Log & resolve incidents, add readme
This commit is contained in:
@@ -1,7 +1,12 @@
|
|||||||
package cachet
|
package cachet
|
||||||
|
|
||||||
import (
|
import (
|
||||||
"time"
|
"fmt"
|
||||||
|
"bytes"
|
||||||
|
"io/ioutil"
|
||||||
|
"strconv"
|
||||||
|
"net/http"
|
||||||
|
"encoding/json"
|
||||||
)
|
)
|
||||||
|
|
||||||
type Incident struct {
|
type Incident struct {
|
||||||
@@ -12,6 +17,77 @@ type Incident struct {
|
|||||||
Human_status string `json:"human_status"`
|
Human_status string `json:"human_status"`
|
||||||
Component *Component `json:"component"`
|
Component *Component `json:"component"`
|
||||||
Component_id *int `json:"component_id"`
|
Component_id *int `json:"component_id"`
|
||||||
Created_at *time.Time `json:"created_at"`
|
Created_at int `json:"created_at"`
|
||||||
Updated_at *time.Time `json:"updated_at"`
|
Updated_at int `json:"updated_at"`
|
||||||
|
}
|
||||||
|
|
||||||
|
type IncidentData struct {
|
||||||
|
Incident Incident `json:"data"`
|
||||||
|
}
|
||||||
|
|
||||||
|
func (incident *Incident) Send() {
|
||||||
|
jsonBytes, err := json.Marshal(incident)
|
||||||
|
if err != nil {
|
||||||
|
panic(err)
|
||||||
|
}
|
||||||
|
|
||||||
|
var req *http.Request
|
||||||
|
if incident.Id == 0 {
|
||||||
|
req, err = http.NewRequest("POST", apiUrl + "/incidents", bytes.NewBuffer(jsonBytes))
|
||||||
|
} else {
|
||||||
|
req, err = http.NewRequest("PUT", apiUrl + "/incidents/" + strconv.Itoa(incident.Id), bytes.NewBuffer(jsonBytes))
|
||||||
|
}
|
||||||
|
|
||||||
|
if err != nil {
|
||||||
|
panic(err)
|
||||||
|
}
|
||||||
|
|
||||||
|
req.Header.Set("Content-Type", "application/json")
|
||||||
|
req.Header.Set("X-Cachet-Token", apiToken)
|
||||||
|
|
||||||
|
client := &http.Client{}
|
||||||
|
resp, err := client.Do(req)
|
||||||
|
if err != nil {
|
||||||
|
panic(err)
|
||||||
|
}
|
||||||
|
|
||||||
|
defer resp.Body.Close()
|
||||||
|
|
||||||
|
body, _ := ioutil.ReadAll(resp.Body)
|
||||||
|
fmt.Println(strconv.Itoa(resp.StatusCode) + " " + string(body))
|
||||||
|
|
||||||
|
var data IncidentData
|
||||||
|
err = json.Unmarshal(body, &data)
|
||||||
|
if err != nil {
|
||||||
|
fmt.Println("Cannot parse incident body.")
|
||||||
|
panic(err)
|
||||||
|
} else {
|
||||||
|
incident.Id = data.Incident.Id
|
||||||
|
}
|
||||||
|
|
||||||
|
fmt.Println("ID:"+strconv.Itoa(incident.Id))
|
||||||
|
|
||||||
|
if resp.StatusCode != 200 {
|
||||||
|
fmt.Println("Could not create/update incident!")
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
func (incident *Incident) SetInvestigating() {
|
||||||
|
incident.Status = 1
|
||||||
|
incident.Human_status = "Investigating"
|
||||||
|
}
|
||||||
|
|
||||||
|
func (incident *Incident) SetIdentified() {
|
||||||
|
incident.Status = 2
|
||||||
|
incident.Human_status = "Identified"
|
||||||
|
}
|
||||||
|
|
||||||
|
func (incident *Incident) SetWatching() {
|
||||||
|
incident.Status = 3
|
||||||
|
incident.Human_status = "Watching"
|
||||||
|
}
|
||||||
|
|
||||||
|
func (incident *Incident) SetFixed() {
|
||||||
|
incident.Status = 4
|
||||||
|
incident.Human_status = "Fixed"
|
||||||
}
|
}
|
||||||
@@ -9,6 +9,7 @@ import (
|
|||||||
const timeout = time.Duration(time.Second)
|
const timeout = time.Duration(time.Second)
|
||||||
|
|
||||||
type Monitor struct {
|
type Monitor struct {
|
||||||
|
Name string `json:"name"`
|
||||||
Url string `json:"url"`
|
Url string `json:"url"`
|
||||||
MetricId int `json:"metric_id"`
|
MetricId int `json:"metric_id"`
|
||||||
Threshold float32 `json:"threshold"`
|
Threshold float32 `json:"threshold"`
|
||||||
@@ -30,9 +31,8 @@ func (monitor *Monitor) Run() {
|
|||||||
monitor.History = append(monitor.History, isUp)
|
monitor.History = append(monitor.History, isUp)
|
||||||
monitor.AnalyseData()
|
monitor.AnalyseData()
|
||||||
|
|
||||||
if isUp == true {
|
if isUp == true && monitor.MetricId > 0 {
|
||||||
SendMetric(monitor.MetricId, lag)
|
SendMetric(monitor.MetricId, lag)
|
||||||
return
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
@@ -52,11 +52,6 @@ func (monitor *Monitor) doRequest() bool {
|
|||||||
|
|
||||||
func (monitor *Monitor) AnalyseData() {
|
func (monitor *Monitor) AnalyseData() {
|
||||||
// look at the past few incidents
|
// look at the past few incidents
|
||||||
if len(monitor.History) != 10 {
|
|
||||||
// not enough data
|
|
||||||
return
|
|
||||||
}
|
|
||||||
|
|
||||||
numDown := 0
|
numDown := 0
|
||||||
for _, wasUp := range monitor.History {
|
for _, wasUp := range monitor.History {
|
||||||
if wasUp == false {
|
if wasUp == false {
|
||||||
@@ -65,14 +60,31 @@ func (monitor *Monitor) AnalyseData() {
|
|||||||
}
|
}
|
||||||
|
|
||||||
t := (float32(numDown) / float32(len(monitor.History))) * 100
|
t := (float32(numDown) / float32(len(monitor.History))) * 100
|
||||||
fmt.Printf("%s %.2f%% Down. Threshold: %.2f%%\n", monitor.Url, t, monitor.Threshold)
|
fmt.Printf("%s %.2f%% Down at %v. Threshold: %.2f%%\n", monitor.Url, t, time.Now().UnixNano() / int64(time.Second), monitor.Threshold)
|
||||||
|
|
||||||
|
if len(monitor.History) != 10 {
|
||||||
|
// not enough data
|
||||||
|
return
|
||||||
|
}
|
||||||
|
|
||||||
if t > monitor.Threshold && monitor.Incident == nil {
|
if t > monitor.Threshold && monitor.Incident == nil {
|
||||||
// is down, create an incident
|
// is down, create an incident
|
||||||
fmt.Println("Creating incident...")
