NVIDIA-596: pass DPU lease config via env vars on dpu-host/dpu DaemonSets

jkary · tsorya · commit 62c31b1dd672 · 2026-03-18T23:57:57.000-04:00
Add configurable DPU node lease renew interval and duration as env vars on ovnkube-controller, gated to dpu-host/dpu modes. Script-lib builds CLI flags from env vars. Values read from hardware-offload-config ConfigMap with defaults 10s/40s. Setting either to 0 disables the health check. Lease namespace derived via fieldRef. Jira: https://issues.redhat.com/browse/NVIDIA-596
diff --git a/bindata/network/ovn-kubernetes/common/008-script-lib.yaml b/bindata/network/ovn-kubernetes/common/008-script-lib.yaml
@@ -508,6 +508,9 @@ data:
       # Ensure openflow_probe_flag is always defined
       openflow_probe_flag=
 
+      # Ensure dpu_lease_flags is always defined
+      dpu_lease_flags=
+
       if [[ $# -ne 3 ]]; then
         echo "Expected three arguments but got $#"
         exit 1
@@ -550,7 +553,15 @@ data:
 
         # disable init-ovnkube-controller for dpu-host mode as it is not supported
         init_ovnkube_controller=""
+      fi
 
+      if [ "${OVN_NODE_MODE}" == "dpu-host" ] || [ "${OVN_NODE_MODE}" == "dpu" ]; then
+        if [[ -n "${OVNKUBE_NODE_LEASE_RENEW_INTERVAL}" ]]; then
+          dpu_lease_flags="--dpu-node-lease-renew-interval ${OVNKUBE_NODE_LEASE_RENEW_INTERVAL}"
+        fi
+        if [[ -n "${OVNKUBE_NODE_LEASE_DURATION}" ]]; then
+          dpu_lease_flags="$dpu_lease_flags --dpu-node-lease-duration ${OVNKUBE_NODE_LEASE_DURATION}"
+        fi
       fi
 
       if [ "{{.OVN_GATEWAY_MODE}}" == "shared" ]; then
@@ -702,5 +713,6 @@ data:
         ${ovn_v4_masquerade_subnet_opt} \
         ${ovn_v6_masquerade_subnet_opt} \
         ${ovn_v4_transit_switch_subnet_opt} \
-        ${ovn_v6_transit_switch_subnet_opt}
+        ${ovn_v6_transit_switch_subnet_opt} \
+        ${dpu_lease_flags}
     }
diff --git a/bindata/network/ovn-kubernetes/managed/ovnkube-node.yaml b/bindata/network/ovn-kubernetes/managed/ovnkube-node.yaml
@@ -437,6 +437,12 @@ spec:
         - name: OVNKUBE_NODE_MGMT_PORT_DP_RESOURCE_NAME
           value: {{ .MgmtPortResourceName }}
         {{ end }}
+        {{ if and (or (eq .OVN_NODE_MODE "dpu-host") (eq .OVN_NODE_MODE "dpu")) (ne .DpuNodeLeaseRenewInterval "0") }}
+        - name: OVNKUBE_NODE_LEASE_RENEW_INTERVAL
+          value: "{{.DpuNodeLeaseRenewInterval}}"
+        - name: OVNKUBE_NODE_LEASE_DURATION
+          value: "{{.DpuNodeLeaseDuration}}"
+        {{ end }}
 {{ if .HTTP_PROXY }}
         - name: "HTTP_PROXY"
           value: "{{ .HTTP_PROXY}}"
diff --git a/bindata/network/ovn-kubernetes/self-hosted/ovnkube-node.yaml b/bindata/network/ovn-kubernetes/self-hosted/ovnkube-node.yaml
@@ -472,6 +472,12 @@ spec:
         - name: OVNKUBE_NODE_MGMT_PORT_DP_RESOURCE_NAME
           value: {{ .MgmtPortResourceName }}
         {{ end }}
+        {{ if and (or (eq .OVN_NODE_MODE "dpu-host") (eq .OVN_NODE_MODE "dpu")) (ne .DpuNodeLeaseRenewInterval "0") }}
+        - name: OVNKUBE_NODE_LEASE_RENEW_INTERVAL
+          value: "{{.DpuNodeLeaseRenewInterval}}"
+        - name: OVNKUBE_NODE_LEASE_DURATION
+          value: "{{.DpuNodeLeaseDuration}}"
+        {{ end }}
         - name: K8S_NODE
           valueFrom:
             fieldRef:
diff --git a/hack/hardware-offload-config.yaml b/hack/hardware-offload-config.yaml
@@ -12,3 +12,5 @@ data:
     dpu-mode-label: "network.operator.openshift.io/dpu="
     smart-nic-mode-label: "network.operator.openshift.io/smart-nic="
     mgmt-port-resource-name: "openshift.io/mgmtvf"
+    dpu-node-lease-renew-interval: "10"
+    dpu-node-lease-duration: "40"
diff --git a/pkg/bootstrap/types.go b/pkg/bootstrap/types.go
@@ -30,18 +30,20 @@ type OVNHyperShiftBootstrapResult struct {
 }
 
