fix: delete chassis from sb when delete node

dist/images/start-controller.sh

@@ -1,4 +1,8 @@
 #!/usr/bin/env bash
 set -euo pipefail
 export OVN_NB_DAEMON=$(ovn-nbctl --db=tcp:["${OVN_NB_SERVICE_HOST}"]:"${OVN_NB_SERVICE_PORT}" --pidfile --detach --overwrite-pidfile)
-exec ./kube-ovn-controller --ovn-nb-host="${OVN_NB_SERVICE_HOST}" --ovn-nb-port="${OVN_NB_SERVICE_PORT}" $@
+exec ./kube-ovn-controller --ovn-nb-host="${OVN_NB_SERVICE_HOST}" \
+                           --ovn-nb-port="${OVN_NB_SERVICE_PORT}" \
+                           --ovn-sb-host="${OVN_SB_SERVICE_HOST}" \
+                           --ovn-sb-port="${OVN_SB_SERVICE_PORT}" \
+                           $@

pkg/controller/config.go

@@ -19,7 +19,9 @@ type Configuration struct {
 	OvnNbSocket    string
 	OvnNbHost      string
 	OvnNbPort      int
-	OvnNbTimeout   int
+	OvnSbHost      string
+	OvnSbPort      int
+	OvnTimeout     int
 	KubeConfigFile string
 	KubeClient     kubernetes.Interface
 	KubeOvnClient  clientset.Interface
@@ -51,7 +53,9 @@ func ParseFlags() (*Configuration, error) {
 		argOvnNbSocket    = pflag.String("ovn-nb-socket", "", "The ovn-nb socket file. (If not set use ovn-nb-address)")
 		argOvnNbHost      = pflag.String("ovn-nb-host", "0.0.0.0", "The ovn-nb host address. (If not set use ovn-nb-socket)")
 		argOvnNbPort      = pflag.Int("ovn-nb-port", 6641, "")
-		argOvnNbTimeout   = pflag.Int("ovn-nb-timeout", 30, "")
+		argOvnSbHost      = pflag.String("ovn-sb-host", "0.0.0.0", "The ovn-sb host address.")
+		argOvnSbPort      = pflag.Int("ovn-sb-port", 6642, "")
+		argOvnTimeout     = pflag.Int("ovn-timeout", 30, "")
 		argKubeConfigFile = pflag.String("kubeconfig", "", "Path to kubeconfig file with authorization and master location information. If not set use the inCluster token.")
 
 		argDefaultLogicalSwitch = pflag.String("default-ls", "ovn-default", "The default logical switch name, default: ovn-default")
@@ -93,7 +97,9 @@ func ParseFlags() (*Configuration, error) {
 		OvnNbSocket:            *argOvnNbSocket,
 		OvnNbHost:              *argOvnNbHost,
 		OvnNbPort:              *argOvnNbPort,
-		OvnNbTimeout:           *argOvnNbTimeout,
+		OvnSbHost:              *argOvnSbHost,
+		OvnSbPort:              *argOvnSbPort,
+		OvnTimeout:             *argOvnTimeout,
 		KubeConfigFile:         *argKubeConfigFile,
 		DefaultLogicalSwitch:   *argDefaultLogicalSwitch,
 		DefaultCIDR:            *argDefaultCIDR,

pkg/controller/controller.go

@@ -112,7 +112,7 @@ func NewController(config *Configuration) *Controller {
 
 	controller := &Controller{
 		config:    config,
-		ovnClient: ovs.NewClient(config.OvnNbHost, config.OvnNbPort, config.OvnNbTimeout, "", 0, config.ClusterRouter, config.ClusterTcpLoadBalancer, config.ClusterUdpLoadBalancer, config.NodeSwitch, config.NodeSwitchCIDR),
+		ovnClient: ovs.NewClient(config.OvnNbHost, config.OvnNbPort, config.OvnTimeout, config.OvnSbHost, config.OvnSbPort, config.ClusterRouter, config.ClusterTcpLoadBalancer, config.ClusterUdpLoadBalancer, config.NodeSwitch, config.NodeSwitchCIDR),
 		ipam:      ovnipam.NewIPAM(),
 
