ovs: fix memory leak in qos

Author: zhangzujian

.github/workflows/build-x86-image.yaml

@@ -271,7 +271,7 @@ jobs:
       - build-kube-ovn
       - build-e2e-binaries
     runs-on: ubuntu-22.04
-    timeout-minutes: 15
+    timeout-minutes: 20
     strategy:
       fail-fast: false
       matrix:

dist/images/Dockerfile.base

@@ -21,7 +21,9 @@ RUN cd /usr/src/ && \
     # ovsdb-tool: add optional server id parameter for "join-cluster" command
     curl -s https://github.com/kubeovn/ovs/commit/ebf61515da71fa2e23125a92859fbdb96dcbffe7.patch | git apply && \
     # Add jitter parameter patch for netem qos
-    curl -s https://github.com/kubeovn/ovs/commit/2eaaf89fbf3ee2172719ed10d045fd79900edc8e.patch | git apply
+    curl -s https://github.com/kubeovn/ovs/commit/2eaaf89fbf3ee2172719ed10d045fd79900edc8e.patch | git apply && \
+    # fix memory leak in qos
+    curl -s https://github.com/kubeovn/ovs/commit/6a4dd2f4b9311a227cc26fef7c398ae9b241311b.patch | git apply
 
 RUN cd /usr/src/ && git clone -b branch-22.12 --depth=1 https://github.com/ovn-org/ovn.git && \
     cd ovn && \
@@ -69,14 +71,12 @@ RUN mkdir /packages/ && \
 
 FROM ubuntu:22.04
 
-ARG DEBUG=false
 ARG DEBIAN_FRONTEND=noninteractive
 RUN apt update && apt upgrade -y && apt install ca-certificates python3 hostname libunwind8 netbase \
         ethtool iproute2 ncat libunbound-dev procps libatomic1 kmod iptables python3-netifaces python3-sortedcontainers \
         tcpdump ipset curl uuid-runtime openssl inetutils-ping arping ndisc6 \
         logrotate dnsutils net-tools strongswan strongswan-pki libcharon-extra-plugins \
         libcharon-extauth-plugins libstrongswan-extra-plugins libstrongswan-standard-plugins -y --no-install-recommends && \
-        if [ "${DEBUG}" = "true" ]; then apt install -y --no-install-recommends valgrind; fi && \
         rm -rf /var/lib/apt/lists/* && \
         rm -rf /etc/localtime
 
@@ -94,12 +94,6 @@ ENV KUBE_VERSION="v1.27.2"
 RUN curl -L https://dl.k8s.io/${KUBE_VERSION}/kubernetes-client-linux-${ARCH}.tar.gz | tar -xz -C . && cp ./kubernetes/client/bin/kubectl /usr/bin/kubectl \
  && chmod +x /usr/bin/kubectl && rm -rf ./kubernetes
 
-RUN --mount=type=bind,target=/packages,from=ovs-builder,source=/packages  \
-    dpkg -i /packages/openvswitch-*.deb /packages/python3-openvswitch*.deb && \
-    dpkg -i --ignore-depends=openvswitch-switch,openvswitch-common /packages/ovn-*.deb && \
-    if [ "${DEBUG}" = "true" ]; then dpkg -i --ignore-depends=openvswitch-switch,openvswitch-common /packages/*.ddeb; fi && \
-    rm -rf /var/lib/openvswitch/pki/
-
 ENV BFDD_VERSION="v0.5.4"
 RUN curl -sSf -L --retry 3 -o /usr/local/bin/bfdd-control https://github.com/bobz965/bfd-binary-for-kube-ovn-cni/releases/download/${BFDD_VERSION}/bfdd-control && \
     curl -sSf -L --retry 3 -o /usr/local/bin/bfdd-beacon https://github.com/bobz965/bfd-binary-for-kube-ovn-cni/releases/download/${BFDD_VERSION}/bfdd-beacon && \
@@ -111,4 +105,17 @@ RUN dumb_init_arch="x86_64"; \
     curl -sSf -L --retry 5 -o /usr/bin/dumb-init https://github.com/Yelp/dumb-init/releases/download/v${DUMB_INIT_VERSION}/dumb-init_${DUMB_INIT_VERSION}_${dumb_init_arch} && \
     chmod +x /usr/bin/dumb-init
 
+RUN --mount=type=bind,target=/packages,from=ovs-builder,source=/packages  \
+    dpkg -i /packages/openvswitch-*.deb /packages/python3-openvswitch*.deb && \
+    dpkg -i --ignore-depends=openvswitch-switch,openvswitch-common /packages/ovn-*.deb && \
+    rm -rf /var/lib/openvswitch/pki/
+
+ARG DEBUG=false
+RUN --mount=type=bind,target=/packages,from=ovs-builder,source=/packages \
+    if [ "${DEBUG}" = "true" ]; then \
+        apt update && apt install -y --no-install-recommends valgrind && \
+        rm -rf /var/lib/apt/lists/* && \
+        dpkg -i --ignore-depends=openvswitch-switch,openvswitch-common /packages/*.ddeb; \
+    fi
+
 ENTRYPOINT ["/usr/bin/dumb-init", "--"]

test/e2e/kube-ovn/qos/qos.go

@@ -75,27 +75,23 @@ func waitOvsQosForPod(f *framework.Framework, table string, pod *corev1.Pod, exp
 var _ = framework.Describe("[group:qos]", func() {
 	f := framework.NewDefaultFramework("qos")
 