 type OVNConfigBoostrapResult struct {
-	GatewayMode           string
-	HyperShiftConfig      *OVNHyperShiftBootstrapResult
-	DisableUDPAggregation bool
-	DpuHostModeLabel      string
-	DpuHostModeNodes      []string
-	DpuHostModeValue      string
-	DpuModeLabel          string
-	DpuModeNodes          []string
-	SmartNicModeLabel     string
-	SmartNicModeNodes     []string
-	SmartNicModeValue     string
-	MgmtPortResourceName  string
+	GatewayMode               string
+	HyperShiftConfig          *OVNHyperShiftBootstrapResult
+	DisableUDPAggregation     bool
+	DpuHostModeLabel          string
+	DpuHostModeNodes          []string
+	DpuHostModeValue          string
+	DpuModeLabel              string
+	DpuModeNodes              []string
+	SmartNicModeLabel         string
+	SmartNicModeNodes         []string
+	SmartNicModeValue         string
+	MgmtPortResourceName      string
+	DpuNodeLeaseRenewInterval int
+	DpuNodeLeaseDuration      int
 	// ConfigOverrides contains the overrides for the OVN Kubernetes configuration
 	// This is used to set the hidden OVN Kubernetes configuration in the cluster
 	// It is a map of key-value pairs where the key is the configuration option and the
diff --git a/pkg/network/kube_proxy_test.go b/pkg/network/kube_proxy_test.go
@@ -374,10 +374,12 @@ func TestFillKubeProxyDefaults(t *testing.T) {
 var FakeKubeProxyBootstrapResult = bootstrap.BootstrapResult{
 	OVN: bootstrap.OVNBootstrapResult{
 		OVNKubernetesConfig: &bootstrap.OVNConfigBoostrapResult{
-			DpuHostModeLabel:     OVN_NODE_SELECTOR_DEFAULT_DPU_HOST,
-			DpuModeLabel:         OVN_NODE_SELECTOR_DEFAULT_DPU,
-			SmartNicModeLabel:    OVN_NODE_SELECTOR_DEFAULT_SMART_NIC,
-			MgmtPortResourceName: "",
+			DpuHostModeLabel:          OVN_NODE_SELECTOR_DEFAULT_DPU_HOST,
+			DpuModeLabel:              OVN_NODE_SELECTOR_DEFAULT_DPU,
+			SmartNicModeLabel:         OVN_NODE_SELECTOR_DEFAULT_SMART_NIC,
+			MgmtPortResourceName:      "",
+			DpuNodeLeaseRenewInterval: DPU_NODE_LEASE_RENEW_INTERVAL_DEFAULT,
+			DpuNodeLeaseDuration:      DPU_NODE_LEASE_DURATION_DEFAULT,
 		},
 	},
 }
diff --git a/pkg/network/ovn_kubernetes.go b/pkg/network/ovn_kubernetes.go
@@ -64,6 +64,11 @@ const OVN_NODE_SELECTOR_DEFAULT_DPU = "network.operator.openshift.io/dpu="
 const OVN_NODE_SELECTOR_DEFAULT_SMART_NIC = "network.operator.openshift.io/smart-nic="
 const OVN_NODE_IDENTITY_CERT_DURATION = "24h"
 
+// Default DPU health check lease configuration.
+// Setting renew-interval to 0 disables the health check.
+const DPU_NODE_LEASE_RENEW_INTERVAL_DEFAULT = 10
+const DPU_NODE_LEASE_DURATION_DEFAULT = 40
+
 // gRPC healthcheck port. See: https://github.com/openshift/enhancements/pull/1209
 const OVN_EGRESSIP_HEALTHCHECK_PORT = "9107"
 
