zfsmon/main.go

package main

import (
	"flag"
	"fmt"
	"io/ioutil"
	"log"
	"os"
	"os/exec"
	"path/filepath"
	"strings"
	"time"

	"code.cloudfoundry.org/bytefmt"
	zfs "github.com/bicomsystems/go-libzfs"
	"github.com/davidscholberg/go-durationfmt"
	"gopkg.in/yaml.v2"
)

const LAST_SCAN_FILE = "last_scan.yaml"
const LOG_FILE = "zfsmon.log"
const POOL_SPACE_WARN = 90
const DATE_FORMAT = "02.01.2006"
const TIME_FORMAT = "02.01.2006 15:04:05"

type ScanStats struct {
	Func         string
	State        uint64
	Errors       uint64
	Duration     uint64
	LastFinished uint64
}

type DeviceState struct {
	State          string
	ReadErrors     uint64 `yaml:"read_errors"`
	WriteErrors    uint64 `yaml:"write_errors"`
	ChecksumErrors uint64 `yaml:"checksum_errors"`
}

type PoolState struct {
	Devices    map[string]DeviceState
	Scan       *ScanStats
	TotalSpace uint64  `yaml:"total_space"`
	FreeSpace  uint64  `yaml:"free_space"`
	FillLevel  float64 `yaml:"fill_level"`
}

func getStateOfDevices(tree zfs.VDevTree, data map[string]DeviceState) {
	data[tree.Name] = DeviceState{
		State:          tree.Stat.State.String(),
		ReadErrors:     tree.Stat.ReadErrors,
		WriteErrors:    tree.Stat.WriteErrors,
		ChecksumErrors: tree.Stat.ChecksumErrors,
	}

	if tree.Devices != nil {
		for _, subtree := range tree.Devices {
			getStateOfDevices(subtree, data)
		}
	}
}

func getScanFunc(scanFunc uint64) string {
	switch scanFunc {
	case zfs.PoolScanScrub:
		return "Scrub"
	case zfs.PoolScanResilver:
		return "Resilver"
	default:
		return "None"
	}
}

func getScanStats(scanStat zfs.PoolScanStat) *ScanStats {
	return &ScanStats{
		Func:         getScanFunc(scanStat.Func),
		State:        scanStat.State,
		Errors:       scanStat.Errors,
		Duration:     scanStat.EndTime - scanStat.StartTime,
		LastFinished: scanStat.EndTime,
	}
}

func getPoolState(poolName string) (*PoolState, error) {
	pool, err := zfs.PoolOpen(poolName)
	if err != nil {
		return nil, err
	}

	tree, err := pool.VDevTree()
	if err != nil {
		return nil, err
	}

	deviceData := map[string]DeviceState{}
	getStateOfDevices(tree, deviceData)

	scanStats := getScanStats(tree.ScanStat)
	level := float64(tree.Stat.Alloc) / float64(tree.Stat.Space) * 100.0

	return &PoolState{
		Devices:    deviceData,
		Scan:       scanStats,
		TotalSpace: tree.Stat.Space,
		FreeSpace:  tree.Stat.Space - tree.Stat.Alloc,
		FillLevel:  level,
	}, nil
}

func (s *PoolState) numDeviceErrors() uint64 {
	var n uint64

	for _, dev := range s.Devices {
		n += (dev.ReadErrors + dev.WriteErrors + dev.ChecksumErrors)

		if dev.State != "ONLINE" {
			n++
		}
	}
	return n
}

func readLastScanFile(lastScanFile string) *PoolState {
	historyFile, err := os.Open(lastScanFile)
	if err != nil {
		return nil
	}
	defer historyFile.Close()

	byteValue, _ := ioutil.ReadAll(historyFile)
	var poolState *PoolState
	_ = yaml.Unmarshal(byteValue, &poolState)

	return poolState
}

func (s *PoolState) writeLastScanFile(lastScanFile string) {
	data, _ := yaml.Marshal(s)
	data = append([]byte("# Scan history from "+time.Now().Format(DATE_FORMAT)+"\n\n"), data...)

