bennetgallein/cachet-monitor
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

huge refactor

Browse Source 
- extendable backends
- better project structure
- better cli interface
This commit is contained in:
Matej Kramny
2019-02-20 11:14:45 +08:00
parent df31238a1f
commit 162d55b3f3
17 changed files with 946 additions and 705 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
.gitignore vendored
View File

@@ -1,3 +1,5 @@
/config.yml
/config.json
examples/
vendor/
cachet-monitor

85
api.go
View File

@@ -1,85 +0,0 @@
package cachet
import (
"bytes"
"crypto/tls"
"encoding/json"
"errors"
"net/http"
"strconv"
"time"
"github.com/Sirupsen/logrus"
)
type CachetAPI struct {
URL string `json:"url"`
Token string `json:"token"`
Insecure bool `json:"insecure"`
}
type CachetResponse struct {
Data json.RawMessage `json:"data"`
}
// TODO: test
func (api CachetAPI) Ping() error {
resp, _, err := api.NewRequest("GET", "/ping", nil)
if err != nil {
return err
}
if resp.StatusCode != 200 {
return errors.New("API Responded with non-200 status code")
}
defer resp.Body.Close()
return nil
}
// SendMetric adds a data point to a cachet monitor
func (api CachetAPI) SendMetric(id int, lag int64) {
logrus.Debugf("Sending lag metric ID:%d RTT %vms", id, lag)
jsonBytes, _ := json.Marshal(map[string]interface{}{
"value": lag,
"timestamp": time.Now().Unix(),
})
resp, _, err := api.NewRequest("POST", "/metrics/"+strconv.Itoa(id)+"/points", jsonBytes)
if err != nil || resp.StatusCode != 200 {
logrus.Warnf("Could not log metric! ID: %d, err: %v", id, err)
}
defer resp.Body.Close()
}
// TODO: test
// NewRequest wraps http.NewRequest
func (api CachetAPI) NewRequest(requestType, url string, reqBody []byte) (*http.Response, CachetResponse, error) {
req, err := http.NewRequest(requestType, api.URL+url, bytes.NewBuffer(reqBody))
req.Header.Set("Content-Type", "application/json")
req.Header.Set("X-Cachet-Token", api.Token)
transport := http.DefaultTransport.(*http.Transport)
transport.TLSClientConfig = &tls.Config{InsecureSkipVerify: api.Insecure}
client := &http.Client{
Transport: transport,
}
res, err := client.Do(req)
if err != nil {
return nil, CachetResponse{}, err
}
var body struct {
Data json.RawMessage `json:"data"`
}
err = json.NewDecoder(res.Body).Decode(&body)
defer req.Body.Close()
return res, body, err
}

