From 41fb4a24f1acb3803c137a329011a25be75b0b42 Mon Sep 17 00:00:00 2001
From: Gallardot <gallardot@apache.org>
Date: Fri, 20 Sep 2024 12:14:41 +0800
Subject: [PATCH] fix: excessive records by adding the `MaxEvents` field
 (#4402)

Signed-off-by: Gallardot <gallardot@apache.org>
Signed-off-by: Yue Yang <g1enyy0ung@gmail.com>
Co-authored-by: Yue Yang <g1enyy0ung@gmail.com>
---
 CHANGELOG.md                                 |  1 +
 controllers/common/records/controller.go     | 13 +++++++++++++
 controllers/utils/chaosdaemon/chaosdaemon.go |  2 +-
 pkg/config/controller.go                     |  2 ++
 4 files changed, 17 insertions(+), 1 deletion(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 58e4c5f633..c83c3c2b30 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -67,6 +67,7 @@ For more information and how-to, see [RFC: Keep A Changelog](https://github.com/
 - Fix TTL configuration from environment variables [#4338](https://github.com/chaos-mesh/chaos-mesh/pull/4338)
 - Fix dashboard panic while replacing query namespace with targetNamespace in namespace scoped mode [#4409](https://github.com/chaos-mesh/chaos-mesh/issues/4409)
 - Fix incorrect mmap args for IOChaos [#3680](https://github.com/chaos-mesh/chaos-mesh/issues/3680)
+- Fix excessive records by adding the `MaxEvents` field [#4402](https://github.com/chaos-mesh/chaos-mesh/pull/4402)
 - Fix chaos controller can't find daemonIP over 1000 nodes using endpoints [#4421](https://github.com/chaos-mesh/chaos-mesh/pull/4421)
 - Minor fixes in certificates to make them ArgoCD friendly [#4482](https://github.com/chaos-mesh/chaos-mesh/pull/4482)
 
diff --git a/controllers/common/records/controller.go b/controllers/common/records/controller.go
index be913279e6..3e4ec9dae7 100644
--- a/controllers/common/records/controller.go
+++ b/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,
diff --git a/controllers/utils/chaosdaemon/chaosdaemon.go b/controllers/utils/chaosdaemon/chaosdaemon.go
index 47338ed03c..90c784d78a 100644
--- a/controllers/utils/chaosdaemon/chaosdaemon.go
+++ b/controllers/utils/chaosdaemon/chaosdaemon.go
@@ -67,7 +67,7 @@ func (b *ChaosDaemonClientBuilder) FindDaemonIP(ctx context.Context, pod *v1.Pod
 
 	daemonIP := findIPOnEndpointSlice(&endpointSliceList, nodeName)
 	if len(daemonIP) == 0 {
-		return "", errors.Errorf("cannot find daemonIP on node %s in endpointSliceList %v", nodeName, endpointSliceList)
+		return "", errors.Errorf("cannot find daemonIP on node %s", nodeName)
 	}
 
 	return daemonIP, nil
diff --git a/pkg/config/controller.go b/pkg/config/controller.go
index db3ec5acd2..b22d9d27c1 100644
--- a/pkg/config/controller.go
+++ b/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.