Update Kube-OVN v1.13.13 and enable db healthcheck (#1047)

This PR updates Kube-OVN to the latest version and also includes fix
https://github.com/kubeovn/kube-ovn/pull/5294

Ref
https://github.com/kubeovn/kube-ovn/issues/5125#issuecomment-2921920661

Signed-off-by: Andrei Kvapil <kvapss@gmail.com>

packages/system/kubeovn/Makefile

@@ -15,6 +15,7 @@ update:
 	patch --no-backup-if-mismatch -p4 < patches/mtu.diff
 	version=$$(awk '$$1 == "version:" {print $$2}' charts/kube-ovn/Chart.yaml) && \
 	sed -i "s/ARG VERSION=.*/ARG VERSION=$${version}/" images/kubeovn/Dockerfile
+	sed -i "s/ARG TAG=.*/ARG TAG=$${version}/" images/kubeovn/Dockerfile
 
 image:
 	docker buildx build images/kubeovn \

packages/system/kubeovn/charts/kube-ovn/Chart.yaml

@@ -15,12 +15,12 @@ type: application
 # This is the chart version. This version number should be incremented each time you make changes
 # to the chart and its templates, including the app version.
 # Versions are expected to follow Semantic Versioning (https://semver.org/)
-version: v1.13.11
+version: v1.13.13
 
 # This is the version number of the application being deployed. This version number should be
 # incremented each time you make changes to the application. Versions are not expected to
 # follow Semantic Versioning. They should reflect the version the application is using.
 # It is recommended to use it with quotes.
-appVersion: "1.13.11"
+appVersion: "1.13.13"
 
 kubeVersion: ">= 1.23.0-0"

packages/system/kubeovn/charts/kube-ovn/values.yaml

@@ -10,7 +10,7 @@ global:
       repository: kube-ovn
       dpdkRepository: kube-ovn-dpdk
       vpcRepository: vpc-nat-gateway
-      tag: v1.13.11
+      tag: v1.13.13
       support_arm: true
       thirdparty: true
 

packages/system/kubeovn/images/kubeovn/Dockerfile

@@ -1,2 +1,47 @@
-ARG VERSION=v1.13.11
-FROM kubeovn/kube-ovn:${VERSION}
+# syntax = docker/dockerfile:experimental
+ARG VERSION=v1.13.13
+ARG BASE_TAG=$VERSION
+
+FROM golang:1.23-bookworm as builder
+
+ARG TAG=v1.13.13
+RUN git clone --branch ${TAG} --depth 1 https://github.com/kubeovn/kube-ovn /source
+
+WORKDIR /source
+
+COPY patches /patches
+RUN git apply /patches/*.diff
+RUN make build-go
+
+WORKDIR /source/dist/images
+
+# imported from https://github.com/kubeovn/kube-ovn/blob/master/dist/images/Dockerfile
+FROM kubeovn/kube-ovn-base:$BASE_TAG AS setcap
+
+COPY --from=builder /source/dist/images/*.sh /kube-ovn/
+COPY --from=builder /source/dist/images/kubectl-ko /kube-ovn/kubectl-ko
+COPY --from=builder /source/dist/images/01-kube-ovn.conflist /kube-ovn/01-kube-ovn.conflist
+
+COPY --from=builder /source/dist/images/kube-ovn /kube-ovn/kube-ovn
+COPY --from=builder /source/dist/images/kube-ovn-cmd /kube-ovn/kube-ovn-cmd
+COPY --from=builder /source/dist/images/kube-ovn-daemon /kube-ovn/kube-ovn-daemon
+COPY --from=builder /source/dist/images/kube-ovn-controller /kube-ovn/kube-ovn-controller
+RUN ln -s /kube-ovn/kube-ovn-cmd /kube-ovn/kube-ovn-monitor && \
+    ln -s /kube-ovn/kube-ovn-cmd /kube-ovn/kube-ovn-speaker && \
+    ln -s /kube-ovn/kube-ovn-cmd /kube-ovn/kube-ovn-webhook && \
+    ln -s /kube-ovn/kube-ovn-cmd /kube-ovn/kube-ovn-leader-checker && \
+    ln -s /kube-ovn/kube-ovn-cmd /kube-ovn/kube-ovn-ic-controller && \
+    ln -s /kube-ovn/kube-ovn-controller /kube-ovn/kube-ovn-pinger && \
+    setcap CAP_NET_BIND_SERVICE+eip /kube-ovn/kube-ovn-cmd && \
+    setcap CAP_NET_RAW,CAP_NET_BIND_SERVICE+eip /kube-ovn/kube-ovn-controller && \
+    setcap CAP_NET_ADMIN,CAP_NET_RAW,CAP_NET_BIND_SERVICE,CAP_SYS_ADMIN+eip /kube-ovn/kube-ovn-daemon
+
+FROM kubeovn/kube-ovn-base:$BASE_TAG
+
+COPY --chmod=0644 --from=builder /source/dist/images/logrotate/* /etc/logrotate.d/
+COPY --from=builder /source/dist/images/grace_stop_ovn_controller /usr/share/ovn/scripts/grace_stop_ovn_controller
+
+COPY --from=setcap /kube-ovn /kube-ovn
+RUN /kube-ovn/iptables-wrapper-installer.sh --no-sanity-check
+
+WORKDIR /kube-ovn

