Merge pull request #600 from NebulousLabs/healthcheck-cron

Healthcheck cron

docker-compose.yml

@@ -156,7 +156,8 @@ services:
         ipv4_address: 10.10.10.60
     environment:
       - HOSTNAME=0.0.0.0
-      - PORTAL_URL=nginx
+      - PORTAL_URL=http://nginx
+      - STATE_DIR=/usr/app/state
     expose:
       - 3100
     depends_on:

packages/handshake-api/package.json

@@ -10,6 +10,6 @@
     "punycode": "^2.1.1"
   },
   "devDependencies": {
-    "prettier": "^2.0.5"
+    "prettier": "^2.2.1"
   }
 }

packages/health-check/Dockerfile

@@ -7,6 +7,9 @@ RUN yarn --no-lockfile
 COPY src src
 COPY cli cli
 
+RUN echo '*/5 * * * * /usr/app/cli/run critical' >> /etc/crontabs/root
+RUN echo '0 * * * * /usr/app/cli/run verbose' >> /etc/crontabs/root
+
 EXPOSE 3100
 ENV NODE_ENV production
-CMD [ "node", "--max-http-header-size=64000", "src/index.js" ]
+CMD [ "sh", "-c", "crond ; node --max-http-header-size=64000 src/index.js" ]

packages/health-check/cli/run

@@ -0,0 +1,5 @@
+#!/usr/bin/env node
+
+process.env.NODE_ENV = process.env.NODE_ENV || "production";
+
+require("../src/run.js");

packages/health-check/package.json

@@ -5,17 +5,17 @@
   "license": "MIT",
   "dependencies": {
     "body-parser": "^1.19.0",
-    "cron": "^1.8.2",
     "deep-object-diff": "^1.1.0",
     "express": "^4.17.1",
     "http-status-codes": "^2.1.2",
     "lodash": "^4.17.20",
     "lowdb": "^1.0.0",
-    "ms": "^2.1.2",
-    "object-hash": "^2.0.3",
-    "superagent": "^6.0.0"
+    "object-hash": "^2.1.1",
+    "superagent": "^6.0.0",
+    "tmp": "^0.2.1",
+    "yargs": "^16.2.0"
   },
   "devDependencies": {
-    "prettier": "^2.0.5"
+    "prettier": "^2.2.1"
   }
 }

packages/health-check/src/checks/critical.js

@@ -1,15 +1,22 @@
+const fs = require("fs");
 const superagent = require("superagent");
+const tmp = require('tmp');
 const { StatusCodes } = require("http-status-codes");
 const { calculateElapsedTime, getResponseContent } = require("../utils");
 
 // uploadCheck returns the result of uploading a sample file
 async function uploadCheck(done) {
   const time = process.hrtime();
+  const file = tmp.fileSync();
+  
+  fs.writeSync(file.fd, Buffer.from(new Date())); // write current date to temp file
 
   superagent
-    .post(`http://${process.env.PORTAL_URL}/skynet/skyfile`)
-    .attach("file", "package.json", "package.json")
+    .post(`${process.env.PORTAL_URL}/skynet/skyfile`)
+    .attach("file", file.name, file.name)
     .end((error, response) => {
+      file.removeCallback();
+
       const statusCode = (response && response.statusCode) || (error && error.statusCode) || null;
 
       done({
@@ -29,7 +36,7 @@ async function downloadCheck(done) {
   let statusCode, errorResponseContent;
 
   try {
-    const response = await superagent.get(`http://${process.env.PORTAL_URL}/${skylink}?nocache=true`);
+    const response = await superagent.get(`${process.env.PORTAL_URL}/${skylink}?nocache=true`);
 
     statusCode = response.statusCode;
   } catch (error) {
@@ -46,4 +53,4 @@ async function downloadCheck(done) {
   });
 }
 
-module.exports.criticalChecks = [uploadCheck, downloadCheck];
+module.exports = [uploadCheck, downloadCheck];

packages/health-check/src/checks/verbose.js

@@ -1037,7 +1037,7 @@ function skylinkVerification(done, { name, skylink, bodyHash, headers }) {
   const time = process.hrtime();
 
   // Create the query for the skylink
-  const query = `http://${process.env.PORTAL_URL}/${skylink}?nocache=true`;
+  const query = `${process.env.PORTAL_URL}/${skylink}?nocache=true`;
 