288
backends/cachet/backend.go Normal file
View File

@@ -0,0 +1,288 @@
package cachetbackend
import (
"bytes"
"crypto/tls"
"encoding/json"
"errors"
"fmt"
"net/http"
"strconv"
"strings"
"time"
"github.com/castawaylabs/cachet-monitor/monitors"
"github.com/sirupsen/logrus"
)
const DefaultTimeFormat = "15:04:05 Jan 2 MST"
type CachetBackend struct {
URL string `json:"url" yaml:"url"`
Token string `json:"token" yaml:"token"`
Insecure bool `json:"insecure" yaml:"insecure"`
DateFormat string `json:"date_format" yaml:"date_format"`
}
type CachetResponse struct {
Data json.RawMessage `json:"data"`
}
func (api CachetBackend) ValidateMonitor(mon *monitors.AbstractMonitor) []string {
errs := []string{}
params := mon.Params
componentID, componentIDOk := params["component_id"]
metricID, metricIDOk := params["metric_id"]
if !componentIDOk && !metricIDOk {
errs = append(errs, "component_id and metric_id is unset")
}
if _, ok := componentID.(int); !ok && componentIDOk {
errs = append(errs, "component_id not integer")
}
if _, ok := metricID.(int); !ok && metricIDOk {
errs = append(errs, "metric_id not integer")
}
return errs
}
func (api CachetBackend) Validate() []string {
errs := []string{}
if len(api.URL) == 0 {
errs = append(errs, "Cachet API URL invalid")
}
if len(api.Token) == 0 {
errs = append(errs, "Cachet API Token invalid")
}
if len(api.DateFormat) == 0 {
api.DateFormat = DefaultTimeFormat
}
return errs
}
// TODO: test
func (api CachetBackend) Ping() error {
resp, _, err := api.NewRequest("GET", "/ping", nil)
if err != nil {
return err
}
if resp.StatusCode != 200 {
return errors.New("API Responded with non-200 status code")
}
defer resp.Body.Close()
return nil
}
// TODO: test
// NewRequest wraps http.NewRequest
func (api CachetBackend) NewRequest(requestType, url string, reqBody []byte) (*http.Response, interface{}, error) {
req, err := http.NewRequest(requestType, api.URL+url, bytes.NewBuffer(reqBody))
req.Header.Set("Content-Type", "application/json")
req.Header.Set("X-Cachet-Token", api.Token)
transport := http.DefaultTransport.(*http.Transport)
transport.TLSClientConfig = &tls.Config{InsecureSkipVerify: api.Insecure}
client := &http.Client{
Transport: transport,
}
res, err := client.Do(req)
if err != nil {
return nil, CachetResponse{}, err
}
defer res.Body.Close()
defer req.Body.Close()
var body struct {
Data json.RawMessage `json:"data"`
}
err = json.NewDecoder(res.Body).Decode(&body)
return res, body, err
}
func (mon CachetBackend) Describe() []string {
features := []string{"Cachet API"}
return features
}
func (api CachetBackend) SendMetric(monitor monitors.MonitorInterface, lag int64) error {
mon := monitor.GetMonitor()
if _, ok := mon.Params["metric_id"]; !ok {
return nil
}
metricID := mon.Params["metric_id"].(int)
// report lag
logrus.Debugf("Sending lag metric ID: %d RTT %vms", metricID, lag)
jsonBytes, _ := json.Marshal(map[string]interface{}{
"value": lag,
"timestamp": time.Now().Unix(),
})
resp, _, err := api.NewRequest("POST", "/metrics/"+strconv.Itoa(metricID)+"/points", jsonBytes)
if err != nil || resp.StatusCode != 200 {
logrus.Warnf("Could not log metric! ID: %d, err: %v", metricID, err)
}
if resp != nil && resp.Body != nil {
defer resp.Body.Close()
}
return nil
}
func (api CachetBackend) UpdateMonitor(mon monitors.MonitorInterface, status, previousStatus monitors.MonitorStatus, errs []error) error {
monitor := mon.GetMonitor()
l := logrus.WithFields(logrus.Fields{
"monitor": monitor.Name,
"time": time.Now().Format(api.DateFormat),
})
errors := make([]string, len(errs))
for i, err := range errs {
errors[i] = err.Error()
}
fmt.Println("errs", errs)
componentID := monitor.Params["component_id"].(int)
incident, err := api.findIncident(componentID)
if err != nil {
l.Errorf("Couldn't find existing incidents: %v", err)
}
if incident == nil {
// create a new one
incident = &Incident{
Name: "",
ComponentID: componentID,
Message: "",
Notify: true,
}
} else {
// find component status
component, err := api.getComponent(incident.ComponentID)
if err != nil {
panic(err)
}
incident.ComponentStatus = component.Status
}
tpls := monitor.Template
tplData := api.getTemplateData(monitor)
var tpl monitors.MessageTemplate
if status == monitors.MonitorStatusDown {
tpl = tpls.Investigating
tplData["FailReason"] = strings.Join(errors, "\n - ")
l.Warnf("updating component. Monitor is down: %v", tplData["FailReason"])
} else {
// was down, created an incident, its now ok, make it resolved.
tpl = tpls.Fixed
l.Warn("Resolving incident")
}
tplData["incident"] = incident
subject, message := tpl.Exec(tplData)
if incident.ID == 0 {
incident.Name = subject
incident.Message = message
} else {
incident.Message += "\n\n---\n\n" + subject + ":\n\n" + message
}
if status == monitors.MonitorStatusDown && (incident.ComponentStatus == 0 || incident.ComponentStatus > 2) {
incident.Status = 1
fmt.Println("incident status", incident.ComponentStatus)
if incident.ComponentStatus >= 3 {
// major outage
incident.ComponentStatus = 4
} else {
incident.ComponentStatus = 3
}
} else if status == monitors.MonitorStatusUp {
incident.Status = 4
incident.ComponentStatus = 1
}
incident.Notify = true
// create/update incident
if err := incident.Send(api); err != nil {
l.Errorf("Error sending incident: %v", err)
return err
}
return nil
}
func (api CachetBackend) Tick(monitor monitors.MonitorInterface, status monitors.MonitorStatus, errs []error, lag int64) {
mon := monitor.GetMonitor()
if mon.GetLastStatus() == status || status == monitors.MonitorStatusNotSaturated {
return
}
logrus.Infof("updating backend for monitor")
lastStatus := mon.UpdateLastStatus(status)
api.UpdateMonitor(monitor, status, lastStatus, errs)
if _, ok := mon.Params["metric_id"]; ok && lag > 0 {
api.SendMetric(monitor, lag)
}
}
func (api CachetBackend) getComponent(componentID int) (*Component, error) {
return nil, nil
}
func (api CachetBackend) findIncident(componentID int) (*Incident, error) {
// fetch watching, identified & investigating
statuses := []int{3, 2, 1}
for _, status := range statuses {
incidents, err := api.findIncidents(componentID, status)
if err != nil {
return nil, err
}
for _, incident := range incidents {
incident.Status = status
return incident, nil
}
}
return nil, nil
}
func (api CachetBackend) findIncidents(componentID int, status int) ([]*Incident, error) {
resp, body, err := api.NewRequest("GET", "/incidents?component_id="+strconv.Itoa(componentID)+"&status="+strconv.Itoa(status), nil)
if err != nil {
return nil, err
}
var data []*Incident
if err := json.Unmarshal(body.(CachetResponse).Data, &data); err != nil {
return nil, fmt.Errorf("Cannot find incidents: %v", err)
}
if resp.StatusCode != 200 {
return nil, fmt.Errorf("Could not fetch incidents! %v", err)
}
return data, nil
}

14
backends/cachet/component.go Normal file
View File

@@ -0,0 +1,14 @@
package cachetbackend
// Incident Cachet data model
type Component struct {
ID int `json:"id"`
Name string `json:"name"`
Message string `json:"message"`
Status int `json:"status"`
Visible int `json:"visible"`
Notify bool `json:"notify"`
ComponentID int `json:"component_id"`
ComponentStatus int `json:"component_status"`
}

71
backends/cachet/incident.go Normal file
View File

@@ -0,0 +1,71 @@
package cachetbackend
import (
"encoding/json"
"fmt"
"strconv"
"time"
"github.com/castawaylabs/cachet-monitor/backends"
"github.com/castawaylabs/cachet-monitor/monitors"
)
// "encoding/json"
// "fmt"
// "strconv"
// "github.com/sirupsen/logrus"
// Incident Cachet data model
type Incident struct {
ID int `json:"id"`
Name string `json:"name"`
Message string `json:"message"`
Status int `json:"status"`
Visible int `json:"visible"`
Notify bool `json:"notify"`
ComponentID int `json:"component_id"`
ComponentStatus int `json:"component_status"`
}
// Send - Create or Update incident
func (incident *Incident) Send(backend backends.BackendInterface) error {
requestURL := "/incidents"
requestMethod := "POST"
jsonBytes, _ := json.Marshal(incident)
if incident.ID > 0 {
// create an incident update
requestMethod = "PUT"
requestURL += "/" + strconv.Itoa(incident.ID)
}
resp, body, err := backend.NewRequest(requestMethod, requestURL, jsonBytes)
if err != nil {
return err
}
var data struct {
ID int `json:"id"`
}
if err := json.Unmarshal(body.(CachetResponse).Data, &data); err != nil {
return fmt.Errorf("Cannot parse incident body: %v, %v", err, string(body.(CachetResponse).Data))
}
incident.ID = data.ID
if resp.StatusCode != 200 {
return fmt.Errorf("Could not update/create incident!")
}
return nil
}
func (api *CachetBackend) getTemplateData(monitor *monitors.AbstractMonitor) map[string]interface{} {
return map[string]interface{}{
// "SystemName": monitor.config.SystemName,
"Monitor": monitor,
"now": time.Now().Format(api.DateFormat),
// "incident": monitor.incident,
}
}

