feat(monitor): optimize monitor (#2141)

1. fix bug in prometheus rule 2. add ttl for messagerequest Co-authored-by: willzgli <willzgli@tencent.com>
tkestack · Nov 3, 2022 · f42cc9a · f42cc9a
1 parent 13b64d0
commit f42cc9a
Show file tree

Hide file tree

Showing 7 changed files with 41 additions and 14 deletions.
diff --git a/cmd/tke-installer/app/installer/manifests/tke-notify-api/tke-notify-api.yaml b/cmd/tke-installer/app/installer/manifests/tke-notify-api/tke-notify-api.yaml
@@ -141,6 +141,11 @@ data:
       api_server = "https://tke-platform-api"
       api_server_client_config = "/app/conf/tke-platform-config.yaml"
 
+    [features]
+       # Valid time units are "ns", "us" (or "µs"), "ms", "s", "m", "h"
+       messagerequest_ttl = "1h"
+       message_ttl = "360h"
+
   tke-auth-webhook.yaml: |
     apiVersion: v1
     kind: Config

diff --git a/cmd/tke-notify-api/app/config/config.go b/cmd/tke-notify-api/app/config/config.go
@@ -57,6 +57,7 @@ type Config struct {
 	PrivilegedUsername             string
 	ExternalHost                   string
 	ExternalPort                   int
+	MessageRequestTTL              time.Duration
 	MessageTTL                     time.Duration
 }
 
@@ -139,6 +140,7 @@ func CreateConfigFromOptions(serverName string, opts *options.Options) (*Config,
 		PrivilegedUsername:             opts.Authentication.PrivilegedUsername,
 		ExternalHost:                   opts.Generic.ExternalHost,
 		ExternalPort:                   opts.Generic.ExternalPort,
+		MessageRequestTTL:              opts.FeatureOptions.MessageRequestTTL,
 		MessageTTL:                     opts.FeatureOptions.MessageTTL,
 	}, nil
 }
diff --git a/cmd/tke-notify-api/app/options/feature.go b/cmd/tke-notify-api/app/options/feature.go
@@ -19,34 +19,48 @@
 package options
 
 import (
+	"time"
+
 	"github.com/spf13/pflag"
 	"github.com/spf13/viper"
-	"time"
 )
 
 const (
-	flagMessageTTL    = "message-ttl"
-	configMessageTTL  = "features.message_ttl"
-	defaultMessageTTL = time.Hour * 24 * 30
+	flagMessageTTL           = "message-ttl"
+	flagMessageRequestTTL    = "messagerequest-ttl"
+	configMessageRequestTTL  = "features.messagerequest_ttl"
+	configMessageTTL         = "features.message_ttl"
+	defaultMessageRequestTTL = time.Hour
+	defaultMessageTTL        = time.Hour * 24 * 30
 )
 
 type FeatureOptions struct {
-	MessageTTL time.Duration
+	MessageRequestTTL time.Duration
+	MessageTTL        time.Duration
 }
 
 func NewFeatureOptions() *FeatureOptions {
-	return &FeatureOptions{MessageTTL: defaultMessageTTL}
+	return &FeatureOptions{
+		MessageRequestTTL: defaultMessageRequestTTL,
+		MessageTTL:        defaultMessageTTL,
+	}
 }
 
 func (o *FeatureOptions) AddFlags(fs *pflag.FlagSet) {
+	fs.DurationVar(&o.MessageRequestTTL, flagMessageRequestTTL, o.MessageRequestTTL,
+		"How long to retain messagerequests")
 	fs.DurationVar(&o.MessageTTL, flagMessageTTL, o.MessageTTL,
-		"How long to retain messages and messagerequests")
+		"How long to retain messages")
+	_ = viper.BindPFlag(configMessageRequestTTL, fs.Lookup(flagMessageRequestTTL))
 	_ = viper.BindPFlag(configMessageTTL, fs.Lookup(flagMessageTTL))
+
 }
 
 func (o *FeatureOptions) ApplyFlags() []error {
 	var errs []error
 
+	// Valid time units are "ns", "us" (or "µs"), "ms", "s", "m", "h".
+	o.MessageRequestTTL = viper.GetDuration(configMessageRequestTTL)
 	o.MessageTTL = viper.GetDuration(configMessageTTL)
 
 	return errs

diff --git a/cmd/tke-notify-api/app/server.go b/cmd/tke-notify-api/app/server.go
@@ -99,6 +99,7 @@ func createAPIServerConfig(cfg *config.Config) *apiserver.Config {
 			StorageFactory:          cfg.StorageFactory,
 			APIResourceConfigSource: cfg.StorageFactory.APIResourceConfigSource,
 			PrivilegedUsername:      cfg.PrivilegedUsername,
+			MessageRequestTTL:       cfg.MessageRequestTTL,
 			MessageTTL:              cfg.MessageTTL,
 		},
 	}

diff --git a/pkg/monitor/controller/prometheus/yamls.go b/pkg/monitor/controller/prometheus/yamls.go
@@ -668,7 +668,7 @@ groups:
     expr: sum(k8s_container_fs_write_times) without (container_name,container_id)
 
