Optmize refreshing SriovNetworkNodeState

There is a SriovNetworkNodeState watcher running.
It should be used to keep updated state of the object.
Requesting latest state of the object very often
overloads the kube-apiserver in big clusters.

Signed-of-by: Alan Kutniewski <[email protected]>

Author: alan-kut

pkg/daemon/daemon.go

@@ -14,6 +14,7 @@ import (
 	"sync"
 	"time"
 
+	v1 "github.com/k8snetworkplumbingwg/sriov-network-operator/pkg/client/listers/sriovnetwork/v1"
 	mcfgv1 "github.com/openshift/machine-config-operator/pkg/apis/machineconfiguration.openshift.io/v1"
 	daemonconsts "github.com/openshift/machine-config-operator/pkg/daemon/constants"
 	mcfginformers "github.com/openshift/machine-config-operator/pkg/generated/informers/externalversions"
@@ -98,6 +99,8 @@ type Daemon struct {
 
 	disableDrain bool
 
+	sriovNetworkNodeStateLister v1.SriovNetworkNodeStateLister
+
 	nodeLister listerv1.NodeLister
 
 	workqueue workqueue.RateLimitingInterface
@@ -220,6 +223,7 @@ func (dn *Daemon) Run(stopCh <-chan struct{}, exitCh <-chan error) error {
 		},
 	)
 
+	dn.sriovNetworkNodeStateLister = informerFactory.Sriovnetwork().V1().SriovNetworkNodeStates().Lister()
 	informer := informerFactory.Sriovnetwork().V1().SriovNetworkNodeStates().Informer()
 	informer.AddEventHandler(cache.ResourceEventHandlerFuncs{
 		AddFunc: dn.enqueueNodeState,
@@ -288,6 +292,13 @@ func (dn *Daemon) Run(stopCh <-chan struct{}, exitCh <-chan error) error {
 	}
 }
 
+func (dn *Daemon) getSriovNetworkNodeState() (*sriovnetworkv1.SriovNetworkNodeState, error) {
+	if dn.sriovNetworkNodeStateLister != nil {
+		return dn.sriovNetworkNodeStateLister.SriovNetworkNodeStates(vars.Namespace).Get(vars.NodeName)
+	}
+	return dn.client.SriovnetworkV1().SriovNetworkNodeStates(vars.Namespace).Get(context.Background(), vars.NodeName, metav1.GetOptions{ResourceVersion: "0"})
+}
+
 func (dn *Daemon) runWorker() {
 	for dn.processNextWorkItem() {
 	}
@@ -413,7 +424,7 @@ func (dn *Daemon) nodeStateSyncHandler() error {
 	// Get the latest NodeState
 	var latestState *sriovnetworkv1.SriovNetworkNodeState
 	var sriovResult = &systemd.SriovResult{SyncStatus: consts.SyncStatusSucceeded, LastSyncError: ""}
-	latestState, err = dn.client.SriovnetworkV1().SriovNetworkNodeStates(vars.Namespace).Get(context.Background(), vars.NodeName, metav1.GetOptions{})
+	latestState, err = dn.getSriovNetworkNodeState()
 	if err != nil {
 		log.Log.Error(err, "nodeStateSyncHandler(): Failed to fetch node state", "name", vars.NodeName)
 		return err
@@ -511,7 +522,7 @@ func (dn *Daemon) nodeStateSyncHandler() error {
 	// we need to load the latest status to our object
 	// if we don't do it we can have a race here where the user remove the virtual functions but the operator didn't
 	// trigger the refresh
-	updatedState, err := dn.client.SriovnetworkV1().SriovNetworkNodeStates(vars.Namespace).Get(context.Background(), vars.NodeName, metav1.GetOptions{})
+	updatedState, err := dn.getSriovNetworkNodeState()
 	if err != nil {
 		log.Log.Error(err, "nodeStateSyncHandler(): Failed to fetch node state", "name", vars.NodeName)
 		return err
@@ -1062,11 +1073,7 @@ func (dn *Daemon) drainNode() error {
 // TODO: move this to host interface
 func (dn *Daemon) tryCreateSwitchdevUdevRule() error {
 	log.Log.V(2).Info("tryCreateSwitchdevUdevRule()")
-	nodeState, nodeStateErr := dn.client.SriovnetworkV1().SriovNetworkNodeStates(vars.Namespace).Get(
-		context.Background(),
-		vars.NodeName,
-		metav1.GetOptions{},
-	)
+	nodeState, nodeStateErr := dn.getSriovNetworkNodeState()
 	if nodeStateErr != nil {
 		log.Log.Error(nodeStateErr, "could not fetch node state, skip updating switchdev udev rules", "name", vars.NodeName)
 		return nil

pkg/daemon/writer.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"os"
 	"path/filepath"
+	"reflect"
 	"time"
 
 	"github.com/pkg/errors"
@@ -88,6 +89,7 @@ func (w *NodeStateStatusWriter) RunOnce() error {
 func (w *NodeStateStatusWriter) Run(stop <-chan struct{}, refresh <-chan Message, syncCh chan<- struct{}) error {
 	log.Log.V(0).Info("Run(): start writer")
 	msg := Message{}
+	lastInterfaces := w.status.Interfaces
 
 	for {
 		select {
@@ -99,6 +101,7 @@ func (w *NodeStateStatusWriter) Run(stop <-chan struct{}, refresh <-chan Message
 			if err := w.pollNicStatus(); err != nil {
 				continue
 			}
+			lastInterfaces = w.status.Interfaces
 			_, err := w.setNodeStateStatus(msg)
 			if err != nil {
 				log.Log.Error(err, "Run() refresh: writing to node status failed")
@@ -109,7 +112,10 @@ func (w *NodeStateStatusWriter) Run(stop <-chan struct{}, refresh <-chan Message
 			if err := w.pollNicStatus(); err != nil {
 				continue
 			}
-			w.setNodeStateStatus(msg)
+			if !reflect.DeepEqual(lastInterfaces, w.status.Interfaces) {
+				lastInterfaces = w.status.Interfaces
+				w.setNodeStateStatus(msg)
+			}
 		}
 	}
 }
@@ -207,7 +213,7 @@ func (w *NodeStateStatusWriter) getNodeState() (*sriovnetworkv1.SriovNetworkNode
 	var lastErr error
 	var n *sriovnetworkv1.SriovNetworkNodeState
 	err := wait.PollImmediate(10*time.Second, 5*time.Minute, func() (bool, error) {
-		n, lastErr = w.client.SriovnetworkV1().SriovNetworkNodeStates(vars.Namespace).Get(context.Background(), vars.NodeName, metav1.GetOptions{})
+		n, lastErr = w.client.SriovnetworkV1().SriovNetworkNodeStates(vars.Namespace).Get(context.Background(), vars.NodeName, metav1.GetOptions{ResourceVersion: "0"})
 		if lastErr == nil {
 			return true, nil
 		}