19
backends/interface.go Normal file
View File

@@ -0,0 +1,19 @@
package backends
import (
"net/http"
"github.com/castawaylabs/cachet-monitor/monitors"
)
type BackendInterface interface {
Ping() error
Tick(monitor monitors.MonitorInterface, status monitors.MonitorStatus, errs []error, lag int64)
SendMetric(monitor monitors.MonitorInterface, lag int64) error
UpdateMonitor(monitor monitors.MonitorInterface, status, previousStatus monitors.MonitorStatus, errs []error) error
NewRequest(requestType, url string, reqBody []byte) (*http.Response, interface{}, error)
Describe() []string
Validate() []string
ValidateMonitor(monitor *monitors.AbstractMonitor) []string
}

195
cli/main.go
View File

@@ -1,198 +1,5 @@
package main
import (
"encoding/json"
"errors"
"io/ioutil"
"net/http"
"net/url"
"os"
"os/signal"
"strings"
"sync"
"github.com/Sirupsen/logrus"
cachet "github.com/castawaylabs/cachet-monitor"
docopt "github.com/docopt/docopt-go"
"github.com/mitchellh/mapstructure"
"gopkg.in/yaml.v2"
)
const usage = `cachet-monitor
Usage:
cachet-monitor (-c PATH | --config PATH) [--log=LOGPATH] [--name=NAME] [--immediate]
cachet-monitor -h | --help | --version
Arguments:
PATH path to config.json
LOGPATH path to log output (defaults to STDOUT)
NAME name of this logger
Examples:
cachet-monitor -c /root/cachet-monitor.json
cachet-monitor -c /root/cachet-monitor.json --log=/var/log/cachet-monitor.log --name="development machine"
Options:
-c PATH.json --config PATH Path to configuration file
-h --help Show this screen.
--version Show version
--immediate Tick immediately (by default waits for first defined interval)
Environment varaibles:
CACHET_API override API url from configuration
CACHET_TOKEN override API token from configuration
CACHET_DEV set to enable dev logging`
var version string
func main() {
arguments, _ := docopt.Parse(usage, nil, true, version, false)
cfg, err := getConfiguration(arguments["--config"].(string))
if err != nil {
logrus.Panicf("Unable to start (reading config): %v", err)
}
if immediate, ok := arguments["--immediate"]; ok {
cfg.Immediate = immediate.(bool)
}
if name := arguments["--name"]; name != nil {
cfg.SystemName = name.(string)
}
logrus.SetOutput(getLogger(arguments["--log"]))
if len(os.Getenv("CACHET_API")) > 0 {
cfg.API.URL = os.Getenv("CACHET_API")
}
if len(os.Getenv("CACHET_TOKEN")) > 0 {
cfg.API.Token = os.Getenv("CACHET_TOKEN")
}
if len(os.Getenv("CACHET_DEV")) > 0 {
logrus.SetLevel(logrus.DebugLevel)
}
if valid := cfg.Validate(); !valid {
logrus.Errorf("Invalid configuration")
os.Exit(1)
}
logrus.Debug("Configuration valid")
logrus.Infof("System: %s", cfg.SystemName)
logrus.Infof("API: %s", cfg.API.URL)
logrus.Infof("Monitors: %d\n", len(cfg.Monitors))
logrus.Infof("Pinging cachet")
if err := cfg.API.Ping(); err != nil {
logrus.Errorf("Cannot ping cachet!\n%v", err)
os.Exit(1)
}
logrus.Infof("Ping OK")
wg := &sync.WaitGroup{}
for index, monitor := range cfg.Monitors {
logrus.Infof("Starting Monitor #%d: ", index)
logrus.Infof("Features: \n - %v", strings.Join(monitor.Describe(), "\n - "))
go monitor.ClockStart(cfg, monitor, wg)
}
signals := make(chan os.Signal, 1)
signal.Notify(signals, os.Interrupt, os.Kill)
<-signals
logrus.Warnf("Abort: Waiting monitors to finish")
for _, mon := range cfg.Monitors {
mon.GetMonitor().ClockStop()
}
wg.Wait()
}
func getLogger(logPath interface{}) *os.File {
if logPath == nil || len(logPath.(string)) == 0 {
return os.Stdout
}
file, err := os.Create(logPath.(string))
if err != nil {
logrus.Errorf("Unable to open file '%v' for logging: \n%v", logPath, err)
os.Exit(1)
}
return file
}
func getConfiguration(path string) (*cachet.CachetMonitor, error) {
var cfg cachet.CachetMonitor
var data []byte
// test if its a url
url, err := url.ParseRequestURI(path)
if err == nil && len(url.Scheme) > 0 {
// download config
response, err := http.Get(path)
if err != nil {
logrus.Warn("Unable to download network configuration")
return nil, err
}
defer response.Body.Close()
data, _ = ioutil.ReadAll(response.Body)
logrus.Info("Downloaded network configuration.")
} else {
data, err = ioutil.ReadFile(path)
if err != nil {
return nil, errors.New("Unable to open file: '" + path + "'")
}
}
if strings.HasSuffix(path, ".yaml") || strings.HasSuffix(path, ".yml") {
err = yaml.Unmarshal(data, &cfg)
} else {
err = json.Unmarshal(data, &cfg)
}
if err != nil {
logrus.Warnf("Unable to parse configuration file")
}
cfg.Monitors = make([]cachet.MonitorInterface, len(cfg.RawMonitors))
for index, rawMonitor := range cfg.RawMonitors {
var t cachet.MonitorInterface
var err error
// get default type
monType := cachet.GetMonitorType("")
if t, ok := rawMonitor["type"].(string); ok {
monType = cachet.GetMonitorType(t)
}
switch monType {
case "http":
var s cachet.HTTPMonitor
err = mapstructure.Decode(rawMonitor, &s)
t = &s
case "dns":
var s cachet.DNSMonitor
err = mapstructure.Decode(rawMonitor, &s)
t = &s
default:
logrus.Errorf("Invalid monitor type (index: %d) %v", index, monType)
continue
}
t.GetMonitor().Type = monType
if err != nil {
logrus.Errorf("Unable to unmarshal monitor to type (index: %d): %v", index, err)
continue
}
cfg.Monitors[index] = t
}
return &cfg, err
Execute()
}