@@ -218,6 +223,8 @@ func renderOVNKubernetes(conf *operv1.NetworkSpec, bootstrapResult *bootstrap.Bo
 	data.Data["SmartNicModeLabel"] = bootstrapResult.OVN.OVNKubernetesConfig.SmartNicModeLabel
 	data.Data["SmartNicModeValue"] = bootstrapResult.OVN.OVNKubernetesConfig.SmartNicModeValue
 	data.Data["MgmtPortResourceName"] = bootstrapResult.OVN.OVNKubernetesConfig.MgmtPortResourceName
+	data.Data["DpuNodeLeaseRenewInterval"] = strconv.Itoa(bootstrapResult.OVN.OVNKubernetesConfig.DpuNodeLeaseRenewInterval)
+	data.Data["DpuNodeLeaseDuration"] = strconv.Itoa(bootstrapResult.OVN.OVNKubernetesConfig.DpuNodeLeaseDuration)
 	data.Data["OVN_CONTROLLER_INACTIVITY_PROBE"] = os.Getenv("OVN_CONTROLLER_INACTIVITY_PROBE")
 	controller_inactivity_probe := os.Getenv("OVN_CONTROLLER_INACTIVITY_PROBE")
 	if len(controller_inactivity_probe) == 0 {
@@ -927,10 +934,12 @@ func findCommonNode(nodeLists ...[]string) (bool, string) {
 // if it exists, otherwise returns default configuration for OCP clusters using OVN-Kubernetes
 func bootstrapOVNConfig(conf *operv1.Network, kubeClient cnoclient.Client, hc *hypershift.HyperShiftConfig, infraStatus *bootstrap.InfraStatus) (*bootstrap.OVNConfigBoostrapResult, error) {
 	ovnConfigResult := &bootstrap.OVNConfigBoostrapResult{
-		DpuHostModeLabel:     OVN_NODE_SELECTOR_DEFAULT_DPU_HOST,
-		DpuModeLabel:         OVN_NODE_SELECTOR_DEFAULT_DPU,
-		SmartNicModeLabel:    OVN_NODE_SELECTOR_DEFAULT_SMART_NIC,
-		MgmtPortResourceName: "",
+		DpuHostModeLabel:          OVN_NODE_SELECTOR_DEFAULT_DPU_HOST,
+		DpuModeLabel:              OVN_NODE_SELECTOR_DEFAULT_DPU,
+		SmartNicModeLabel:         OVN_NODE_SELECTOR_DEFAULT_SMART_NIC,
+		MgmtPortResourceName:      "",
+		DpuNodeLeaseRenewInterval: DPU_NODE_LEASE_RENEW_INTERVAL_DEFAULT,
+		DpuNodeLeaseDuration:      DPU_NODE_LEASE_DURATION_DEFAULT,
 	}
 	if conf.Spec.DefaultNetwork.OVNKubernetesConfig.GatewayConfig == nil {
 		bootstrapOVNGatewayConfig(conf, kubeClient.ClientFor("").CRClient())
@@ -976,6 +985,33 @@ func bootstrapOVNConfig(conf *operv1.Network, kubeClient cnoclient.Client, hc *h
 		if exists {
 			ovnConfigResult.MgmtPortResourceName = mgmtPortresourceName
 		}
+
+		if val, exists := cm.Data["dpu-node-lease-renew-interval"]; exists {
+			parsed, err := strconv.Atoi(val)
+			if err == nil && parsed >= 0 {
+				ovnConfigResult.DpuNodeLeaseRenewInterval = parsed
+			} else {
+				klog.Warningf("Invalid dpu-node-lease-renew-interval %q, using default %d", val, DPU_NODE_LEASE_RENEW_INTERVAL_DEFAULT)
+			}
+		}
+		if val, exists := cm.Data["dpu-node-lease-duration"]; exists {
+			parsed, err := strconv.Atoi(val)
+			if err == nil && parsed >= 0 {
+				ovnConfigResult.DpuNodeLeaseDuration = parsed
+			} else {
+				klog.Warningf("Invalid dpu-node-lease-duration %q, using default %d", val, DPU_NODE_LEASE_DURATION_DEFAULT)
+			}
+		}
+
+		// Setting either value to 0 disables the DPU health check.
+		// When both are non-zero, duration must be greater than interval.
+		if ovnConfigResult.DpuNodeLeaseRenewInterval != 0 && ovnConfigResult.DpuNodeLeaseDuration != 0 &&
+			ovnConfigResult.DpuNodeLeaseDuration <= ovnConfigResult.DpuNodeLeaseRenewInterval {
+			klog.Warningf("dpu-node-lease-duration (%d) must be greater than dpu-node-lease-renew-interval (%d), using defaults",
+				ovnConfigResult.DpuNodeLeaseDuration, ovnConfigResult.DpuNodeLeaseRenewInterval)
+			ovnConfigResult.DpuNodeLeaseRenewInterval = DPU_NODE_LEASE_RENEW_INTERVAL_DEFAULT
+			ovnConfigResult.DpuNodeLeaseDuration = DPU_NODE_LEASE_DURATION_DEFAULT
+		}
 	}
 
