From 9a32299fb160a4b33731071f0f5805be6a34f9ae Mon Sep 17 00:00:00 2001
From: Knut Ahlers <knut@ahlers.me>
Date: Fri, 3 Jun 2016 12:55:19 +0200
Subject: [PATCH] Initital version

---
 .gitignore              |   1 +
 elb-instance-status.yml |  16 +++++
 main.go                 | 150 ++++++++++++++++++++++++++++++++++++++++
 3 files changed, 167 insertions(+)
 create mode 100644 .gitignore
 create mode 100644 elb-instance-status.yml
 create mode 100644 main.go

diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..5842232
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1 @@
+elb-instance-status
diff --git a/elb-instance-status.yml b/elb-instance-status.yml
new file mode 100644
index 0000000..a6c18b4
--- /dev/null
+++ b/elb-instance-status.yml
@@ -0,0 +1,16 @@
+---
+
+- name: Ensure there are at least 30% free inodes on /
+  command: test $(df -i | grep "/$" | xargs | cut -d ' ' -f 5 | sed "s/%//") -lt 70
+
+- name: Ensure there are at least 30% free inodes on /var/lib/docker
+  command: test $(df -i | grep "/var/lib/docker$" | xargs | cut -d ' ' -f 5 | sed "s/%//") -lt 70
+
+- name: Ensure there is at least 30% free disk space on /var/lib/docker
+  command: test $(df | grep "/var/lib/docker$" | xargs | cut -d ' ' -f 5 | sed "s/%//") -lt 70
+
+- name: Ensure volume on /var/lib/docker is mounted
+  command: mount | grep -q /var/lib/docker
+
+- name: Ensure docker can start a small container
+  command: docker run --rm alpine /bin/sh -c "echo testing123" | grep -q testing123
\ No newline at end of file
diff --git a/main.go b/main.go
new file mode 100644
index 0000000..3810e24
--- /dev/null
+++ b/main.go
@@ -0,0 +1,150 @@
+package main
+
+import (
+	"bytes"
+	"fmt"
+	"io"
+	"io/ioutil"
+	"log"
+	"net/http"
+	"os"
+	"os/exec"
+	"strconv"
+	"sync"
+	"time"
+
+	"gopkg.in/yaml.v2"
+
+	"github.com/Luzifer/rconfig"
+	"github.com/gorilla/mux"
+	"github.com/robfig/cron"
+)
+
+var (
+	cfg = struct {
+		CheckDefinitionsFile string `flag:"check-definitions-file,c" default:"/etc/elb-instance-status.yml" description:"File containing checks to perform for instance health"`
+		UnhealthyThreshold   int64  `flag:"unhealthy-threshold" default:"5" description:"How often does a check have to fail to mark the machine unhealthy"`
+		Listen               string `flag:"listen" default:":3000" description:"IP/Port to listen on for ELB health checks"`
+		VersionAndExit       bool   `flag:"version" default:"false" description:"Print version and exit"`
+	}{}
+
+	version = "dev"
+
+	checks               = []checkCommand{}
+	checkResults         = map[string]*checkResult{}
+	checkResultsLock     sync.RWMutex
+	lastResultRegistered time.Time
+)
+
+type checkCommand struct {
+	Name     string `yaml:"name"`
+	Command  string `yaml:"command"`
+	WarnOnly bool   `yaml:"warn-only"`
+}
+
+type checkResult struct {
+	Check     checkCommand
+	IsSuccess bool
+	Streak    int64
+}
+
+func init() {
+	rconfig.Parse(&cfg)
+
+	if cfg.VersionAndExit {
+		fmt.Printf("elb-instance-status %s\n", version)
+		os.Exit(0)
+	}
+}
+
+func loadChecks() error {
+	rawChecks, err := ioutil.ReadFile(cfg.CheckDefinitionsFile)
+	if err != nil {
+		return err
+	}
+	return yaml.Unmarshal(rawChecks, &checks)
+}
+
+func main() {
+	if err := loadChecks(); err != nil {
+		log.Fatalf("Unable to read definitions file: %s", err)
+	}
+
+	c := cron.New()
+	c.AddFunc("@every 1m", spawnChecks)
+	c.Start()
+
+	spawnChecks()
+
+	r := mux.NewRouter()
+	r.HandleFunc("/status", handleELBHealthCheck)
+	http.ListenAndServe(cfg.Listen, r)
+}
+
+func spawnChecks() {
+	for i := range checks {
+		go executeAndRegisterCheck(i)
+	}
+}
+
+func executeAndRegisterCheck(checkIndex int) {
+	check := checks[checkIndex]
+
+	cmd := exec.Command("/bin/bash", "-c", check.Command)
+	err := cmd.Run()
+
+	success := err == nil
+
+	checkResultsLock.Lock()
+
+	if _, ok := checkResults[check.Name]; !ok {
+		checkResults[check.Name] = &checkResult{
+			Check: check,
+		}
+	}
+
+	if success == checkResults[check.Name].IsSuccess {
+		checkResults[check.Name].Streak++
+	} else {
+		checkResults[check.Name].IsSuccess = success
+		checkResults[check.Name].Streak = 1
+	}
+
+	lastResultRegistered = time.Now()
+
+	checkResultsLock.Unlock()
+}
+
+func handleELBHealthCheck(res http.ResponseWriter, r *http.Request) {
+	healthy := true
+	start := time.Now()
+	buf := bytes.NewBuffer([]byte{})
+
+	checkResultsLock.RLock()
+	for cn, cr := range checkResults {
+		state := ""
+		switch {
+		case cr.IsSuccess:
+			state = "PASS"
+		case !cr.IsSuccess && cr.Check.WarnOnly:
+			state = "WARN"
+		case !cr.IsSuccess && !cr.Check.WarnOnly && cr.Streak < cfg.UnhealthyThreshold:
+			state = "CRIT"
+		case !cr.IsSuccess && !cr.Check.WarnOnly && cr.Streak >= cfg.UnhealthyThreshold:
+			state = "CRIT"
+			healthy = false
+		}
+		fmt.Fprintf(buf, "[%s] %s\n", state, cn)
+	}
+	checkResultsLock.RUnlock()
+
+	res.Header().Set("X-Collection-Parsed-In", strconv.FormatInt(time.Since(start).Nanoseconds()/int64(time.Microsecond), 10)+"ms")
+	res.Header().Set("X-Last-Result-Registered-At", lastResultRegistered.Format(time.RFC1123))
+	if healthy {
+		res.WriteHeader(http.StatusOK)
+	} else {
+		res.WriteHeader(http.StatusInternalServerError)
+	}
+
+	io.Copy(res, buf)
+}