fix: dial the ochre appliance customer ENI, not the management NIC (#851)

Both the endpoint ENI and the management NIC carry the rds-db-instance tag,
so a first-match pick non-deterministically dialed the unreachable management
NIC. Select the endpoint ENI by its RDS description, falling back to any usable
tagged NIC.

spinifex/handlers/ochrevector/hostport.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"log/slog"
 	"net/netip"
+	"strings"
 
 	"github.com/aws/aws-sdk-go/aws"
 	"github.com/aws/aws-sdk-go/service/ec2"
@@ -135,6 +136,11 @@ func ensureDaemonENI(ctx context.Context, vpcSvc vpcProvisioner, subnetID, nodeI
 // (tests/e2e/harness/rds.go) duplicates it the same way.
 const applianceInstanceTagKey = "spinifex:rds-db-instance"
 
+// endpointENIDescriptionPrefix mirrors the description handlers/rds gives the
+// customer (endpoint) ENI in resolveCustomerENI. The management NIC shares the
+// rds-db-instance tag, so description is what tells the reachable ENI apart.
+const endpointENIDescriptionPrefix = "RDS endpoint ENI for "
+
 // resolveApplianceTarget finds the appliance's own customer ENI by the tag
 // handlers/rds stamps on it at launch, returning the real private IP to dial
 // and the subnet it lives in -- never the vanity endpoint hostname, which is
@@ -152,13 +158,22 @@ func resolveApplianceTarget(ctx context.Context, deps HostPortDeps, identifier s
 	if err != nil {
 		return "", "", "", fmt.Errorf("ochrevector: describe appliance ENI for %s: %w", identifier, err)
 	}
+	// Both the endpoint ENI and the management NIC carry the rds-db-instance
+	// tag, but only the endpoint ENI (in the customer VPC) is reachable from the
+	// daemon. Prefer it by description; fall back to any usable tagged NIC.
 	var ni *ec2.NetworkInterface
 	if out != nil {
 		for _, cand := range out.NetworkInterfaces {
-			if aws.StringValue(cand.PrivateIpAddress) != "" && aws.StringValue(cand.SubnetId) != "" {
+			if aws.StringValue(cand.PrivateIpAddress) == "" || aws.StringValue(cand.SubnetId) == "" {
+				continue
+			}
+			if strings.HasPrefix(aws.StringValue(cand.Description), endpointENIDescriptionPrefix) {
 				ni = cand
 				break
 			}
+			if ni == nil {
+				ni = cand
+			}
 		}
 	}
 	if ni == nil {

spinifex/handlers/ochrevector/hostport_test.go

@@ -98,6 +98,26 @@ func (f *fakeVPC) putApplianceENI(eniID, identifier, subnetID, ip string) {
 	}
 }
 
+// putTaggedENIWithDescription seeds an appliance-tagged ENI that also carries
+// a description, so a test can present both the endpoint ENI and the
+// management NIC that share the rds-db-instance tag at launch.
+func (f *fakeVPC) putTaggedENIWithDescription(eniID, identifier, subnetID, ip, description string) {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	if f.records == nil {
+		f.records = map[string]*ec2.NetworkInterface{}
+	}
+	f.records[eniID] = &ec2.NetworkInterface{
+		NetworkInterfaceId: aws.String(eniID),
+		SubnetId:           aws.String(subnetID),
+		PrivateIpAddress:   aws.String(ip),
+		Description:        aws.String(description),
+		TagSet: []*ec2.Tag{
+			{Key: aws.String(applianceInstanceTagKey), Value: aws.String(identifier)},
+		},
+	}
+}
+
 // tagValue returns the value of key in tags, or "" if absent.
 func tagValue(tags []*ec2.Tag, key string) string {
 	for _, t := range tags {
@@ -404,6 +424,31 @@ func TestResolveApplianceTarget_FindsTheTaggedApplianceENI(t *testing.T) {
 	assert.Equal(t, "10.244.1.0/24", cidr)
 }
 
+// The endpoint ENI and the management NIC share the rds-db-instance tag, but
+// only the endpoint ENI (described "RDS endpoint ENI for ...") is reachable
+// from the daemon. The resolver must pick it by description regardless of the
+// order DescribeNetworkInterfaces returns them -- map iteration is unordered,
+// so a repeat loop exercises both orderings.
+func TestResolveApplianceTarget_PrefersEndpointENIOverManagementNIC(t *testing.T) {
+	const mgmtSubnet = "subnet-mgmt-0001"
+	const mgmtIP = "10.251.185.4"
+	for range 16 {
+		h := newHostPortHarness()
+		h.putSubnet(testApplianceSubnet, "10.244.1.0/24")
+		h.putSubnet(mgmtSubnet, "10.251.185.0/24")
+		h.vpc.putTaggedENIWithDescription("eni-mgmt", testApplianceIdentifier, mgmtSubnet, mgmtIP,
+			"RDS management NIC for "+testApplianceIdentifier)
+		h.vpc.putTaggedENIWithDescription("eni-endpoint", testApplianceIdentifier, testApplianceSubnet, testApplianceEndpoint,
+			endpointENIDescriptionPrefix+testApplianceIdentifier)
+
+		dialIP, subnetID, cidr, err := resolveApplianceTarget(t.Context(), h.deps(), testApplianceIdentifier)
+		require.NoError(t, err)
+		require.Equal(t, testApplianceEndpoint, dialIP, "must dial the endpoint ENI, never the management NIC")
+		require.Equal(t, testApplianceSubnet, subnetID)
+		require.Equal(t, "10.244.1.0/24", cidr)
+	}
+}
+
 // An ENI tagged for a different identifier must never match.
 func TestResolveApplianceTarget_IgnoresOtherIdentifiersTags(t *testing.T) {
 	h := newHostPortHarness()