Merge pull request #15 from fguimond/dedup-key

Dedup key & severity mapping

.gitignore

@@ -1,4 +1,5 @@
 # Binaries for programs and plugins
+sensu-pagerduty-handler
 *.exe
 *.exe~
 *.dll
@@ -16,4 +17,7 @@ vendor/
 
 # deploy
 bonsai/
-dist/
+dist/
+
+# IDE's
+.idea/

Gopkg.toml

@@ -35,7 +35,11 @@
 
 [[constraint]]
   name = "github.com/sensu/sensu-plugins-go-library"
-  version = "0.2.0"
+  version = "0.3.0"
+
+[[constraint]]
+  name = "github.com/stretchr/testify"
+  version = "1.3.0"
 
 [prune]
   go-tests = true

README.md

@@ -32,6 +32,9 @@ Example Sensu Go handler definition:
         "command": "sensu-pagerduty-handler",
         "env_vars": [
           "PAGERDUTY_TOKEN=SECRET",
+          "PAGERDUTY_DEDUP_KEY=SENSU_EVENT_LABEL",
+          "PAGERDUTY_DEDUP_KEY_TEMPLATE={{.Entity.Name}}-{{.Check.Name}}",
+          "PAGERDUTY_STATUS_MAP={\"info\":[130,10],\"error\":[4]}"
         ],
         "timeout": 10,
         "filters": [
@@ -73,13 +76,58 @@ Usage:
   sensu-pagerduty-handler [flags]
 
