usage / 1000; // convert to msecs (to match system metrics)

        metrics[containerName] = { networkRead, networkWrite, blockRead, blockWrite, memUsed, memMax, cpuUsage };
    }

    return metrics;
}

async function getMemoryMetrics() {
    const output = await fs.promises.readFile('/proc/meminfo', { encoding: 'utf8' });

    const totalMemoryMatch = output.match(/^MemTotal:\s+(\d+)/m);
    const freeMemoryMatch = output.match(/^MemFree:\s+(\d+)/m);
    const buffersMatch = output.match(/^Buffers:\s+(\d+)/m);
    const cachedMatch = output.match(/^Cached:\s+(\d+)/m);

    if (!totalMemoryMatch || !freeMemoryMatch || !buffersMatch || !cachedMatch) throw new BoxError(BoxError.EXTERNAL_ERROR, 'Could not find memory used');

    const memoryUsed = parseInt(totalMemoryMatch[1]) * 1024 - parseInt(freeMemoryMatch[1]) * 1024 - parseInt(buffersMatch[1]) * 1024 - parseInt(cachedMatch[1]) * 1024;

    const swapTotalMatch = output.match(/^SwapTotal:\s+(\d+)/m);
    const swapFreeMatch = output.match(/^SwapFree:\s+(\d+)/m);

    if (!swapTotalMatch || !swapFreeMatch) throw new BoxError(BoxError.EXTERNAL_ERROR, 'Could not find swap used');

    const swapUsed = parseInt(swapTotalMatch[1]) * 1024 - parseInt(swapFreeMatch[1]) * 1024;

    return {
        memoryUsed,
        swapUsed
    };
}

async function getCpuMetrics() {
    const cpus = os.cpus();
    const userMsecs = cpus.map(c => c.times.user).reduce((p, c) => p+c);
    const sysMsecs = cpus.map(c => c.times.sys).reduce((p, c) => p+c);

    return { userMsecs, sysMsecs }; // these values are the times spent since system start
}

async function sendToGraphite() {
    // debug('sendStatsToGraphite: collecting stats');

    const graphiteMetrics = [];

    const memoryMetrics = await getMemoryMetrics();
    graphiteMetrics.push({ path: `cloudron.system.memory-used`, value: memoryMetrics.memoryUsed });
    graphiteMetrics.push({ path: `cloudron.system.swap-used`, value: memoryMetrics.swapUsed });

    const cpuMetrics = await getCpuMetrics();
    graphiteMetrics.push({ path: `cloudron.system.cpu-user`, value: cpuMetrics.userMsecs  });
    graphiteMetrics.push({ path: `cloudron.system.cpu-sys`, value: cpuMetrics.sysMsecs  });

    const dockerMetrics = await getDockerMetrics();
    for (const [name, value] of Object.entries(dockerMetrics)) {
        graphiteMetrics.push(
            { path: `cloudron.container-${name}.network-read`, value: value.networkRead },
            { path: `cloudron.container-${name}.network-write`, value: value.networkWrite },
            { path: `cloudron.container-${name}.blockio-read`, value: value.blockRead },
            { path: `cloudron.container-${name}.blockio-write`, value: value.blockWrite },
            { path: `cloudron.container-${name}.mem-used`, value: value.memUsed },
            { path: `cloudron.container-${name}.mem-max`, value: value.memMax },
            { path: `cloudron.container-${name}.cpu-usage`, value: value.cpuUsage },
        );
    }

    return new Promise((resolve) => {
        const client = new net.Socket();
        client.connect(constants.GRAPHITE_PORT, '127.0.0.1', () => {
            debug('connected to graphite');

            const now = Math.floor(Date.now() / 1000);

            for (const metric of graphiteMetrics) {
                client.write(`${metric.path} ${metric.value} ${now}\n`);
            }
            client.end();
        });

        client.on('error', (error) => {
            debug(error);
            resolve();
        });

        client.on('end', () => {
            debug('sent to graphite');
            resolve();
        });
    });
}

// for testing locally: curl 'http://${graphite-ip}:8000/graphite-web/render?format=json&from=-1min&target=absolute(collectd.localhost.du-docker.capacity-usage)'
// the datapoint is (value, timestamp) https://graphite.readthedocs.io/en/latest/
async function getGraphiteUrl() {
    const [error, result] = await safe(docker.inspect('graphite'));
    if (error && error.reason === BoxError.NOT_FOUND) return { status: exports.SERVICE_STATUS_STOPPED };
    if (error) throw error;

    const ip = safe.query(result, 'NetworkSettings.Networks.cloudron.IPAddress', null);
    if (!ip) throw new BoxError(BoxError.INACTIVE, 'Error getting IP of graphite service');

    return `http://${ip}:8000/graphite-web/render`;
}

async function getContainer(name, options) {
    assert.strictEqual(typeof name, 'string');
    assert.strictEqual(typeof options, 'object');

    const { fromSecs, intervalSecs, noNullPoints } = options;

    const graphiteUrl = await getGraphiteUrl();