-	var subnetName, namespaceName string
+	var podName, namespaceName string
 	var podClient *framework.PodClient
-	var subnetClient *framework.SubnetClient
 
 	ginkgo.BeforeEach(func() {
 		podClient = f.PodClient()
-		subnetClient = f.SubnetClient()
 		namespaceName = f.Namespace.Name
+		podName = "pod-" + framework.RandomSuffix()
 	})
 	ginkgo.AfterEach(func() {
-		if subnetName != "" {
-			ginkgo.By("Deleting subnet " + subnetName)
-			subnetClient.DeleteSync(subnetName)
-		}
+		ginkgo.By("Deleting pod " + podName)
+		podClient.DeleteSync(podName)
 	})
 
 	framework.ConformanceIt("should support netem QoS", func() {
 		f.SkipVersionPriorTo(1, 9, "Support for netem QoS was introduced in v1.9")
 
-		name := "pod-" + framework.RandomSuffix()
-		ginkgo.By("Creating pod " + name)
+		ginkgo.By("Creating pod " + podName)
 		latency, jitter, limit, loss := 600, 400, 2000, 10
 		annotations := map[string]string{
 			util.NetemQosLatencyAnnotation: strconv.Itoa(latency),
@@ -105,7 +101,7 @@ var _ = framework.Describe("[group:qos]", func() {
 		if !f.VersionPriorTo(1, 12) {
 			annotations[util.NetemQosJitterAnnotation] = strconv.Itoa(jitter)
 		}
-		pod := framework.MakePod(namespaceName, name, nil, annotations, "", nil, nil)
+		pod := framework.MakePod(namespaceName, podName, nil, annotations, "", nil, nil)
 		pod = podClient.CreateSync(pod)
 
 		ginkgo.By("Validating pod annotations")
@@ -126,17 +122,13 @@ var _ = framework.Describe("[group:qos]", func() {
 		}
 		framework.ExpectHaveKeyWithValue(qos, "limit", strconv.Itoa(limit))
 		framework.ExpectHaveKeyWithValue(qos, "loss", strconv.Itoa(loss))
-
-		ginkgo.By("Deleting pod " + name)
-		podClient.DeleteSync(pod.Name)
 	})
 
 	framework.ConformanceIt("should be able to update netem QoS", func() {
 		f.SkipVersionPriorTo(1, 9, "Support for netem QoS was introduced in v1.9")
 
-		name := "pod-" + framework.RandomSuffix()
-		ginkgo.By("Creating pod " + name + " without QoS")
-		pod := framework.MakePod(namespaceName, name, nil, nil, "", nil, nil)
+		ginkgo.By("Creating pod " + podName + " without QoS")
+		pod := framework.MakePod(namespaceName, podName, nil, nil, "", nil, nil)
 		pod = podClient.CreateSync(pod)
 
 		ginkgo.By("Validating pod annotations")
@@ -176,21 +168,17 @@ var _ = framework.Describe("[group:qos]", func() {
 		}
 		framework.ExpectHaveKeyWithValue(qos, "limit", strconv.Itoa(limit))
 		framework.ExpectHaveKeyWithValue(qos, "loss", strconv.Itoa(loss))
-
-		ginkgo.By("Deleting pod " + name)
-		podClient.DeleteSync(pod.Name)
 	})
 
 	framework.ConformanceIt("should support htb QoS", func() {
 		f.SkipVersionPriorTo(1, 9, "Support for htb QoS with priority was introduced in v1.9")
 
-		name := "pod-" + framework.RandomSuffix()
-		ginkgo.By("Creating pod " + name)
+		ginkgo.By("Creating pod " + podName)
 		ingressRate := 300
 		annotations := map[string]string{
 			util.IngressRateAnnotation: strconv.Itoa(ingressRate),
 		}
-		pod := framework.MakePod(namespaceName, name, nil, annotations, "", nil, nil)
+		pod := framework.MakePod(namespaceName, podName, nil, annotations, "", nil, nil)
 		pod = podClient.CreateSync(pod)
 
 		ginkgo.By("Validating pod annotations")
@@ -201,8 +189,5 @@ var _ = framework.Describe("[group:qos]", func() {
 		ginkgo.By("Validating OVS Queue")
 		queue := getOvsQosForPod(f, "queue", pod)
 		framework.ExpectHaveKeyWithValue(queue, "max-rate", strconv.Itoa(ingressRate*1000*1000))
-
-		ginkgo.By("Deleting pod " + name)
-		podClient.DeleteSync(pod.Name)
 	})
 })