From 3ee977a29b961b090bf94f65b3d4e5496c9d1be8 Mon Sep 17 00:00:00 2001
From: hzma <hzma@alauda.io>
Date: Wed, 9 Aug 2023 11:41:43 +0800
Subject: [PATCH] move unnecessary init process after startWorkers (#3124)

---
 pkg/controller/controller.go | 60 ++++++++++++++++++++----------------
 pkg/controller/gc.go         | 10 +-----
 pkg/controller/init.go       |  1 +
 3 files changed, 36 insertions(+), 35 deletions(-)

diff --git a/pkg/controller/controller.go b/pkg/controller/controller.go
index 0bbea18848a..cb228e49ed1 100644
--- a/pkg/controller/controller.go
+++ b/pkg/controller/controller.go
@@ -633,6 +633,8 @@ func NewController(config *Configuration) *Controller {
 // is closed, at which point it will shutdown the workqueue and wait for
 // workers to finish processing their current work items.
 func (c *Controller) Run(ctx context.Context) {
+	// The init process can only be placed here if the init process do really affect the normal process of controller, such as Nodes/Pods/Subnets...
+	// Otherwise, the init process should be placed after all workers have already started working
 	defer c.shutdown()
 	klog.Info("Starting OVN controller")
 
@@ -691,24 +693,10 @@ func (c *Controller) Run(ctx context.Context) {
 		util.LogFatalAndExit(err, "failed to initialize ipam")
 	}
 
-	if err := c.initNodeChassis(); err != nil {
-		util.LogFatalAndExit(err, "failed to initialize node chassis")
-	}
-
 	if err := c.initNodeRoutes(); err != nil {
 		util.LogFatalAndExit(err, "failed to initialize node routes")
 	}
 
-	if err := c.initDenyAllSecurityGroup(); err != nil {
-		util.LogFatalAndExit(err, "failed to initialize 'deny_all' security group")
-	}
-
-	// remove resources in ovndb that not exist any more in kubernetes resources
-	if err := c.gc(); err != nil {
-		util.LogFatalAndExit(err, "failed to run gc")
-	}
-
-	c.registerSubnetMetrics()
 	if err := c.initSyncCrdSubnets(); err != nil {
 		util.LogFatalAndExit(err, "failed to sync crd subnets")
 	}
@@ -716,24 +704,14 @@ func (c *Controller) Run(ctx context.Context) {
 		util.LogFatalAndExit(err, "failed to sync crd vlans")
 	}
 
-	if c.config.PodDefaultFipType == util.IptablesFip {
-		if err := c.initSyncCrdVpcNatGw(); err != nil {
-			util.LogFatalAndExit(err, "failed to sync crd vpc nat gateways")
-		}
-	}
-
-	if c.config.EnableLb {
-		if err := c.initVpcDnsConfig(); err != nil {
-			util.LogFatalAndExit(err, "failed to initialize vpc-dns")
-		}
-	}
-
 	if err := c.addNodeGwStaticRoute(); err != nil {
 		util.LogFatalAndExit(err, "failed to add static route for node gateway")
 	}
 
 	// start workers to do all the network operations
 	c.startWorkers(ctx)
+
+	c.initResourceOnce()
 	<-ctx.Done()
 	klog.Info("Shutting down workers")
 }
@@ -1035,3 +1013,33 @@ func (c *Controller) startWorkers(ctx context.Context) {
 		go wait.Until(c.runDelPodAnnotatedIptablesFipWorker, time.Second, ctx.Done())
 	}
 }
+
+func (c *Controller) initResourceOnce() {
+	c.registerSubnetMetrics()
+
+	if err := c.initNodeChassis(); err != nil {
+		util.LogFatalAndExit(err, "failed to initialize node chassis")
+	}
+
+	if err := c.initDenyAllSecurityGroup(); err != nil {
+		util.LogFatalAndExit(err, "failed to initialize 'deny_all' security group")
+	}
+
+	if c.config.PodDefaultFipType == util.IptablesFip {
+		if err := c.initSyncCrdVpcNatGw(); err != nil {
+			util.LogFatalAndExit(err, "failed to sync crd vpc nat gateways")
+		}
+	}
+
+	if c.config.EnableLb {
+		if err := c.initVpcDnsConfig(); err != nil {
+			util.LogFatalAndExit(err, "failed to initialize vpc-dns")
+		}
+	}
+
+	// remove resources in ovndb that not exist any more in kubernetes resources
+	// process gc at last in case of affecting other init process
+	if err := c.gc(); err != nil {
+		util.LogFatalAndExit(err, "failed to run gc")
+	}
+}
diff --git a/pkg/controller/gc.go b/pkg/controller/gc.go
index 920fe02c34a..35e965c7398 100644
--- a/pkg/controller/gc.go
+++ b/pkg/controller/gc.go
@@ -26,7 +26,7 @@ func (c *Controller) gc() error {
 		c.gcChassis,
 		c.gcLogicalSwitch,
 		c.gcCustomLogicalRouter,
-		c.gcLogicalSwitchPort,
+		// The lsp gc is processed periodically by markAndCleanLSP, will not gc lsp when init
 		c.gcLoadBalancer,
 		c.gcPortGroup,
 		c.gcStaticRoute,
@@ -277,14 +277,6 @@ func (c *Controller) gcVip() error {
 	return nil
 }
 
-func (c *Controller) gcLogicalSwitchPort() error {
-	klog.Info("start to gc logical switch port")
-	if err := c.markAndCleanLSP(); err != nil {
-		return err
-	}
-	return c.markAndCleanLSP()
-}
-
 func (c *Controller) markAndCleanLSP() error {
 	klog.V(4).Infof("start to gc logical switch ports")
 	pods, err := c.podsLister.List(labels.Everything())
diff --git a/pkg/controller/init.go b/pkg/controller/init.go
index b5d7a4286ce..99338eb752e 100644
--- a/pkg/controller/init.go
+++ b/pkg/controller/init.go
@@ -422,6 +422,7 @@ func (c *Controller) InitIPAM() error {
 			}
 		}
 	}
+
 	nodes, err := c.nodesLister.List(labels.Everything())
 	if err != nil {
 		klog.Errorf("failed to list nodes: %v", err)