    const targets = [
        // perSecond is nonNegativeDerivative over time . this value is the cpu usage in msecs .
        // (cpu usage msecs) / (cpus * 1000) is the percent but over all cpus. times 100 is the percent.
        // but the y-scale is cpus times 100. so, we only need to scale by 0.1
        `scale(perSecond(cloudron.container-${name}.cpu-usage),0.1)`,
        `summarize(cloudron.container-${name}.mem-used, "${intervalSecs}s", "avg")`,
        `summarize(cloudron.container-${name}.blockio-read, "${intervalSecs}s", "sum")`,
        `summarize(cloudron.container-${name}.blockio-write, "${intervalSecs}s", "sum")`,
        `summarize(cloudron.container-${name}.network-read, "${intervalSecs}s", "sum")`,
        `summarize(cloudron.container-${name}.network-write, "${intervalSecs}s", "sum")`,
        `summarize(cloudron.container-${name}.blockio-read, "${intervalSecs}s", "max")`,
        `summarize(cloudron.container-${name}.blockio-write, "${intervalSecs}s", "max")`,
        `summarize(cloudron.container-${name}.network-read, "${intervalSecs}s", "max")`,
        `summarize(cloudron.container-${name}.network-write, "${intervalSecs}s", "max")`,
    ];

    const results = [];

    for (const target of targets) {
        const query = {
            target: target,
            format: 'json',
            from: `-${fromSecs}s`,
            until: 'now',
            noNullPoints: !!noNullPoints
        };

        const [error, response] = await safe(superagent.get(graphiteUrl).query(query).timeout(30 * 1000).ok(() => true));
        if (error) throw new BoxError(BoxError.NETWORK_ERROR, error);
        if (response.status !== 200) throw new BoxError(BoxError.EXTERNAL_ERROR, `Unknown error with ${target}: ${response.status} ${response.text}`);

        results.push(response.body[0] && response.body[0].datapoints ? response.body[0].datapoints : []);
    }

    // results are datapoints[[value, ts], [value, ts], ...];
    return {
        cpu: results[0],
        memory: results[1],
        blockRead: results[2],
        blockWrite: results[3],
        networkRead: results[4],
        networkWrite: results[5],
        blockReadTotal: results[6][0] && results[6][0][0] ? results[6][0][0] : 0,
        blockWriteTotal: results[7][0] && results[7][0][0] ? results[7][0][0] : 0,
        networkReadTotal: results[8][0] && results[8][0][0] ? results[8][0][0] : 0,
        networkWriteTotal: results[9][0] && results[9][0][0] ? results[9][0][0] : 0,
        cpuCount: os.cpus().length
    };
}

async function readSystemFromGraphite(options) {
    assert.strictEqual(typeof options, 'object');

    const { fromSecs, intervalSecs, noNullPoints } = options;

    const graphiteUrl = await getGraphiteUrl();

    // example: curl 'http://172.18.30.5:8000/graphite-web/render?target=cloudron.system.cpu-user&target=cloudron.system.cpu-sys&format=json&from=-1min&until=now&noNullPoints=false' | python3 -m json.tool
    const targets = [
        // perSecond is nonNegativeDerivative over time . this value is the cpu usage in msecs .
        // (cpu usage msecs) / (cpus * 1000) is the percent but over all cpus. times 100 is the percent.
        // but the y-scale is cpus times 100. so, we only need to scale by 0.1
        `scale(perSecond(sumSeries(cloudron.system.cpu-user,cloudron.system.cpu-sys)),0.1)`,
        `summarize(cloudron.system.memory-used, "${intervalSecs}s", "avg")`,
        `summarize(cloudron.system.swap-used, "${intervalSecs}s", "avg")`,
    ];

    const results = [];

    for (const target of targets) {
        const query = {
            target: target,
            format: 'json',
            from: `-${fromSecs}s`,
            until: 'now',
            noNullPoints: !!noNullPoints
        };

        const [error, response] = await safe(superagent.get(graphiteUrl).query(query).timeout(30 * 1000).ok(() => true));
        if (error) throw new BoxError(BoxError.NETWORK_ERROR, error);
        if (response.status !== 200) throw new BoxError(BoxError.EXTERNAL_ERROR, `Unknown error with ${target}: ${response.status} ${response.text}`);

        results.push(response.body[0] && response.body[0].datapoints ? response.body[0].datapoints : []);
    }

    return {
        cpu: results[0],
        memory: results[1],
        swap: results[2]
    };
}

async function getSystem(options) {
    assert.strictEqual(typeof options, 'object');

    const systemStats = await readSystemFromGraphite(options);

    const appStats = {};
    for (const app of await apps.list()) {
        appStats[app.id] = await getContainer(app.id, options);
    }

    const serviceStats = {};
    for (const serviceId of await services.listServices()) {
        serviceStats[serviceId] = await getContainer(serviceId, options);
    }

    return {
        cpu: systemStats.cpu,
        memory: systemStats.memory,
        swap: systemStats.swap,
        apps: appStats,
        services: serviceStats,
        cpuCount: os.cpus().length
    };
}

async function getSystemStream(options) {
    assert.strictEqual(typeof options, 'object');

    const INTERVAL_MSECS = options.intervalMsecs || 5000;
    let intervalId = null, oldCpuMetrics = null;

    const metricsStream = new Readable({
        read(/*size*/) { /* ignored, we push via interval */ },
        destroy(error, callback) {
            clearInterval(intervalId);
            callback(error);
        }
    });

    intervalId = setInterval(async () => {
        const memoryMetrics = await getMemoryMetrics();
        const cpuMetrics = await getCpuMetrics();

        const cpuPercent = oldCpuMetrics ? (cpuMetrics.userMsecs + cpuMetrics.sysMsecs - oldCpuMetrics.userMsecs - oldCpuMetrics.sysMsecs) * 0.1 / (INTERVAL_MSECS/1000) : null;
        oldCpuMetrics = cpuMetrics;

        const now = Date.now() / 1000;
        metricsStream.push(JSON.stringify({
            cpu: [ cpuPercent, now ],
            memory: [ memoryMetrics.memoryUsed, now ],
            swap: [ memoryMetrics.swapUsed, now ],
        }));
    }, INTERVAL_MSECS);

    return metricsStream;
}