117
cli/root.go Normal file
View File

@@ -0,0 +1,117 @@
package main
import (
"os"
"os/signal"
"strings"
"sync"
cachet "github.com/castawaylabs/cachet-monitor"
"github.com/sirupsen/logrus"
"github.com/spf13/cobra"
)
var cfgFile string
// rootCmd represents the base command when called without any subcommands
var rootCmd = &cobra.Command{
Use: "cmd",
Short: "cachet-monitor",
// Uncomment the following line if your bare application
// has an action associated with it:
Run: func(cmd *cobra.Command, args []string) {
Action(cmd, args)
},
}
// Execute adds all child commands to the root command and sets flags appropriately.
// This is called by main.main(). It only needs to happen once to the rootCmd.
func Execute() {
if err := rootCmd.Execute(); err != nil {
panic(err)
}
}
func init() {
// Here you will define your flags and configuration settings.
// Cobra supports persistent flags, which, if defined here,
// will be global for your application.
pf := rootCmd.PersistentFlags()
pf.StringVarP(&cfgFile, "config", "c", "", "config file (default is $(pwd)/config.yml)")
pf.String("log", "", "log output")
pf.String("format", "text", "log format [text/json]")
pf.String("name", "", "machine name")
pf.Bool("immediate", false, "Tick immediately (by default waits for first defined")
}
func Action(cmd *cobra.Command, args []string) {
cfg, err := cachet.New(cfgFile)
if err != nil {
logrus.Panicf("Unable to start (reading config): %v", err)
}
if immediate, err := cmd.Flags().GetBool("immediate"); err == nil && immediate {
cfg.Immediate = immediate
}
if name, err := cmd.Flags().GetString("name"); err == nil && len(name) > 0 {
cfg.SystemName = name
}
logrus.SetOutput(getLogger(cmd))
if format, err := cmd.Flags().GetString("format"); err == nil && format == "json" {
logrus.SetFormatter(&logrus.JSONFormatter{})
}
if valid := cfg.Validate(); !valid {
logrus.Errorf("Invalid configuration")
os.Exit(1)
}
logrus.Debug("Configuration valid")
logrus.Infof("System: %s", cfg.SystemName)
// logrus.Infof("API: %s", cfg.API.URL)
logrus.Infof("Monitors: %d", len(cfg.Monitors))
logrus.Infof("Backend: %v", strings.Join(cfg.Backend.Describe(), "\n - "))
logrus.Infof("Pinging backend")
if err := cfg.Backend.Ping(); err != nil {
logrus.Errorf("Cannot ping backend!\n%v", err)
// os.Exit(1)
}
logrus.Infof("Ping OK")
logrus.Warnf("Starting!")
wg := &sync.WaitGroup{}
for index, monitor := range cfg.Monitors {
logrus.Infof("Starting Monitor #%d: ", index)
logrus.Infof("Features: \n - %v", strings.Join(monitor.Describe(), "\n - "))
go monitor.Start(monitor.GetTestFunc(), wg, cfg.Backend.Tick, cfg.Immediate)
}
signals := make(chan os.Signal, 1)
signal.Notify(signals, os.Interrupt, os.Kill)
<-signals
logrus.Warnf("Abort: Waiting for monitors to finish")
for _, mon := range cfg.Monitors {
mon.GetMonitor().Stop()
}
wg.Wait()
}
func getLogger(cmd *cobra.Command) *os.File {
logPath, _ := cmd.Flags().GetString("log")
if len(logPath) == 0 {
return os.Stdout
}
file, err := os.Create(logPath)
if err != nil {
logrus.Errorf("Unable to open file '%v' for logging: \n%v", logPath, err)
os.Exit(1)
}
return file
}