|
fmt.Println("Creating incident...")
|
||||||
monitor.Incident = &Incident{}
|
|
||||||
|
monitor.Incident = &Incident{
|
||||||
|
Name: monitor.Name,
|
||||||
|
Message: monitor.Name + " is unreachable.",
|
||||||
|
}
|
||||||
|
|
||||||
|
monitor.Incident.SetInvestigating()
|
||||||
|
monitor.Incident.Send()
|
||||||
} else if t < monitor.Threshold && monitor.Incident != nil {
|
} else if t < monitor.Threshold && monitor.Incident != nil {
|
||||||
// was down, created an incident, its now ok, make it resolved.
|
// was down, created an incident, its now ok, make it resolved.
|
||||||
fmt.Println("Updating incident to resolved...")
|
fmt.Println("Updating incident to resolved...")
|
||||||
|
|
||||||
|
monitor.Incident.SetFixed()
|
||||||
|
monitor.Incident.Send()
|
||||||
|
|
||||||
monitor.Incident = nil
|
monitor.Incident = nil
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|||||||
17
main.go
17
main.go
@@ -1,25 +1,38 @@
|
|||||||
package main
|
package main
|
||||||
|
|
||||||
import (
|
import (
|
||||||
|
"fmt"
|
||||||
"time"
|
"time"
|
||||||
"./cachet"
|
"github.com/castawaylabs/cachet-monitor/cachet"
|
||||||
)
|
)
|
||||||
|
|
||||||
func main() {
|
func main() {
|
||||||
monitors := []*cachet.Monitor{
|
monitors := []*cachet.Monitor{
|
||||||
/*&cachet.Monitor{
|
/*&cachet.Monitor{
|
||||||
|
Name: "nodegear frontend",
|
||||||
Url: "https://nodegear.io/ping",
|
Url: "https://nodegear.io/ping",
|
||||||
MetricId: 1,
|
MetricId: 1,
|
||||||
Threshold: 80.0,
|
Threshold: 80.0,
|
||||||
|
ExpectedStatusCode: 200,
|
||||||
},*/
|
},*/
|
||||||
&cachet.Monitor{
|
&cachet.Monitor{
|
||||||
|
Name: "local test server",
|
||||||
Url: "http://localhost:1337",
|
Url: "http://localhost:1337",
|
||||||
MetricId: 1,
|
|
||||||
Threshold: 80.0,
|
Threshold: 80.0,
|
||||||
ExpectedStatusCode: 200,
|
ExpectedStatusCode: 200,
|
||||||
},
|
},
|
||||||
}
|
}
|
||||||
|
|
||||||
|
fmt.Printf("Starting %d monitors:\n", len(monitors))
|
||||||
|
for _, monitor := range monitors {
|
||||||
|
fmt.Printf(" %s: GET %s & Expect HTTP %d\n", monitor.Name, monitor.Url, monitor.ExpectedStatusCode)
|
||||||
|
if monitor.MetricId > 0 {
|
||||||
|
fmt.Printf(" - Logs lag to metric id: %d\n", monitor.MetricId)
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
fmt.Println()
|
||||||
|
|
||||||
ticker := time.NewTicker(time.Second)
|
ticker := time.NewTicker(time.Second)
|
||||||
for _ = range ticker.C {
|
for _ = range ticker.C {
|
||||||
for _, monitor := range monitors {
|
for _, monitor := range monitors {
|
||||||
|
|||||||
18
readme.md
Normal file
18
readme.md
Normal file
@@ -0,0 +1,18 @@
|
|||||||
|
Cachet Monitor plugin
|
||||||
|
=====================
|
||||||
|
|
||||||
|
This is a monitoring plugin for CachetHQ.
|
||||||
|
|
||||||
|
How to run:
|
||||||
|
-----------
|
||||||
|
|
||||||
|
1. Set up [Go](https://golang.org)
|
||||||
|
2. `go install github.com/castawaylabs/cachet-monitor`
|
||||||
|
3. `cachet-monitor`
|
||||||
|
|
||||||
|
Environment variables:
|
||||||
|
----------------------
|
||||||
|
|
||||||
|
| Name | Example Value | Description |
|
||||||
|
| CACHET_API | http://demo.cachethq.io/api | URL endpoint for cachet api |
|
||||||
|
| CACHET_TOKEN | randomvalue | API Authentication token |
|
||||||
Reference in New Issue
Block a user