   // Get the Skylink
   superagent
@@ -1087,7 +1087,7 @@ function skylinkVerification(done, { name, skylink, bodyHash, headers }) {
     );
 }
 
-module.exports.verboseChecks = [
+module.exports = [
   audioExampleCheck,
   covid19PaperCheck,
   covid19CoroNopePaperCheck,

packages/health-check/src/db.js

@@ -1,11 +1,10 @@
 const fs = require("fs");
 const low = require("lowdb");
 const FileSync = require("lowdb/adapters/FileSync");
-const Memory = require("lowdb/adapters/Memory");
 
-if (!fs.existsSync("state")) fs.mkdirSync("state");
+if (!fs.existsSync(process.env.STATE_DIR)) fs.mkdirSync(process.env.STATE_DIR);
 
-const adapter = new FileSync("state/state.json");
+const adapter = new FileSync(`${process.env.STATE_DIR}/state.json`);
 const db = low(adapter);
 
 db.defaults({ disabled: false, critical: [], verbose: [] }).write();

packages/health-check/src/index.js

@@ -8,8 +8,6 @@ const express = require("express");
 const bodyparser = require("body-parser");
 const db = require("./db");
 
-require("./schedule");
-
 const host = process.env.HOSTNAME || "0.0.0.0";
 const port = Number(process.env.PORT) || 3100;
 

packages/health-check/src/run.js

@@ -0,0 +1,33 @@
+require('yargs/yargs')(process.argv.slice(2))
+  .command('$0 <type>', 'Skynet portal health checks', (yargs) => {
+    yargs.positional('type', {
+        describe: 'Type of checks to run',
+        type: 'string',
+        choices: ["critical", "verbose"]
+    })  
+    .option("portal-url", {
+        describe: "Skynet portal url",
+        default: process.env.PORTAL_URL || "https://siasky.net",
+        type: "string",
+    })
+    .option("state-dir", {
+        describe: "State directory",
+        default: process.env.STATE_DIR || "state",
+        type: "string",
+    })
+  }, async ({ type, portalUrl, stateDir }) => {
+    process.env.PORTAL_URL = portalUrl;
+    process.env.STATE_DIR = stateDir;
+
+    const db = require("../src/db");
+    const checks = require(`../src/checks/${type}`);
+
+    const entry = {
+        date: new Date().toISOString(),
+        checks: await Promise.all(checks.map((check) => new Promise(check))),
+    };
+
+    // read before writing to make sure no external changes are overwritten
+    db.read().get(type).push(entry).write();
+  })
+  .argv

packages/health-check/src/schedule.js

@@ -1,45 +0,0 @@
-const { CronJob } = require("cron");
-const ms = require("ms");
-const db = require("./db");
-const { criticalChecks } = require("./checks/critical");
-const { verboseChecks } = require("./checks/verbose");
-
-// use this timezone to run all cron instances at the same time regardless of the server location
-const timezone = "UTC";
-
-// critical health checks job definition
-const criticalJobSchedule = "*/5 * * * *"; // on every 5 minute mark
-const criticalJobOnTick = async () => {
-  const entry = {
-    date: new Date().toISOString(),
-    checks: await Promise.all(criticalChecks.map((check) => new Promise(check))),
-  };
-
-  // read before writing to make sure no external changes are overwritten
-  db.read().get("critical").push(entry).write();
-};
-const criticalJob = new CronJob(criticalJobSchedule, criticalJobOnTick, null, false, timezone);
-
-// verbose health checks job definition
-const verboseJobSchedule = "0 * * * *"; // on every full hour mark
-const verboseJobOnTick = async () => {
-  const entry = {
-    date: new Date().toISOString(),
-    checks: await Promise.all(verboseChecks.map((check) => new Promise(check))),
-  };
-
-  // read before writing to make sure no external changes are overwritten
-  db.read().get("verbose").push(entry).write();
-};
-const verboseJob = new CronJob(verboseJobSchedule, verboseJobOnTick, null, false, timezone);
-
-// fire all health checks on startup (with delay for other services to boot)
-setTimeout(() => {
-  // fire first run manually
-  criticalJob.fireOnTick();
-  verboseJob.fireOnTick();
-
-  // start cron schedule
-  criticalJob.start();
-  verboseJob.start();
-}, ms("1 minute"));