Files
cloudron-box/src/metrics.js
2025-05-21 18:09:02 +02:00

324 lines
12 KiB
JavaScript

'use strict';
exports = module.exports = {
getSystem,
getSystemStream,
getContainers,
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'),
execSync = require('child_process').execSync,
net = require('net'),
os = require('os'),
{ Readable } = require('stream'),
safe = require('safetydance'),
services = require('./services.js'),
superagent = require('./superagent.js');
function parseSiSize(size) {
const units = {
"B": 1,
"KB": Math.pow(10, 3),
"MB": Math.pow(10, 6),
"GB": Math.pow(10, 9),
"TB": Math.pow(10, 12)
};
const match = size.toUpperCase().match(/^(\d+(\.\d+)?)\s*(\D+)$/);
if (!match) {
throw new Error("Invalid size format");
}
const number = parseFloat(match[1]);
const unit = match[3];
return Math.floor(number * units[unit]);
}
function parseBinarySize(size) {
const units = {
"B": 1,
"KIB": Math.pow(2, 10),
"MIB": Math.pow(2, 20),
"GIB": Math.pow(2, 30),
"TIB": Math.pow(2, 40)
};
const match = size.toUpperCase().match(/^(\d+(\.\d+)?)\s*(\D+)$/);
if (!match) {
throw new Error("Invalid size format");
}
const number = parseFloat(match[1]);
const unit = match[3];
return Math.floor(number * units[unit]);
}
async function getDockerMetrics() {
const lines = execSync('docker stats --format "{{ json . }}" --no-stream --no-trunc', { encoding: 'utf8' }).trim().split('\n');
const metrics = {};
for (const line of lines) {
const stat = JSON.parse(line);
const name = stat.Name; // appid or addon name
const networkData = stat.NetIO.split('/');
const networkRead = parseSiSize(networkData[0].trim());
const networkWrite = parseSiSize(networkData[1].trim());
const blockData = stat.BlockIO.split('/');
const blockRead = parseSiSize(blockData[0].trim());
const blockWrite = parseSiSize(blockData[1].trim());
const memUsageData = stat.MemUsage.split('/');
const memUsed = parseBinarySize(memUsageData[0].trim());
const memMax = parseBinarySize(memUsageData[1].trim());
const cpuPercData = stat.CPUPerc.trim().replace('%', '');
const cpuPerc = parseFloat(cpuPercData);
metrics[name] = { networkRead, networkWrite, blockRead, blockWrite, memUsed, memMax, cpuPerc };
}
return metrics;
}
async function getMemoryMetrics() {
// we can also read /proc/meminfo but complicated to match the 'used' output of free
const output = execSync('free --bytes --wide', { encoding: 'utf8' }).trim(); // --line is not in older ubuntu
const memoryRe = /Mem:\s+(?<total>\d+)\s+(?<used>\d+)\s+(?<free>\d+)\s+(?<shared>\d+)\s+(?<buffers>\d+)\s+(?<cache>\d+)\s+(?<available>\d+)/;
const memory = output.match(memoryRe);
if (!memory) throw new BoxError(BoxError.EXTERNAL_ERROR, 'Could not find memory used');
return { used: memory.groups.used };
}
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.used });
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-${name}.cpu-perc`, value: value.cpuPerc },
);
}
return new Promise((resolve) => {
const client = new net.Socket();
client.connect(constants.GRAPHITE_PORT, '127.0.0.1', () => {
debug('connected to graphite');
const now = Math.floor(Date.now() / 1000);
for (const metric of graphiteMetrics) {
client.write(`${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://${ip}:8000/graphite-web/render`;
}
async function getContainers(name, options) {
assert.strictEqual(typeof name, 'string');
assert.strictEqual(typeof options, 'object');
const { fromSecs, intervalSecs, noNullPoints } = options;
const graphiteUrl = await getGraphiteUrl();
const targets = [
`summarize(cloudron.container-${name}.cpu-perc, "${intervalSecs}s", "avg")`,
`summarize(cloudron.container-${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")`
];
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]
};
}
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 getContainers(app.id, options);
}
const serviceStats = {};
for (const serviceId of await services.listServices()) {
serviceStats[serviceId] = await getContainers(serviceId, options);
}
return {
cpu: systemStats.cpu,
memory: systemStats.memory,
apps: appStats,
services: serviceStats,
cpuCount: os.cpus().length
};
}
async function getSystemStream() {
const INTERVAL_SECS = 2;
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_SECS : null;
oldCpuMetrics = cpuMetrics;
const now = Date.now();
metricsStream.push(JSON.stringify({
cpu: [ cpuPercent, now ],
memory: [ memoryMetrics.used, now ]
}));
}, INTERVAL_SECS*1000);
return metricsStream;
}