 	// We want to see if there are any nodes that are labeled for specific modes such as Full/SmartNIC/DPU Host/DPU
diff --git a/pkg/network/ovn_kubernetes_dpu_host_test.go b/pkg/network/ovn_kubernetes_dpu_host_test.go
@@ -1,6 +1,7 @@
 package network
 
 import (
+	"strconv"
 	"testing"
 
 	"github.com/ghodss/yaml"
@@ -160,6 +161,8 @@ func createTestRenderData(ovnNodeMode string) render.RenderData {
 	data.Data["SmartNicModeValue"] = ""
 	data.Data["DpuModeLabel"] = ""
 	data.Data["MgmtPortResourceName"] = ""
+	data.Data["DpuNodeLeaseRenewInterval"] = strconv.Itoa(DPU_NODE_LEASE_RENEW_INTERVAL_DEFAULT)
+	data.Data["DpuNodeLeaseDuration"] = strconv.Itoa(DPU_NODE_LEASE_DURATION_DEFAULT)
 	data.Data["HTTP_PROXY"] = ""
 	data.Data["HTTPS_PROXY"] = ""
 	data.Data["NO_PROXY"] = ""
@@ -210,6 +213,94 @@ func getMatchExpression(g *WithT, ds *appsv1.DaemonSet, label string) (corev1.No
 	return corev1.NodeSelectorOpDoesNotExist, ""
 }
 
+// TestOVNKubernetesLeaseEnvVars tests that DPU lease env vars are set
+// for DPU and DPU-host modes but not for full mode
+func TestOVNKubernetesLeaseEnvVars(t *testing.T) {
+	templates := []struct {
+		name         string
+		templatePath string
+	}{
+		{
+			name:         "managed",
+			templatePath: "../../bindata/network/ovn-kubernetes/managed/ovnkube-node.yaml",
+		},
+		{
+			name:         "self-hosted",
+			templatePath: "../../bindata/network/ovn-kubernetes/self-hosted/ovnkube-node.yaml",
+		},
+	}
+
+	testCases := []struct {
+		name        string
+		ovnNodeMode string
+		expectSet   bool
+	}{
+		{
+			name:        "full mode should not have lease env vars",
+			ovnNodeMode: "full",
+			expectSet:   false,
+		},
+		{
+			name:        "dpu-host mode should have lease env vars",
+			ovnNodeMode: "dpu-host",
+			expectSet:   true,
+		},
+		{
+			name:        "dpu mode should have lease env vars",
+			ovnNodeMode: "dpu",
+			expectSet:   true,
+		},
+	}
+
+	// Env vars with literal values
+	leaseEnvVars := map[string]string{
+		"OVNKUBE_NODE_LEASE_RENEW_INTERVAL": strconv.Itoa(DPU_NODE_LEASE_RENEW_INTERVAL_DEFAULT),
+		"OVNKUBE_NODE_LEASE_DURATION":       strconv.Itoa(DPU_NODE_LEASE_DURATION_DEFAULT),
+	}
+	for _, template := range templates {
+		for _, tc := range testCases {
+			testName := template.name + "_" + tc.name
+			t.Run(testName, func(t *testing.T) {
+				g := NewGomegaWithT(t)
+
+				data := createTestRenderData(tc.ovnNodeMode)
+
+				objs, err := render.RenderTemplate(template.templatePath, &data)
+				g.Expect(err).NotTo(HaveOccurred())
+				g.Expect(objs).To(HaveLen(1))
+
+				yamlBytes, err := yaml.Marshal(objs[0])
+				g.Expect(err).NotTo(HaveOccurred())
+
+				ds := &appsv1.DaemonSet{}
+				err = yaml.Unmarshal(yamlBytes, ds)
+				g.Expect(err).NotTo(HaveOccurred())
+
+				for envName, expectedValue := range leaseEnvVars {
+					found := false
+					for _, container := range ds.Spec.Template.Spec.Containers {
+						for _, env := range container.Env {
+							if env.Name == envName {
+								found = true
+								g.Expect(env.Value).To(Equal(expectedValue),
+									"%s should be set to %s", envName, expectedValue)
+							}
+						}
+					}
+
+					if tc.expectSet {
+						g.Expect(found).To(BeTrue(),
+							"%s should be set for %s mode", envName, tc.ovnNodeMode)
+					} else {
+						g.Expect(found).To(BeFalse(),
+							"%s should not be set for %s mode", envName, tc.ovnNodeMode)
+					}
+				}
+			})
+		}
+	}
+}
+
 // TestOVNKubernetesNodeSelectorOperator tests that the node selector operator works correctly with label values of different Full/SmartNIC/DPU modes
 func TestOVNKubernetesNodeSelectorOperator(t *testing.T) {
 	templates := []struct {
diff --git a/pkg/network/ovn_kubernetes_test.go b/pkg/network/ovn_kubernetes_test.go