146
config.go
View File

@@ -1,22 +1,130 @@
package cachet
import (
"encoding/json"
"errors"
"io/ioutil"
"net"
"net/http"
"net/url"
"os"
"strings"
"time"
"github.com/Sirupsen/logrus"
"github.com/castawaylabs/cachet-monitor/backends"
cachetbackend "github.com/castawaylabs/cachet-monitor/backends/cachet"
"github.com/castawaylabs/cachet-monitor/monitors"
"github.com/mitchellh/mapstructure"
"github.com/sirupsen/logrus"
yaml "gopkg.in/yaml.v2"
)
type CachetMonitor struct {
SystemName string `json:"system_name" yaml:"system_name"`
DateFormat string `json:"date_format" yaml:"date_format"`
API CachetAPI `json:"api"`
RawMonitors []map[string]interface{} `json:"monitors" yaml:"monitors"`
RawBackend map[string]interface{} `json:"backend" yaml:"backend"`
Monitors []MonitorInterface `json:"-" yaml:"-"`
Immediate bool `json:"-" yaml:"-"`
SystemName string `json:"system_name" yaml:"system_name"`
Backend backends.BackendInterface `json:"-" yaml:"-"`
Monitors []monitors.MonitorInterface `json:"-" yaml:"-"`
Immediate bool `json:"-" yaml:"-"`
}
func New(path string) (*CachetMonitor, error) {
var cfg *CachetMonitor
var data []byte
// test if its a url
url, err := url.ParseRequestURI(path)
if err == nil && len(url.Scheme) > 0 {
// download config
response, err := http.Get(path)
if err != nil {
logrus.Warn("Unable to download network configuration")
return nil, err
}
defer response.Body.Close()
data, _ = ioutil.ReadAll(response.Body)
logrus.Info("Downloaded network configuration.")
} else {
data, err = ioutil.ReadFile(path)
if err != nil {
return nil, errors.New("Unable to open file: '" + path + "'")
}
}
if strings.HasSuffix(path, ".yaml") || strings.HasSuffix(path, ".yml") {
err = yaml.Unmarshal(data, &cfg)
} else {
err = json.Unmarshal(data, &cfg)
}
if err != nil {
logrus.Warnf("Unable to parse configuration file")
return nil, err
}
// get default type
if backend, ok := cfg.RawBackend["type"].(string); !ok {
err = errors.New("Cannot determine backend type")
} else {
switch backend {
case "cachet":
var backend cachetbackend.CachetBackend
err = mapstructure.Decode(cfg.RawBackend, &backend)
cfg.Backend = &backend
// backend.config = cfg
default:
err = errors.New("Invalid backend type: %v" + backend)
}
}
if errs := cfg.Backend.Validate(); len(errs) > 0 {
logrus.Errorf("Backend validation errors: %v", errs)
os.Exit(1)
}
if err != nil {
logrus.Errorf("Unable to unmarshal backend: %v", err)
return nil, err
}
cfg.Monitors = make([]monitors.MonitorInterface, len(cfg.RawMonitors))
for index, rawMonitor := range cfg.RawMonitors {
var t monitors.MonitorInterface
// get default type
monType := GetMonitorType("")
if t, ok := rawMonitor["type"].(string); ok {
monType = GetMonitorType(t)
}
switch monType {
case "http":
var mon monitors.HTTPMonitor
err = mapstructure.Decode(rawMonitor, &mon)
t = &mon
case "dns":
var mon monitors.DNSMonitor
err = mapstructure.Decode(rawMonitor, &mon)
t = &mon
default:
logrus.Errorf("Invalid monitor type (index: %d) %v", index, monType)
continue
}
if err != nil {
logrus.Errorf("Unable to unmarshal monitor to type (index: %d): %v", index, err)
continue
}
mon := t.GetMonitor()
mon.Type = monType
cfg.Monitors[index] = t
}
return cfg, err
}
// Validate configuration
@@ -28,22 +136,13 @@ func (cfg *CachetMonitor) Validate() bool {
cfg.SystemName = getHostname()
}
if len(cfg.DateFormat) == 0 {
cfg.DateFormat = DefaultTimeFormat
}
if len(cfg.API.Token) == 0 || len(cfg.API.URL) == 0 {
logrus.Warnf("API URL or API Token missing.\nGet help at https://github.com/castawaylabs/cachet-monitor")
valid = false
}
if len(cfg.Monitors) == 0 {
logrus.Warnf("No monitors defined!\nSee help for example configuration")
valid = false
}
for index, monitor := range cfg.Monitors {
if errs := monitor.Validate(); len(errs) > 0 {
if errs := monitor.Validate(cfg.Backend.ValidateMonitor); len(errs) > 0 {
logrus.Warnf("Monitor validation errors (index %d): %v", index, "\n - "+strings.Join(errs, "\n - "))
valid = false
}
@@ -67,10 +166,6 @@ func getHostname() string {
return addrs[0].String()
}
func getMs() int64 {
return time.Now().UnixNano() / int64(time.Millisecond)
}
func GetMonitorType(t string) string {
if len(t) == 0 {
return "http"
@@ -78,12 +173,3 @@ func GetMonitorType(t string) string {
return strings.ToLower(t)
}
func getTemplateData(monitor *AbstractMonitor) map[string]interface{} {
return map[string]interface{}{
"SystemName": monitor.config.SystemName,
"API": monitor.config.API,
"Monitor": monitor,
"now": time.Now().Format(monitor.config.DateFormat),
}
}

112
incident.go
View File

@@ -1,112 +0,0 @@
package cachet
import (
"encoding/json"
"fmt"
"strconv"
"github.com/Sirupsen/logrus"
)
// Incident Cachet data model
type Incident struct {
ID int `json:"id"`
Name string `json:"name"`
Message string `json:"message"`
Status int `json:"status"`
Visible int `json"visible"`
Notify bool `json:"notify"`
ComponentID int `json:"component_id"`
ComponentStatus int `json:"component_status"`
}
// Send - Create or Update incident
func (incident *Incident) Send(cfg *CachetMonitor) error {
switch incident.Status {
case 1, 2, 3:
// partial outage
incident.ComponentStatus = 3
componentStatus, err := incident.GetComponentStatus(cfg)
if componentStatus == 3 {
// major outage
incident.ComponentStatus = 4
}
if err != nil {
logrus.Warnf("cannot fetch component: %v", err)
}
case 4:
// fixed
incident.ComponentStatus = 1
}
requestType := "POST"
requestURL := "/incidents"
if incident.ID > 0 {
requestType = "PUT"
requestURL += "/" + strconv.Itoa(incident.ID)
}
jsonBytes, _ := json.Marshal(incident)
resp, body, err := cfg.API.NewRequest(requestType, requestURL, jsonBytes)
if err != nil {
return err
}
var data struct {
ID int `json:"id"`
}
if err := json.Unmarshal(body.Data, &data); err != nil {
return fmt.Errorf("Cannot parse incident body: %v, %v", err, string(body.Data))
}
incident.ID = data.ID
if resp.StatusCode != 200 {
return fmt.Errorf("Could not create/update incident!")
}
return nil
}
func (incident *Incident) GetComponentStatus(cfg *CachetMonitor) (int, error) {
resp, body, err := cfg.API.NewRequest("GET", "/components/"+strconv.Itoa(incident.ComponentID), nil)
if err != nil {
return 0, err
}
if resp.StatusCode != 200 {
return 0, fmt.Errorf("Invalid status code. Received %d", resp.StatusCode)
}
var data struct {
Status int `json:"status,string"`
}
if err := json.Unmarshal(body.Data, &data); err != nil {
return 0, fmt.Errorf("Cannot parse component body: %v. Err = %v", string(body.Data), err)
}
return data.Status, nil
}
// SetInvestigating sets status to Investigating
func (incident *Incident) SetInvestigating() {
incident.Status = 1
}
// SetIdentified sets status to Identified
func (incident *Incident) SetIdentified() {
incident.Status = 2
}
// SetWatching sets status to Watching
func (incident *Incident) SetWatching() {
incident.Status = 3
}
// SetFixed sets status to Fixed
func (incident *Incident) SetFixed() {
incident.Status = 4
}

10
make.sh Executable file
View File

@@ -0,0 +1,10 @@
#!/bin/bash
set -e
if [ "$1" == "test" ]; then
reflex -r '\.go$' -s -d none -- sh -c 'go test ./...'
fi
reflex -r '\.go$' -s -d none -- sh -c 'go build -o ./cachet-monitor ./cli/ && ./cachet-monitor -c config.yml'
exit 0