	err := ioutil.WriteFile(lastScanFile, data, 0644)
	if err != nil {
		log.Println(err)
	}
}

func checkPool(poolName string, lastScanFile string, spaceWarn float64) []string {
	poolState, err := getPoolState(poolName)
	if err != nil {
		msg := fmt.Sprintf("Could not open %s. Error: %s", poolName, err.Error())
		log.Println(msg)
		return []string{msg}
	}

	lastState := readLastScanFile(lastScanFile)
	msgs := []string{}

	if lastState != nil {
		// Compare scan data to last state
		if poolState.numDeviceErrors() > lastState.numDeviceErrors() {
			log.Println("Device errors occurred")

			overviewBts, _ := yaml.Marshal(poolState.Devices)
			msg := fmt.Sprintf(`Oh no! Looks like %s has encountered device errors.
Here is an overview of all devices and their statuses:

%s`, poolName, string(overviewBts))

			msgs = append(msgs, msg)
		}

		// Check if new scan has completed
		if poolState.Scan.LastFinished > lastState.Scan.LastFinished {
			timeStr := time.Unix(int64(poolState.Scan.LastFinished), 0).Format(TIME_FORMAT)
			durationStr, _ := durationfmt.Format(time.Second*time.Duration(poolState.Scan.Duration), "%0h:%0m:%0s")

			log.Printf("%s on %s has finished at %s, took %s, %d errors",
				poolState.Scan.Func, poolName, timeStr, durationStr, poolState.Scan.Errors)

			msg := fmt.Sprintf("%s on %s has finished at %s\nDuration: %s\nErrors: %d",
				poolState.Scan.Func, poolName, timeStr, durationStr, poolState.Scan.Errors)
			msgs = append(msgs, msg)
		}

		// Check if fill level has reached the threshold
		if lastState.FillLevel <= spaceWarn && poolState.FillLevel > spaceWarn {
			byteStr := bytefmt.ByteSize(lastState.FreeSpace) + "B"
			msg := fmt.Sprintf("Disk space warning for %s - only %s left", poolName, byteStr)

			log.Println(msg)
			msgs = append(msgs, msg)
		}
	} else {
		log.Println("This is your first scan. Creating last_scan file.")
	}

	poolState.writeLastScanFile(lastScanFile)
	return msgs
}

func sendMsgs(sendmail string, msgUri string, msgs []string) {
	hostname, _ := os.Hostname()
	if hostname == "" {
		hostname = "unknown"
	}

	mailHeader := fmt.Sprintf(`Subject:[%s] zfsmon status update
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
`, hostname)
	for _, msg := range msgs {
		cmd := exec.Command(sendmail, msgUri)
		cmd.Stdin = strings.NewReader(mailHeader + msg)

		err := cmd.Run()
		if err != nil {
			log.Println(err)
		} else {
			log.Printf("sent mail to %s", msgUri)
		}
	}
}

func main() {
	workDir := ""
	if os.Getuid() == 0 {
		workDir = "/etc/zfsmon"
	}

	// Cmdline args
	poolName := flag.String("pool", "tank", "ZFS pool name")
	sendmail := flag.String("sendmail", "sendmail", "Sendmail path")
	mailAddress := flag.String("m", "", "E-Mail address")
	lastScanFile := flag.String("f", filepath.Join(workDir, LAST_SCAN_FILE), "File to store last scan results")
	logFile := flag.String("log", filepath.Join(workDir, LOG_FILE), "Log file")
	spaceWarn := flag.Int("space", POOL_SPACE_WARN, "Pool fill level in percent to warn at")
	flag.Parse()

	// Log file
	if *logFile != "" {
		f, err := os.OpenFile(*logFile, os.O_RDWR|os.O_CREATE|os.O_APPEND, 0666)
		if err != nil {
			log.Fatalf("error opening logfile: %v", err)
		}
		defer f.Close()

		log.SetOutput(f)
	}

	msgs := checkPool(*poolName, *lastScanFile, float64(*spaceWarn))
	if *mailAddress != "" {
		sendMsgs(*sendmail, *mailAddress, msgs)
	} else {
		log.Println("mail address not set, messaging disabled")
	}
}