packages/system/kubeovn/images/kubeovn/patches/5294-db-healthcheck.diff

@@ -0,0 +1,168 @@
+diff --git a/mocks/pkg/ovs/interface.go b/mocks/pkg/ovs/interface.go
+index e9c472bee..59076f9ed 100644
+--- a/mocks/pkg/ovs/interface.go
++++ b/mocks/pkg/ovs/interface.go
+@@ -10,6 +10,7 @@
+ package ovs
+ 
+ import (
++	context "context"
+ 	reflect "reflect"
+ 
+ 	v1 "github.com/kubeovn/kube-ovn/pkg/apis/kubeovn/v1"
+@@ -3322,6 +3323,20 @@ func (mr *MockNbClientMockRecorder) DeleteSecurityGroup(sgName any) *gomock.Call
+ 	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "DeleteSecurityGroup", reflect.TypeOf((*MockNbClient)(nil).DeleteSecurityGroup), sgName)
+ }
+ 
++// Echo mocks base method.
++func (m *MockNbClient) Echo(arg0 context.Context) error {
++	m.ctrl.T.Helper()
++	ret := m.ctrl.Call(m, "Echo", arg0)
++	ret0, _ := ret[0].(error)
++	return ret0
++}
++
++// Echo indicates an expected call of Echo.
++func (mr *MockNbClientMockRecorder) Echo(arg0 any) *gomock.Call {
++	mr.mock.ctrl.T.Helper()
++	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "Echo", reflect.TypeOf((*MockNbClient)(nil).Echo), arg0)
++}
++
+ // EnablePortLayer2forward mocks base method.
+ func (m *MockNbClient) EnablePortLayer2forward(lspName string) error {
+ 	m.ctrl.T.Helper()
+@@ -4770,6 +4785,20 @@ func (mr *MockSbClientMockRecorder) GetAllChassisByHost(nodeName any) *gomock.Ca
+ 	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "GetAllChassisByHost", reflect.TypeOf((*MockSbClient)(nil).GetAllChassisByHost), nodeName)
+ }
+ 
++// Echo mocks base method.
++func (m *MockSbClient) Echo(arg0 context.Context) error {
++	m.ctrl.T.Helper()
++	ret := m.ctrl.Call(m, "Echo", arg0)
++	ret0, _ := ret[0].(error)
++	return ret0
++}
++
++// Echo indicates an expected call of Echo.
++func (mr *MockSbClientMockRecorder) Echo(arg0 any) *gomock.Call {
++	mr.mock.ctrl.T.Helper()
++	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "Echo", reflect.TypeOf((*MockSbClient)(nil).Echo), arg0)
++}
++
+ // GetChassis mocks base method.
+ func (m *MockSbClient) GetChassis(chassisName string, ignoreNotFound bool) (*ovnsb.Chassis, error) {
+ 	m.ctrl.T.Helper()
+@@ -4915,6 +4944,20 @@ func (m *MockCommon) EXPECT() *MockCommonMockRecorder {
+ 	return m.recorder
+ }
+ 
++// Echo mocks base method.
++func (m *MockCommon) Echo(arg0 context.Context) error {
++	m.ctrl.T.Helper()
++	ret := m.ctrl.Call(m, "Echo", arg0)
++	ret0, _ := ret[0].(error)
++	return ret0
++}
++
++// Echo indicates an expected call of Echo.
++func (mr *MockCommonMockRecorder) Echo(arg0 any) *gomock.Call {
++	mr.mock.ctrl.T.Helper()
++	return mr.mock.ctrl.RecordCallWithMethodType(mr.mock, "Echo", reflect.TypeOf((*MockCommon)(nil).Echo), arg0)
++}
++
+ // GetEntityInfo mocks base method.
+ func (m *MockCommon) GetEntityInfo(entity any) error {
+ 	m.ctrl.T.Helper()
+diff --git a/pkg/controller/controller.go b/pkg/controller/controller.go