252
monitor.go
View File

@@ -1,252 +0,0 @@
package cachet
import (
"sync"
"time"
"github.com/Sirupsen/logrus"
)
const DefaultInterval = time.Second * 60
const DefaultTimeout = time.Second
const DefaultTimeFormat = "15:04:05 Jan 2 MST"
const HistorySize = 10
type MonitorInterface interface {
ClockStart(*CachetMonitor, MonitorInterface, *sync.WaitGroup)
ClockStop()
tick(MonitorInterface)
test() bool
Validate() []string
GetMonitor() *AbstractMonitor
Describe() []string
}
// AbstractMonitor data model
type AbstractMonitor struct {
Name string
Target string
// (default)http / dns
Type string
Strict bool
Interval time.Duration
Timeout time.Duration
MetricID int `mapstructure:"metric_id"`
ComponentID int `mapstructure:"component_id"`
// Templating stuff
Template struct {
Investigating MessageTemplate
Fixed MessageTemplate
}
// Threshold = percentage / number of down incidents
Threshold float32
ThresholdCount bool `mapstructure:"threshold_count"`
// lag / average(lagHistory) * 100 = percentage above average lag
// PerformanceThreshold sets the % limit above which this monitor will trigger degraded-performance
// PerformanceThreshold float32
history []bool
// lagHistory []float32
lastFailReason string
incident *Incident
config *CachetMonitor
// Closed when mon.Stop() is called
stopC chan bool
}
func (mon *AbstractMonitor) Validate() []string {
errs := []string{}
if len(mon.Name) == 0 {
errs = append(errs, "Name is required")
}
if mon.Interval < 1 {
mon.Interval = DefaultInterval
}
if mon.Timeout < 1 {
mon.Timeout = DefaultTimeout
}
if mon.Timeout > mon.Interval {
errs = append(errs, "Timeout greater than interval")
}
if mon.ComponentID == 0 && mon.MetricID == 0 {
errs = append(errs, "component_id & metric_id are unset")
}
if mon.Threshold <= 0 {
mon.Threshold = 100
}
if err := mon.Template.Fixed.Compile(); err != nil {
errs = append(errs, "Could not compile \"fixed\" template: "+err.Error())
}
if err := mon.Template.Investigating.Compile(); err != nil {
errs = append(errs, "Could not compile \"investigating\" template: "+err.Error())
}
return errs
}
func (mon *AbstractMonitor) GetMonitor() *AbstractMonitor {
return mon
}
func (mon *AbstractMonitor) Describe() []string {
features := []string{"Type: " + mon.Type}
if len(mon.Name) > 0 {
features = append(features, "Name: "+mon.Name)
}
return features
}
func (mon *AbstractMonitor) ClockStart(cfg *CachetMonitor, iface MonitorInterface, wg *sync.WaitGroup) {
wg.Add(1)
mon.config = cfg
mon.stopC = make(chan bool)
if cfg.Immediate {
mon.tick(iface)
}
ticker := time.NewTicker(mon.Interval * time.Second)
for {
select {
case <-ticker.C:
mon.tick(iface)
case <-mon.stopC:
wg.Done()
return
}
}
}
func (mon *AbstractMonitor) ClockStop() {
select {
case <-mon.stopC:
return
default:
close(mon.stopC)
}
}
func (mon *AbstractMonitor) test() bool { return false }
func (mon *AbstractMonitor) tick(iface MonitorInterface) {
reqStart := getMs()
up := iface.test()
lag := getMs() - reqStart
histSize := HistorySize
if mon.ThresholdCount {
histSize = int(mon.Threshold)
}
if len(mon.history) == histSize-1 {
logrus.Warnf("%v is now saturated", mon.Name)
}
if len(mon.history) >= histSize {
mon.history = mon.history[len(mon.history)-(histSize-1):]
}
mon.history = append(mon.history, up)
mon.AnalyseData()
// report lag
if mon.MetricID > 0 {
go mon.config.API.SendMetric(mon.MetricID, lag)
}
}
// TODO: test
// AnalyseData decides if the monitor is statistically up or down and creates / resolves an incident
func (mon *AbstractMonitor) AnalyseData() {
// look at the past few incidents
numDown := 0
for _, wasUp := range mon.history {
if wasUp == false {
numDown++
}
}
t := (float32(numDown) / float32(len(mon.history))) * 100
l := logrus.WithFields(logrus.Fields{
"monitor": mon.Name,
"time": time.Now().Format(mon.config.DateFormat),
})
if numDown == 0 {
l.Printf("monitor is up")
} else if mon.ThresholdCount {
l.Printf("monitor down %d/%d", numDown, int(mon.Threshold))
} else {
l.Printf("monitor down %.2f%%/%.2f%%", t, mon.Threshold)
}
histSize := HistorySize
if mon.ThresholdCount {
histSize = int(mon.Threshold)
}
if len(mon.history) != histSize {
// not saturated
return
}
triggered := (mon.ThresholdCount && numDown == int(mon.Threshold)) || (!mon.ThresholdCount && t > mon.Threshold)
if triggered && mon.incident == nil {
// create incident
tplData := getTemplateData(mon)
tplData["FailReason"] = mon.lastFailReason
subject, message := mon.Template.Investigating.Exec(tplData)
mon.incident = &Incident{
Name: subject,
ComponentID: mon.ComponentID,
Message: message,
Notify: true,
}
// is down, create an incident
l.Warnf("creating incident. Monitor is down: %v", mon.lastFailReason)
// set investigating status
mon.incident.SetInvestigating()
// create/update incident
if err := mon.incident.Send(mon.config); err != nil {
l.Printf("Error sending incident: %v", err)
}
return
}
// still triggered or no incident
if triggered || mon.incident == nil {
return
}
// was down, created an incident, its now ok, make it resolved.
l.Warn("Resolving incident")
// resolve incident
tplData := getTemplateData(mon)
tplData["incident"] = mon.incident
subject, message := mon.Template.Fixed.Exec(tplData)
mon.incident.Name = subject
mon.incident.Message = message
mon.incident.SetFixed()
if err := mon.incident.Send(mon.config); err != nil {
l.Printf("Error sending incident: %v", err)
}
mon.lastFailReason = ""
mon.incident = nil
}

7
monitor_test.go
View File

@@ -1,7 +0,0 @@
package cachet
import (
"testing"
)
func TestAnalyseData(t *testing.T) {}