 		subnetsLister:           subnetInformer.Lister(),

pkg/controller/node.go

@@ -308,6 +308,10 @@ func (c *Controller) handleDeleteNode(key string) error {
 		klog.Errorf("failed to delete node switch port node-%s %v", key, err)
 		return err
 	}
+	if err := c.ovnClient.DeleteChassis(key); err != nil {
+		klog.Errorf("failed to delete chassis for node %s %v", key, err)
+		return err
+	}
 
 	if err := c.config.KubeOvnClient.KubeovnV1().IPs().Delete(portName, &metav1.DeleteOptions{}); err != nil && !k8serrors.IsNotFound(err) {
 		return err

pkg/ovs/ovn-nbctl.go

@@ -16,7 +16,7 @@ import (
 
 func (c Client) ovnNbCommand(cmdArgs ...string) (string, error) {
 	start := time.Now()
-	cmdArgs = append([]string{fmt.Sprintf("--timeout=%d", c.OvnNbTimeout)}, cmdArgs...)
+	cmdArgs = append([]string{fmt.Sprintf("--timeout=%d", c.OvnTimeout)}, cmdArgs...)
 	raw, err := exec.Command(OvnNbCtl, cmdArgs...).CombinedOutput()
 	elapsed := float64((time.Since(start)) / time.Millisecond)
 	klog.Infof("%s command %s in %vms", OvnNbCtl, strings.Join(cmdArgs, " "), elapsed)

pkg/ovs/ovn-sbctl.go

@@ -0,0 +1,38 @@
+package ovs
+
+import (
+	"fmt"
+	"os/exec"
+	"strings"
+	"time"
+
+	"k8s.io/klog"
+)
+
+func (c Client) ovnSbCommand(cmdArgs ...string) (string, error) {
+	start := time.Now()
+	cmdArgs = append([]string{fmt.Sprintf("--timeout=%d", c.OvnTimeout), fmt.Sprintf("--db=%s", c.OvnSbAddress)}, cmdArgs...)
+	raw, err := exec.Command(OvnSbCtl, cmdArgs...).CombinedOutput()
+	elapsed := float64((time.Since(start)) / time.Millisecond)
+	klog.Infof("%s command %s in %vms", OvnSbCtl, strings.Join(cmdArgs, " "), elapsed)
+	if err != nil {
+		return "", fmt.Errorf("%s, %q", raw, err)
+	}
+	return trimCommandOutput(raw), nil
+}
+
+func (c Client) DeleteChassis(node string) error {
+	output, err := c.ovnSbCommand("--format=csv", "--no-heading", "--data=bare", "--columns=name", "find", "chassis", fmt.Sprintf("hostname=%s", node))
+	if err != nil {
+		return fmt.Errorf("failed to find node chassis %s, %v", node, err)
+	}
+	for _, chassis := range strings.Split(output, "\n") {
+		chassis = strings.TrimSpace(chassis)
+		if len(chassis) > 0 {
+			if _, err := c.ovnSbCommand("chassis-del", strings.TrimSpace(chassis)); err != nil {
+				return err
+			}
+		}
+	}
+	return nil
+}

pkg/ovs/ovn.go

@@ -13,7 +13,7 @@ var (
 // Client is the ovn client
 type Client struct {
 	OvnNbAddress           string
-	OvnNbTimeout           int
+	OvnTimeout             int
 	OvnSbAddress           string
 	ClusterRouter          string
 	ClusterTcpLoadBalancer string
@@ -24,6 +24,7 @@ type Client struct {
 
 const (
 	OvnNbCtl    = "ovn-nbctl"
+	OvnSbCtl    = "ovn-sbctl"
 	MayExist    = "--may-exist"
 	IfExists    = "--if-exists"
 	Policy      = "--policy"
@@ -36,7 +37,7 @@ func NewClient(ovnNbHost string, ovnNbPort int, ovnNbTimeout int, ovnSbHost stri
 	return &Client{
 		OvnNbAddress:           fmt.Sprintf("tcp:%s:%d", ovnNbHost, ovnNbPort),
 		OvnSbAddress:           fmt.Sprintf("tcp:%s:%d", ovnSbHost, ovnSbPort),
-		OvnNbTimeout:           ovnNbTimeout,
+		OvnTimeout:             ovnNbTimeout,
 		ClusterRouter:          clusterRouter,
 		ClusterTcpLoadBalancer: clusterTcpLoadBalancer,
 		ClusterUdpLoadBalancer: clusterUdpLoadBalancer,