+index cb594a4c8..a2a88eb06 100644
+--- a/pkg/controller/controller.go
++++ b/pkg/controller/controller.go
+@@ -268,6 +268,9 @@ type Controller struct {
+ 	cmInformerFactory      kubeinformers.SharedInformerFactory
+ 	kubeovnInformerFactory kubeovninformer.SharedInformerFactory
+ 	anpInformerFactory     anpinformer.SharedInformerFactory
++
++	// Database health check
++	dbFailureCount int
+ }
+ 
+ func newTypedRateLimitingQueue[T comparable](name string, rateLimiter workqueue.TypedRateLimiter[T]) workqueue.TypedRateLimitingInterface[T] {
+@@ -944,6 +947,48 @@ func (c *Controller) Run(ctx context.Context) {
+ 	klog.Info("Shutting down workers")
+ }
+ 
++func (c *Controller) dbStatus() {
++	const maxFailures = 5
++
++	done := make(chan error, 2)
++	go func() {
++		done <- c.OVNNbClient.Echo(context.Background())
++	}()
++	go func() {
++		done <- c.OVNSbClient.Echo(context.Background())
++	}()
++
++	resultsReceived := 0
++	timeout := time.After(time.Duration(c.config.OvnTimeout) * time.Second)
++
++	for resultsReceived < 2 {
++		select {
++		case err := <-done:
++			resultsReceived++
++			if err != nil {
++				c.dbFailureCount++
++				klog.Errorf("OVN database echo failed (%d/%d): %v", c.dbFailureCount, maxFailures, err)
++				if c.dbFailureCount >= maxFailures {
++					util.LogFatalAndExit(err, "OVN database connection failed after %d attempts", maxFailures)
++				}
++				return
++			}
++		case <-timeout:
++			c.dbFailureCount++
++			klog.Errorf("OVN database echo timeout (%d/%d) after %ds", c.dbFailureCount, maxFailures, c.config.OvnTimeout)
++			if c.dbFailureCount >= maxFailures {
++				util.LogFatalAndExit(nil, "OVN database connection timeout after %d attempts", maxFailures)
++			}
++			return
++		}
++	}
++
++	if c.dbFailureCount > 0 {
++		klog.Infof("OVN database connection recovered after %d failures", c.dbFailureCount)
++		c.dbFailureCount = 0
++	}
++}
++
+ func (c *Controller) shutdown() {
+ 	utilruntime.HandleCrash()
+ 
+@@ -1277,6 +1322,8 @@ func (c *Controller) startWorkers(ctx context.Context) {
+ 	if c.config.EnableLiveMigrationOptimize {
+ 		go wait.Until(runWorker("add/update vmiMigration ", c.addOrUpdateVMIMigrationQueue, c.handleAddOrUpdateVMIMigration), 50*time.Millisecond, ctx.Done())
+ 	}
++
++	go wait.Until(c.dbStatus, 15*time.Second, ctx.Done())
+ }
+ 
+ func (c *Controller) allSubnetReady(subnets ...string) (bool, error) {
+diff --git a/pkg/ovs/interface.go b/pkg/ovs/interface.go
+index df6907c4d..5e70dd6c7 100644
+--- a/pkg/ovs/interface.go
++++ b/pkg/ovs/interface.go
+@@ -1,6 +1,8 @@
+ package ovs
+ 
+ import (
++	"context"
++
+ 	netv1 "k8s.io/api/networking/v1"
+ 
+ 	"github.com/ovn-org/libovsdb/ovsdb"
+@@ -235,6 +237,7 @@ type SbClient interface {
+ }
+ 
+ type Common interface {
++	Echo(context.Context) error
+ 	Transact(method string, operations []ovsdb.Operation) error
+ 	GetEntityInfo(entity interface{}) error
+ }