feat: Parameterize pod restart count (#18)

Co-authored-by: able8 <[email protected]>

CHANGELOG.md

@@ -5,6 +5,10 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 
+## [1.2.0] - 2023-01-03
+### Added
+- Parameterize pod restart count
+
 ## [1.1.0] - 2022-09-19
 ### Added
 - Support ignoring specific namespaces and pods 

README.md

@@ -76,6 +76,7 @@ helm uninstall k8s-pod-restart-info-collector
 | `slackUsername`                     | Slack username (Display on slack message) | default: `"k8s-pod-restart-info-collector"`          |
 | `slackChannel`                      | Slack channel name | default: `"restart-info-nonprod"`          |
 | `muteSeconds`                       | The time to mute duplicate pod alerts | default: `"600"`    
+| `ignoreRestartCount`                | The number of pod restart count to ignore | default: `"30"`
 | `ignoredNamespaces`                 | A comma-separated list of namespaces to ignore | default: `""`    
 | `ignoredPodNamePrefixes`            | A comma-separated list of pod name prefixes to ignore | default: `""`   
 | `slackWebhookUrl`                   | Slack webhook URL | required if slackWebhooUrlSecretKeyRef is not present                       |

build.sh

@@ -1,5 +1,5 @@
 #!/bin/bash
-TAG="v1.1.0"
+TAG="v1.2.0"
 docker buildx build --platform linux/amd64 -t devopsairwallex/k8s-pod-restart-info-collector:${TAG} .
 docker push devopsairwallex/k8s-pod-restart-info-collector:${TAG}
 

controller.go

@@ -37,6 +37,7 @@ type Controller struct {
 // NewController creates a new Controller.
 func NewController(clientset kubernetes.Interface, slack Slack) *Controller {
 	const resyncPeriod = 0
+	ignoreRestartCount := getIgnoreRestartCount()
 
 	queue := workqueue.NewRateLimitingQueue(workqueue.DefaultControllerRateLimiter())
 	informerFactory := informers.NewSharedInformerFactory(clientset, resyncPeriod)
@@ -61,12 +62,10 @@ func NewController(clientset kubernetes.Interface, slack Slack) *Controller {
 				return
 			}
 
-			klog.Infof("Update: %s/%s\n", newPod.Namespace, newPod.Name)
-
 			newPodRestartCount := getPodRestartCount(newPod)
-			// Ignore when restartCount > 30
-			if newPodRestartCount > 30 {
-				klog.Infof("Ignore: %s/%s restartCount: %d > 30\n", newPod.Namespace, newPod.Name, newPodRestartCount)
+			// Ignore when restartCount > ignoreRestartCount
+			if newPodRestartCount > ignoreRestartCount {
+				klog.Infof("Ignore: %s/%s restartCount: %d > %d\n", newPod.Namespace, newPod.Name, newPodRestartCount, ignoreRestartCount)
 				return
 			}
 

helm/templates/deployment.yaml

@@ -39,6 +39,8 @@ spec:
               value: {{ .Values.slackUsername | quote}}
             - name: MUTE_SECONDS
               value: {{ .Values.muteSeconds | quote}}
+            - name: IGNORE_RESTART_COUNT
+              value: {{ .Values.ignoreRestartCount | quote}}
             - name: IGNORED_NAMESPACES
               value: {{ .Values.ignoredNamespaces | quote}}
             - name: IGNORED_POD_NAME_PREFIXES

helm/values.yaml

@@ -11,6 +11,7 @@
 slackChannel: "restart-info-nonprod"
 slackUsername: "k8s-pod-restart-info-collector"
 muteSeconds: 600
+ignoreRestartCount: 30
 
 # A comma-separated list of namespaces to ignore
 ignoredNamespaces: ""

helpers.go

@@ -57,6 +57,15 @@ func isIgnoredPod(name string) bool {
 	return false
 }
 
+func getIgnoreRestartCount() int {
+	ignoreRestartCount, err := strconv.Atoi(os.Getenv("IGNORE_RESTART_COUNT"))
+	if err != nil {
+		ignoreRestartCount = 30
+		klog.Warningf("Environment variable IGNORE_RESTART_COUNT is not set, default: %d\n", ignoreRestartCount)
+	}
+	return ignoreRestartCount
+}
+
 func printPod(pod *v1.Pod) (string, error) {
 	restarts := 0
 	totalContainers := len(pod.Spec.Containers)