   - record: k8s_pod_status_ready
-    expr: sum(kube_pod_status_ready{condition="true"}) by (namespace,pod_name) *  on(namespace, pod_name) group_left(workload_kind,workload_name,node, node_role)  __pod_info2
+    expr: sum(kube_pod_status_phase{phase=~"Running|Succeeded"}) by (namespace,pod_name) *  on(namespace, pod_name) group_left(workload_kind,workload_name,node, node_role)  __pod_info2
 
   - record: k8s_pod_restart_total
     expr: sum(idelta(kube_pod_container_status_restarts_total [2m])) by (namespace,pod_name) *  on(namespace, pod_name) group_left(workload_kind,workload_name,node, node_role)  __pod_info2
@@ -1240,8 +1240,8 @@ func configForAlertManager(webhookAddr string, repeatInterval string) string {
 
     route:
       group_by: ['alertname','alarmPolicyName','version']
-      group_wait: 1s
-      group_interval: 1s
+      group_wait: 30s
+      group_interval: 5m
       repeat_interval: %s
       receiver: 'web.hook'
       routes:

diff --git a/pkg/notify/apiserver/apiserver.go b/pkg/notify/apiserver/apiserver.go
@@ -19,10 +19,11 @@
 package apiserver
 
 import (
+	"time"
+
 	"k8s.io/apiserver/pkg/registry/generic"
 	genericapiserver "k8s.io/apiserver/pkg/server"
 	serverstorage "k8s.io/apiserver/pkg/server/storage"
-	"time"
 	versionedinformers "tkestack.io/tke/api/client/informers/externalversions"
 	notifyv1 "tkestack.io/tke/api/notify/v1"
 	"tkestack.io/tke/pkg/apiserver/storage"
@@ -37,6 +38,7 @@ type ExtraConfig struct {
 	StorageFactory          serverstorage.StorageFactory
 	VersionedInformers      versionedinformers.SharedInformerFactory
 	PrivilegedUsername      string
+	MessageRequestTTL       time.Duration
 	MessageTTL              time.Duration
 }
 
@@ -92,6 +94,7 @@ func (c completedConfig) New(delegationTarget genericapiserver.DelegationTarget)
 		&notifyrest.StorageProvider{
 			LoopbackClientConfig: c.GenericConfig.LoopbackClientConfig,
 			PrivilegedUsername:   c.ExtraConfig.PrivilegedUsername,
+			MessageRequestTTL:    c.ExtraConfig.MessageRequestTTL,
 			MessageTTL:           c.ExtraConfig.MessageTTL,
 		},
 	}

diff --git a/pkg/notify/registry/rest/rest.go b/pkg/notify/registry/rest/rest.go
@@ -19,15 +19,16 @@
 package rest
 
 import (
+	"time"
+
 	"k8s.io/apiserver/pkg/registry/generic"
 	"k8s.io/apiserver/pkg/registry/rest"
 	genericserver "k8s.io/apiserver/pkg/server"
 	serverstorage "k8s.io/apiserver/pkg/server/storage"
 	restclient "k8s.io/client-go/rest"
-	"time"
 	notifyinternalclient "tkestack.io/tke/api/client/clientset/internalversion/typed/notify/internalversion"
 	"tkestack.io/tke/api/notify"
-	"tkestack.io/tke/api/notify/v1"
+	v1 "tkestack.io/tke/api/notify/v1"
 	"tkestack.io/tke/pkg/apiserver/storage"
 	channelstorage "tkestack.io/tke/pkg/notify/registry/channel/storage"
 	configmapstorage "tkestack.io/tke/pkg/notify/registry/configmap/storage"
@@ -43,6 +44,7 @@ import (
 type StorageProvider struct {
 	LoopbackClientConfig *restclient.Config
 	PrivilegedUsername   string
+	MessageRequestTTL    time.Duration
 	MessageTTL           time.Duration
 }
 
@@ -85,7 +87,7 @@ func (s *StorageProvider) v1Storage(apiResourceConfigSource serverstorage.APIRes
 		storageMap["messages"] = messageREST.Message
 		storageMap["messages/status"] = messageREST.Status
 
-		messageRequestREST := messagerequeststorage.NewStorage(restOptionsGetter, notifyClient, s.PrivilegedUsername, s.MessageTTL)
+		messageRequestREST := messagerequeststorage.NewStorage(restOptionsGetter, notifyClient, s.PrivilegedUsername, s.MessageRequestTTL)
 		storageMap["messagerequests"] = messageRequestREST.MessageRequest
 		storageMap["messagerequests/status"] = messageRequestREST.Status