Beginning of incident reporting

cachet/component.go

@@ -0,0 +1,18 @@
+package cachet
+
+import (
+	"time"
+)
+
+type Component struct {
+	Id int `json:"id"`
+	Name string `json:"name"`
+	Description string `json:"description"`
+	Status int `json:"status"`
+	Link *string `json:"link"`
+	Order *int `json:"order"`
+	Group_id *int `json:"group_id"`
+	Created_at *time.Time `json:"created_at"`
+	Updated_at *time.Time `json:"updated_at"`
+	Deleted_at *time.Time `json:"deleted_at"`
+}

cachet/incident.go

@@ -0,0 +1,17 @@
+package cachet
+
+import (
+	"time"
+)
+
+type Incident struct {
+	Id int `json:"id"`
+	Name string `json:"name"`
+	Message string `json:"message"`
+	Status int `json:"status"`// 4?
+	Human_status string `json:"human_status"`
+	Component *Component `json:"component"`
+	Component_id *int `json:"component_id"`
+	Created_at *time.Time `json:"created_at"`
+	Updated_at *time.Time `json:"updated_at"`
+}

cachet/metrics.go

@@ -33,8 +33,8 @@ func SendMetric(metricId int, delay int64) {
 
 	defer resp.Body.Close()
 
-	body, _ := ioutil.ReadAll(resp.Body)
-	fmt.Println(strconv.Itoa(resp.StatusCode) + " " + string(body))
+	_, _ = ioutil.ReadAll(resp.Body)
+	// fmt.Println(strconv.Itoa(resp.StatusCode) + " " + string(body))
 
 	if resp.StatusCode != 200 {
 		fmt.Println("Could not log data point!")

cachet/monitor.go

@@ -11,37 +11,70 @@ const timeout = time.Duration(time.Second)
 type Monitor struct {
 	Url string `json:"url"`
 	MetricId int `json:"metric_id"`
+	Threshold float32 `json:"threshold"`
+	ComponentId *int `json:"component_id"`
+	ExpectedStatusCode int `json:"expected_status_code"`
+
+	History []bool `json:"-"`
+	Incident *Incident `json:"-"`
 }
 
 func (monitor *Monitor) Run() {
 	reqStart := getMs()
-	err := monitor.doRequest()
+	isUp := monitor.doRequest()
 	lag := getMs() - reqStart
 
-	failed := false
-	if err != nil {
-		failed = true
+	if len(monitor.History) >= 10 {
+		monitor.History = monitor.History[len(monitor.History)-9:]
 	}
+	monitor.History = append(monitor.History, isUp)
+	monitor.AnalyseData()
 
-	if failed == true {
-		fmt.Println("Req failed")
+	if isUp == true {
+		SendMetric(monitor.MetricId, lag)
+		return
 	}
-
-	SendMetric(1, lag)
 }
 
-func (monitor *Monitor) doRequest() error {
+func (monitor *Monitor) doRequest() bool {
 	client := &http.Client{
 		Timeout: timeout,
 	}
-	resp, err := client.Get(monitor.Url) // http://127.0.0.1:1337
+	resp, err := client.Get(monitor.Url)
 	if err != nil {
-		return err
+		return false
 	}
 
 	defer resp.Body.Close()
 
-	return nil
+	return resp.StatusCode == monitor.ExpectedStatusCode
+}
+
+func (monitor *Monitor) AnalyseData() {
+	// look at the past few incidents
+	if len(monitor.History) != 10 {
+		// not enough data
+		return
+	}
+
+	numDown := 0
+	for _, wasUp := range monitor.History {
+		if wasUp == false {
+			numDown++
+		}
+	}
+
+	t := (float32(numDown) / float32(len(monitor.History))) * 100
+	fmt.Printf("%s %.2f%% Down. Threshold: %.2f%%\n", monitor.Url, t, monitor.Threshold)
+	if t > monitor.Threshold && monitor.Incident == nil {
+		// is down, create an incident
+		fmt.Println("Creating incident...")
+		monitor.Incident = &Incident{}
+	} else if t < monitor.Threshold && monitor.Incident != nil {
+		// was down, created an incident, its now ok, make it resolved.
+		fmt.Println("Updating incident to resolved...")
+		monitor.Incident = nil
+	}
 }
 
 func getMs() int64 {