40
dns.go → monitors/dns.go
View File

@@ -1,14 +1,33 @@
package cachet
package monitors
import (
"errors"
"net"
"regexp"
"strings"
"github.com/Sirupsen/logrus"
"github.com/miekg/dns"
"github.com/sirupsen/logrus"
)
// Investigating template
var defaultDNSInvestigatingTpl = MessageTemplate{
Subject: `{{ .Monitor.Name }} - {{ .SystemName }}`,
Message: `{{ .Monitor.Name }} DNS check **failed** (server time: {{ .now }})
{{ .FailReason }}`,
}
// Fixed template
var defaultDNSFixedTpl = MessageTemplate{
Subject: `{{ .Monitor.Name }} - {{ .SystemName }}`,
Message: `**Resolved** - {{ .now }}
- - -
{{ .incident.Message }}`,
}
type DNSAnswer struct {
Regex string
regexp *regexp.Regexp
@@ -28,8 +47,11 @@ type DNSMonitor struct {
Answers []DNSAnswer
}
func (monitor *DNSMonitor) Validate() []string {
errs := monitor.AbstractMonitor.Validate()
func (monitor *DNSMonitor) Validate(validate backendValidateFunc) []string {
monitor.Template.Investigating.SetDefault(defaultDNSInvestigatingTpl)
monitor.Template.Fixed.SetDefault(defaultDNSFixedTpl)
errs := monitor.AbstractMonitor.Validate(validate)
if len(monitor.DNS) == 0 {
config, _ := dns.ClientConfigFromFile("/etc/resolv.conf")
@@ -61,7 +83,7 @@ func (monitor *DNSMonitor) Validate() []string {
return errs
}
func (monitor *DNSMonitor) test() bool {
func (monitor *DNSMonitor) test() (bool, []error) {
m := new(dns.Msg)
m.SetQuestion(dns.Fqdn(monitor.Target), monitor.question)
m.RecursionDesired = true
@@ -70,11 +92,11 @@ func (monitor *DNSMonitor) test() bool {
r, _, err := c.Exchange(m, monitor.DNS)
if err != nil {
logrus.Warnf("DNS error: %v", err)
return false
return false, []error{err}
}
if r.Rcode != dns.RcodeSuccess {
return false
return false, []error{errors.New("Invalid status code returned")}
}
for _, check := range monitor.Answers {
@@ -88,11 +110,11 @@ func (monitor *DNSMonitor) test() bool {
if !found {
logrus.Warnf("DNS check failed: %v. Not found in any of %v", check, r.Answer)
return false
return false, []error{errors.New("Record not found")}
}
}
return true
return true, nil
}
func findDNSType(t string) uint16 {

27
http.go → monitors/http.go
View File

@@ -1,7 +1,8 @@
package cachet
package monitors
import (
"crypto/tls"
"errors"
"io/ioutil"
"net/http"
"regexp"
@@ -13,7 +14,7 @@ import (
// Investigating template
var defaultHTTPInvestigatingTpl = MessageTemplate{
Subject: `{{ .Monitor.Name }} - {{ .SystemName }}`,
Message: `{{ .Monitor.Name }} check **failed** (server time: {{ .now }})
Message: `{{ .Monitor.Name }} HTTP check **failed** (server time: {{ .now }})
{{ .FailReason }}`,
}
@@ -41,7 +42,7 @@ type HTTPMonitor struct {
}
// TODO: test
func (monitor *HTTPMonitor) test() bool {
func (monitor *HTTPMonitor) test() (bool, []error) {
req, err := http.NewRequest(monitor.Method, monitor.Target, nil)
for k, v := range monitor.Headers {
req.Header.Add(k, v)
@@ -56,40 +57,38 @@ func (monitor *HTTPMonitor) test() bool {
resp, err := client.Do(req)
if err != nil {
monitor.lastFailReason = err.Error()
return false
return false, []error{err}
}
defer resp.Body.Close()
if monitor.ExpectedStatusCode > 0 && resp.StatusCode != monitor.ExpectedStatusCode {
monitor.lastFailReason = "Expected HTTP response status: " + strconv.Itoa(monitor.ExpectedStatusCode) + ", got: " + strconv.Itoa(resp.StatusCode)
return false
fail := "Expected HTTP response status: " + strconv.Itoa(monitor.ExpectedStatusCode) + ", got: " + strconv.Itoa(resp.StatusCode)
return false, []error{errors.New(fail)}
}
if monitor.bodyRegexp != nil {
// check response body
responseBody, err := ioutil.ReadAll(resp.Body)
if err != nil {
monitor.lastFailReason = err.Error()
return false
return false, []error{err}
}
if !monitor.bodyRegexp.Match(responseBody) {
monitor.lastFailReason = "Unexpected body: " + string(responseBody) + ".\nExpected to match: " + monitor.ExpectedBody
return false
fail := "Unexpected body: " + string(responseBody) + ".\nExpected to match: " + monitor.ExpectedBody
return false, []error{errors.New(fail)}
}
}
return true
return true, nil
}
// TODO: test
func (mon *HTTPMonitor) Validate() []string {
func (mon *HTTPMonitor) Validate(validate backendValidateFunc) []string {
mon.Template.Investigating.SetDefault(defaultHTTPInvestigatingTpl)
mon.Template.Fixed.SetDefault(defaultHTTPFixedTpl)
errs := mon.AbstractMonitor.Validate()
errs := mon.AbstractMonitor.Validate(validate)
if len(mon.ExpectedBody) > 0 {
exp, err := regexp.Compile(mon.ExpectedBody)

