OCPBUGS-78694: [release-4.20]: openstack: Disable LoadBalancer in cloud config when Octavia is unavailable

pkg/asset/manifests/openstack/cloudproviderconfig.go

@@ -2,6 +2,7 @@ package openstack
 
 import (
 	"context"
+	"errors"
 	"os"
 	"strconv"
 	"strings"
@@ -25,6 +26,27 @@ type Error struct {
 func (e Error) Error() string { return e.msg + ": " + e.err.Error() }
 func (e Error) Unwrap() error { return e.err }
 
+// isOctaviaAvailable checks if the Octavia (load-balancer) endpoint exists
+// in the OpenStack service catalog.
+func isOctaviaAvailable(ctx context.Context, clientOpts *clientconfig.ClientOpts) bool {
+	if clientOpts == nil {
+		// If no client options provided, assume Octavia is available
+		// to maintain backward compatibility
+		return true
+	}
+	_, err := openstackdefaults.NewServiceClient(ctx, "load-balancer", clientOpts)
+	if err != nil {
+		var gerr *gophercloud.ErrEndpointNotFound
+		if errors.As(err, &gerr) {
+			return false
+		}
+		// For other errors, assume Octavia might be available
+		// to avoid incorrectly disabling it
+		return true
+	}
+	return true
+}
+
 // CloudProviderConfigSecret generates the cloud provider config for the OpenStack
 // platform, that will be stored in the system secret.
 // TODO: I think this is crud for the legacy cloud-provider and is no longer needed. Burn it with fire?
@@ -78,7 +100,7 @@ func CloudProviderConfigSecret(cloud *clientconfig.Cloud) ([]byte, error) {
 	return []byte(res.String()), nil
 }
 
-func generateCloudProviderConfig(ctx context.Context, networkClient *gophercloud.ServiceClient, cloudConfig *clientconfig.Cloud, installConfig types.InstallConfig) (cloudProviderConfigData, cloudProviderConfigCABundleData string, err error) {
+func generateCloudProviderConfig(ctx context.Context, networkClient *gophercloud.ServiceClient, cloudConfig *clientconfig.Cloud, clientOpts *clientconfig.ClientOpts, installConfig types.InstallConfig) (cloudProviderConfigData, cloudProviderConfigCABundleData string, err error) {
 	cloudProviderConfigData = `[Global]
 secret-name = openstack-credentials
 secret-namespace = kube-system
@@ -96,7 +118,13 @@ secret-namespace = kube-system
 		cloudProviderConfigCABundleData = string(caFile)
 	}
 
-	if installConfig.OpenStack.ExternalNetwork != "" {
+	switch {
+	case !isOctaviaAvailable(ctx, clientOpts):
+		// Explicitly disable LoadBalancer when Octavia is not available
+		// to prevent CCM from crashing on startup.
+		// See: https://issues.redhat.com/browse/OCPBUGS-64842
+		cloudProviderConfigData += "\n[LoadBalancer]\nenabled = false\n"
+	case installConfig.OpenStack.ExternalNetwork != "":
 		networkName := installConfig.OpenStack.ExternalNetwork // Yes, we use a name in install-config.yaml :/
 		networkID, err := networkutils.IDFromName(ctx, networkClient, networkName)
 		if err != nil {
@@ -123,5 +151,5 @@ func GenerateCloudProviderConfig(ctx context.Context, installConfig types.Instal
 		return "", "", Error{err, "failed to create a network client"}
 	}
 
-	return generateCloudProviderConfig(ctx, networkClient, session.CloudConfig, installConfig)
+	return generateCloudProviderConfig(ctx, networkClient, session.CloudConfig, session.ClientOpts, installConfig)
 }

pkg/asset/manifests/openstack/cloudproviderconfig_test.go

@@ -130,7 +130,7 @@ region = my_region
 
 	for _, tc := range cases {
 		t.Run(tc.name, func(t *testing.T) {
-			actualConfig, _, err := generateCloudProviderConfig(context.Background(), nil, &cloud, *tc.installConfig)
+			actualConfig, _, err := generateCloudProviderConfig(context.Background(), nil, &cloud, nil, *tc.installConfig)
 			assert.NoError(t, err, "unexpected error when generating cloud provider config")
 			assert.Equal(t, tc.expectedConfig, actualConfig, "unexpected cloud provider config")
 		})