 Flags:
-  -h, --help           help for sensu-pagerduty-handler
-  -t, --token string   The PagerDuty V2 API authentication token, use default from PAGERDUTY_TOKEN env var
-
+  -d, --dedup-key string            The Sensu event label specifying the PagerDuty V2 API deduplication key, use default from PAGERDUTY_DEDUP_KEY env var
+  -k, --dedup-key-template string   The PagerDuty V2 API deduplication key template, use default from PAGERDUTY_DEDUP_KEY_TEMPLATE env var
+  -h, --help                        help for sensu-pagerduty-handler
+  -s, --status-map string           The status map used to translate a Sensu check status to a PagerDuty severity, use default from PAGERDUTY_STATUS_MAP env var
+  -t, --token string                The PagerDuty V2 API authentication token, use default from PAGERDUTY_TOKEN env var
 ```
 
 **Note:** Make sure to set the `PAGERDUTY_TOKEN` environment variable for sensitive credentials in production to prevent leaking into system process table. Please remember command arguments can be viewed by unprivileged users using commands such as `ps` or `top`. The `--token` argument is provided as an override primarily for testing purposes. 
 
+### Deduplication Key Priority
+
+The deduplication key is determined using the following priority:
+1. --dedup-key  --  specifies the entity label containing the key
+1. --dedup-key-template  --  a template containing the values
+1. the default value containing the entity and check names
+
+### PagerDuty Severity Mapping
+
+Optionally you can provide mapping information between the Sensu check status and the PagerDuty incident severity.
+To provide the mapping you need to use the `--status-map` command line option or the `PAGERDUTY_STATUS_MAP` environment variable.
+The option accepts a JSON document containing the mapping information. Here's an example of the JSON document:
+
+```json
+{
+    "info": [
+        0,
+        1
+    ],
+    "warning": [
+        2
+    ],
+    "critical:": [
+        3
+    ],
+    "error": [
+        4,
+        5,
+        6,
+        7,
+        8,
+        9,
+        10
+    ]
+}
+```
+
+The valid [PagerDuty alert severity levels][5] are the following:
+* `info`
+* `warning`
+* `critical`
+* `error`
+
 ## Contributing
 
 See https://github.com/sensu/sensu-go/blob/master/CONTRIBUTING.md
@@ -88,3 +136,4 @@ See https://github.com/sensu/sensu-go/blob/master/CONTRIBUTING.md
 [2]: https://www.pagerduty.com/ 
 [3]: https://docs.sensu.io/sensu-go/5.0/reference/handlers/#how-do-sensu-handlers-work
 [4]: https://github.com/sensu/sensu-pagerduty-handler/releases
+[5]: https://support.pagerduty.com/docs/dynamic-notifications#section-eventalert-severity-levels

event.json

@@ -1,4 +1,5 @@
 {
+  "timestamp": 1561246706000,
   "entity": {
     "entity_class": "agent",
     "system": {
@@ -58,7 +59,9 @@
     "metadata": {
       "name": "webserver01",
       "namespace": "default",
-      "labels": null,
+      "labels": {
+        "someLabel": "one hell of a deduplication key"
+      },
       "annotations": null
     }
   },
@@ -103,7 +106,7 @@
     "issued": 1542667666,
     "output": "example output",
     "state": "failing",
-    "status": 1,
+    "status": 2,
     "total_state_change": 0,
     "last_ok": 0,
     "occurrences": 1,

main.go

@@ -1,18 +1,25 @@
 package main
 
 import (
+	"encoding/json"
 	"fmt"
-
 	"github.com/PagerDuty/go-pagerduty"
 	corev2 "github.com/sensu/sensu-go/api/core/v2"
 	"github.com/sensu/sensu-plugins-go-library/sensu"
+	"github.com/sensu/sensu-plugins-go-library/templates"
+	"log"
 )
 
 type HandlerConfig struct {
 	sensu.PluginConfig
-	authToken string
+	authToken        string
+	dedupKey         string
+	dedupKeyTemplate string
+	statusMapJson    string
 }
 
+type eventStatusMap map[string][]uint32
+
 var (
 	config = HandlerConfig{
 		PluginConfig: sensu.PluginConfig{
@@ -31,6 +38,33 @@ var (
 			Value:     &config.authToken,
 			Default:   "",
 		},
+		{
+			Path:      "dedup-key",
+			Env:       "PAGERDUTY_DEDUP_KEY",
+			Argument:  "dedup-key",
+			Shorthand: "d",
+			Usage:     "The Sensu event label specifying the PagerDuty V2 API deduplication key, use default from PAGERDUTY_DEDUP_KEY env var",
+			Value:     &config.dedupKey,
+			Default:   "",
+		},
+		{
+			Path:      "dedup-key-template",
+			Env:       "PAGERDUTY_DEDUP_KEY_TEMPLATE",
+			Argument:  "dedup-key-template",
+			Shorthand: "k",
+			Usage:     "The PagerDuty V2 API deduplication key template, use default from PAGERDUTY_DEDUP_KEY_TEMPLATE env var",
+			Value:     &config.dedupKeyTemplate,
+			Default:   "",
+		},
+		{
+			Path:      "status-map",
+			Env:       "PAGERDUTY_STATUS_MAP",
+			Argument:  "status-map",
+			Shorthand: "s",
+			Usage:     "The status map used to translate a Sensu check status to a PagerDuty severity, use default from PAGERDUTY_STATUS_MAP env var",
+			Value:     &config.statusMapJson,
+			Default:   "",
+		},
 	}
 )
 
@@ -47,12 +81,11 @@ func checkArgs(event *corev2.Event) error {
 }
 
 func manageIncident(event *corev2.Event) error {
-	severity := "warning"
-
-	if event.Check.Status < 3 {
-		severities := []string{"info", "warning", "critical"}
-		severity = severities[event.Check.Status]
+	severity, err := getPagerDutySeverity(event, config.statusMapJson)
+	if err != nil {
+		return err
 	}
+	log.Printf("Incident severity: %s", severity)
 
 	summary := fmt.Sprintf("%s/%s : %s", event.Entity.Name, event.Check.Name, event.Check.Output)
 
@@ -70,7 +103,13 @@ func manageIncident(event *corev2.Event) error {
 		action = "resolve"
 	}
 
-	dedupKey := fmt.Sprintf("%s-%s", event.Entity.Name, event.Check.Name)
+	dedupKey, err := getPagerDutyDedupKey(event)
+	if err != nil {
+		return err
+	}
+	if len(dedupKey) == 0 {
+		return fmt.Errorf("pagerduty dedup key is empty")
+	}
 
 	pdEvent := pagerduty.V2Event{
 		RoutingKey: config.authToken,
@@ -79,10 +118,84 @@ func manageIncident(event *corev2.Event) error {
 		DedupKey:   dedupKey,
 	}
 
-	_, err := pagerduty.ManageEvent(pdEvent)
+	_, err = pagerduty.ManageEvent(pdEvent)
 	if err != nil {
 		return err
 	}
 
 	return nil
 }
+
+// getPagerDutyDedupKey returns the PagerDuty deduplication key. The following priority is used to determine the
+// deduplication key.
+// 1. --dedup-key  --  specifies the entity label containing the key
+// 2. --dedup-key-template  --  a template containing the values
+// 3. the default value including the entity and check names
+func getPagerDutyDedupKey(event *corev2.Event) (string, error) {
+	if len(config.dedupKey) > 0 {
+		labelValue := event.Entity.Labels[config.dedupKey]
+		if len(labelValue) > 0 {
+			return labelValue, nil
+		} else {
+			return "", fmt.Errorf("no deduplication key value in label %s", config.dedupKey)
+		}
+	}
+
+	if len(config.dedupKeyTemplate) > 0 {
+		return templates.EvalTemplate("dedupKey", config.dedupKeyTemplate, event)
+	} else {
+		return fmt.Sprintf("%s-%s", event.Entity.Name, event.Check.Name), nil
+	}
+}
+
+func getPagerDutySeverity(event *corev2.Event, statusMapJson string) (string, error) {
+	var statusMap map[uint32]string
+	var err error
+
+	if len(statusMapJson) > 0 {
+		statusMap, err = parseStatusMap(statusMapJson)
+		if err != nil {
+			return "", err
+		}
+	}
+
+	if len(statusMap) > 0 {
+		status := event.Check.Status
+		severity := statusMap[status]
+		if len(severity) > 0 {
+			return severity, nil
+		}
+	}
+
+	// Default to these values is no status map is found
+	severity := "warning"
+	if event.Check.Status < 3 {
+		severities := []string{"info", "warning", "critical"}
+		severity = severities[event.Check.Status]
+	}
+
+	return severity, nil
+}
+
+func parseStatusMap(statusMapJson string) (map[uint32]string, error) {
+	validPagerDutySeverities := map[string]bool{"info": true, "critical": true, "warning": true, "error": true}
+
+	statusMap := eventStatusMap{}
+	err := json.Unmarshal([]byte(statusMapJson), &statusMap)
+	if err != nil {
+		return nil, err
+	}
+
+	// Reverse the map to key it on the status
+	statusToSeverityMap := map[uint32]string{}
+	for severity, statuses := range statusMap {
+		if !validPagerDutySeverities[severity] {
+			return nil, fmt.Errorf("invalid pagerduty severity: %s", severity)
+		}
+		for i := range statuses {
+			statusToSeverityMap[uint32(statuses[i])] = severity
+		}
+	}
+
+	return statusToSeverityMap, nil
+}