Fix: Limit the number of events and minify the events

.github/workflows/integration_test.yml

@@ -109,7 +109,7 @@ jobs:
           helm repo add localstack-repo http://helm.localstack.cloud
           helm upgrade --install localstack localstack-repo/localstack --version 0.1.2
           pip install awscli
-          kubectl wait --timeout=60s --for=condition=ready --all pod
+          kubectl wait --timeout=160s --for=condition=ready --all pod
 
       - name: Run integration test
         run: |

CHANGELOG.md

@@ -62,6 +62,7 @@ For more information and how-to, see [RFC: Keep A Changelog](https://github.com/
 - Fix daemon-server `SetDNSServer` endpoint to validate provided server address [#4246](https://github.com/chaos-mesh/chaos-mesh/pull/4246)
 - Enable prometheus directive within CoreDNS [#4321](https://github.com/chaos-mesh/chaos-mesh/pull/4321)
 - Fix TTL configuration from environment variables [#4338](https://github.com/chaos-mesh/chaos-mesh/pull/4338)
+- Fix limit the number of events and minify the events [#4402](https://github.com/chaos-mesh/chaos-mesh/pull/4402)
 
 ### Security
 

controllers/common/records/controller.go

@@ -29,6 +29,7 @@ import (
 
 	"github.com/chaos-mesh/chaos-mesh/api/v1alpha1"
 	"github.com/chaos-mesh/chaos-mesh/controllers/chaosimpl/types"
+	"github.com/chaos-mesh/chaos-mesh/controllers/config"
 	"github.com/chaos-mesh/chaos-mesh/controllers/utils/recorder"
 	"github.com/chaos-mesh/chaos-mesh/pkg/selector"
 )
@@ -158,6 +159,9 @@ func (r *Reconciler) Reconcile(ctx context.Context, req ctrl.Request) (ctrl.Resu
 				// but the retry shouldn't block other resource process
 				idLogger.Error(err, "fail to apply chaos")
 				applyFailedEvent := newRecordEvent(v1alpha1.TypeFailed, v1alpha1.Apply, err.Error())
+				if len(records[index].Events) >= config.ControllerCfg.MaxEvents {
+					records[index].Events = records[index].Events[1:]
+				}
 				records[index].Events = append(records[index].Events, *applyFailedEvent)
 				r.Recorder.Event(obj, recorder.Failed{
 					Activity: "apply chaos",
@@ -172,6 +176,9 @@ func (r *Reconciler) Reconcile(ctx context.Context, req ctrl.Request) (ctrl.Resu
 			if record.Phase == v1alpha1.Injected {
 				records[index].InjectedCount++
 				applySucceedEvent := newRecordEvent(v1alpha1.TypeSucceeded, v1alpha1.Apply, "")
+				if len(records[index].Events) >= config.ControllerCfg.MaxEvents {
+					records[index].Events = records[index].Events[1:]
+				}
 				records[index].Events = append(records[index].Events, *applySucceedEvent)
 				r.Recorder.Event(obj, recorder.Applied{
 					Id: records[index].Id,
@@ -188,6 +195,9 @@ func (r *Reconciler) Reconcile(ctx context.Context, req ctrl.Request) (ctrl.Resu
 				// but the retry shouldn't block other resource process
 				idLogger.Error(err, "fail to recover chaos")
 				recoverFailedEvent := newRecordEvent(v1alpha1.TypeFailed, v1alpha1.Recover, err.Error())
+				if len(records[index].Events) >= config.ControllerCfg.MaxEvents {
+					records[index].Events = records[index].Events[1:]
+				}
 				records[index].Events = append(records[index].Events, *recoverFailedEvent)
 				r.Recorder.Event(obj, recorder.Failed{
 					Activity: "recover chaos",
@@ -202,6 +212,9 @@ func (r *Reconciler) Reconcile(ctx context.Context, req ctrl.Request) (ctrl.Resu
 			if record.Phase == v1alpha1.NotInjected {
 				records[index].RecoveredCount++
 				recoverSucceedEvent := newRecordEvent(v1alpha1.TypeSucceeded, v1alpha1.Recover, "")
+				if len(records[index].Events) >= config.ControllerCfg.MaxEvents {
+					records[index].Events = records[index].Events[1:]
+				}
 				records[index].Events = append(records[index].Events, *recoverSucceedEvent)
 				r.Recorder.Event(obj, recorder.Recovered{
 					Id: records[index].Id,

controllers/utils/chaosdaemon/chaosdaemon.go

@@ -65,7 +65,7 @@ func (b *ChaosDaemonClientBuilder) FindDaemonIP(ctx context.Context, pod *v1.Pod
 
 	daemonIP := findIPOnEndpoints(&endpoints, nodeName)
 	if len(daemonIP) == 0 {
-		return "", errors.Errorf("cannot find daemonIP on node %s in related Endpoints %v", nodeName, endpoints)
+		return "", errors.Errorf("cannot find daemonIP on node %s ", nodeName)
 	}
 
 	return daemonIP, nil

pkg/config/controller.go

@@ -114,6 +114,8 @@ type ChaosControllerConfig struct {
 	EnabledWebhooks    []string `envconfig:"ENABLED_WEBHOOKS" default:"*"`
 
 	LocalHelmChartPath string `envconfig:"LOCAL_HELM_CHART_PATH" default:""`
+
+	MaxEvents int `envconfig:"MAX_EVENTS" default:"100"`
 }
 
 // EnvironChaosController returns the settings from the environment.