257
monitors/monitor.go Normal file
View File

@@ -0,0 +1,257 @@
package monitors
import (
"sync"
"time"
"github.com/sirupsen/logrus"
)
const DefaultInterval = time.Second * 60
const DefaultTimeout = time.Second
const HistorySize = 10
type MonitorStatus string
const (
MonitorStatusUp = "up"
MonitorStatusDown = "down"
MonitorStatusNotSaturated = "unsaturated"
)
type backendValidateFunc = func(monitor *AbstractMonitor) []string
type MonitorTestFunc func() (up bool, errs []error)
type MonitorTickFunc func(monitor MonitorInterface, status MonitorStatus, errs []error, lag int64)
type MonitorInterface interface {
Start(MonitorTestFunc, *sync.WaitGroup, MonitorTickFunc, bool)
Stop()
tick(MonitorTestFunc) (status MonitorStatus, errors []error, lag int64)
test() (bool, []error)
Validate(validate backendValidateFunc) []string
Describe() []string
GetMonitor() *AbstractMonitor
GetTestFunc() MonitorTestFunc
GetLastStatus() MonitorStatus
UpdateLastStatus(status MonitorStatus) (old MonitorStatus)
}
// AbstractMonitor data model
type AbstractMonitor struct {
Name string
Target string
// (default)http / dns
Type string
Strict bool
Interval time.Duration
Timeout time.Duration
Params map[string]interface{}
// Templating stuff
Template MonitorTemplates
// Threshold = percentage / number of down incidents
Threshold float32
ThresholdCount bool `mapstructure:"threshold_count"`
// lag / average(lagHistory) * 100 = percentage above average lag
// PerformanceThreshold sets the % limit above which this monitor will trigger degraded-performance
// PerformanceThreshold float32
history []bool
lastStatus MonitorStatus
// Closed when mon.Stop() is called
stopC chan bool
}
func (mon *AbstractMonitor) Validate(validate backendValidateFunc) []string {
errs := []string{}
if len(mon.Name) == 0 {
errs = append(errs, "Name is required")
}
if mon.Interval < 1 {
mon.Interval = DefaultInterval
}
if mon.Timeout < 1 {
mon.Timeout = DefaultTimeout
}
if mon.Timeout > mon.Interval {
errs = append(errs, "Timeout greater than interval")
}
// get the backend to validate the monitor
errs = append(errs, validate(mon)...)
if mon.Threshold <= 0 {
mon.Threshold = 100
}
// if len(mon.Template.Fixed.Message) == 0 || len(mon.Template.Fixed.Subject) == 0 {
// errs = append(errs, "\"fixed\" template empty/missing")
// }
// if len(mon.Template.Investigating.Message) == 0 || len(mon.Template.Investigating.Subject) == 0 {
// errs = append(errs, "\"investigating\" template empty/missing")
// }
if err := mon.Template.Fixed.Compile(); err != nil {
errs = append(errs, "Could not compile \"fixed\" template: "+err.Error())
}
if err := mon.Template.Investigating.Compile(); err != nil {
errs = append(errs, "Could not compile \"investigating\" template: "+err.Error())
}
return errs
}
func (mon *AbstractMonitor) GetMonitor() *AbstractMonitor {
return mon
}
func (mon *AbstractMonitor) Describe() []string {
features := []string{"Type: " + mon.Type}
if len(mon.Name) > 0 {
features = append(features, "Name: "+mon.Name)
}
return features
}
func (mon *AbstractMonitor) Start(testFunc MonitorTestFunc, wg *sync.WaitGroup, tickFunc MonitorTickFunc, immediate bool) {
wg.Add(1)
mon.stopC = make(chan bool)
if immediate {
status, errs, lag := mon.tick(testFunc)
tickFunc(mon, status, errs, lag)
}
ticker := time.NewTicker(mon.Interval * time.Second)
for {
select {
case <-ticker.C:
status, errs, lag := mon.tick(testFunc)
tickFunc(mon, status, errs, lag)
case <-mon.stopC:
wg.Done()
return
}
}
}
func (mon *AbstractMonitor) Stop() {
select {
case <-mon.stopC:
return
default:
close(mon.stopC)
}
}
func (mon *AbstractMonitor) tick(testFunc MonitorTestFunc) (status MonitorStatus, errors []error, lag int64) {
reqStart := getMs()
up, errs := testFunc()
lag = getMs() - reqStart
histSize := HistorySize
if mon.ThresholdCount {
histSize = int(mon.Threshold)
}
if len(mon.history) == histSize-1 {
logrus.WithFields(logrus.Fields{
"monitor": mon.Name,
}).Warn("monitor saturated")
}
if len(mon.history) >= histSize {
mon.history = mon.history[len(mon.history)-(histSize-1):]
}
mon.history = append(mon.history, up)
status = mon.GetStatus()
errors = errs
return
}
// TODO: test
// AnalyseData decides if the monitor is statistically up or down and creates / resolves an incident
func (mon *AbstractMonitor) GetStatus() MonitorStatus {
numDown := 0
for _, wasUp := range mon.history {
if wasUp == false {
numDown++
}
}
t := (float32(numDown) / float32(len(mon.history))) * 100
logFields := logrus.Fields{"monitor": mon.Name}
// stop reporting time for jsonformatter, it's there by default
if _, ok := logrus.StandardLogger().Formatter.(*logrus.JSONFormatter); !ok {
logFields["t"] = time.Now()
}
l := logrus.WithFields(logFields)
symbol := "⚠️"
if t == 100 {
symbol = "❌"
}
if numDown == 0 {
l.Printf("👍 up")
} else if mon.ThresholdCount {
l.Printf("%v down (%d/%d)", symbol, numDown, int(mon.Threshold))
} else {
l.Printf("%v down %.0f%%/%.0f%%", symbol, t, mon.Threshold)
}
histSize := HistorySize
if mon.ThresholdCount {
histSize = int(mon.Threshold)
}
if len(mon.history) != histSize {
// not saturated
return MonitorStatusNotSaturated
}
var down bool
if mon.ThresholdCount {
down = numDown >= int(mon.Threshold)
} else {
down = t >= mon.Threshold
}
if !down {
return MonitorStatusUp
}
return MonitorStatusDown
}
func (mon *AbstractMonitor) GetTestFunc() MonitorTestFunc {
return mon.test
}
func (mon *AbstractMonitor) GetLastStatus() MonitorStatus {
return mon.lastStatus
}
func (mon *AbstractMonitor) UpdateLastStatus(status MonitorStatus) (old MonitorStatus) {
old = mon.lastStatus
mon.lastStatus = status
return
}
func (mon *AbstractMonitor) test() (bool, []error) { return false, nil }
func getMs() int64 {
return time.Now().UnixNano() / int64(time.Millisecond)
}

7
template.go → monitors/template.go
View File

@@ -1,10 +1,15 @@
package cachet
package monitors
import (
"bytes"
"text/template"
)
type MonitorTemplates struct {
Investigating MessageTemplate
Fixed MessageTemplate
}
type MessageTemplate struct {
Subject string `json:"subject"`
Message string `json:"message"`