feat(fqdn): support using short name access service

core/manager/bootstrap.go

@@ -19,6 +19,7 @@ package manager
 import (
 	"fmt"
 	"os"
+	"strings"
 	"time"
 
 	"github.com/cloudwego/kitex/pkg/klog"
@@ -30,6 +31,7 @@ import (
 const (
 	PodNamespace   = "POD_NAMESPACE"
 	PodName        = "POD_NAME"
+	MetaNamespace  = "NAMESPACE"
 	InstanceIP     = "INSTANCE_IP"
 	IstiodAddr     = "istiod.istio-system.svc:15010"
 	KitexXdsDomain = "KITEX_XDS_DOMAIN"
@@ -40,8 +42,12 @@ const (
 )
 
 type BootstrapConfig struct {
-	node      *v3core.Node
-	xdsSvrCfg *XDSServerConfig
+	// The namespace to make up fqdn.
+	// Use POD_NAMESPACE default, the meta namespace will override that if set.
+	configNamespace string
+	nodeDomain      string
+	node            *v3core.Node
+	xdsSvrCfg       *XDSServerConfig
 }
 
 type XDSServerConfig struct {
@@ -88,6 +94,23 @@ func nodeId(podIP, podName, namespace, nodeDomain string) string {
 	return fmt.Sprintf("sidecar~%s~%s.%s~%s.svc.%s", podIP, podName, namespace, namespace, nodeDomain)
 }
 
+// tryExpandFQDN try expand fully qualified domain name.
+func (bc *BootstrapConfig) tryExpandFQDN(host string) string {
+	// The kubernetes services following the <serviceName>.<ns>.svc.<suffix> naming convention
+	// and that share a suffix with the domain. If it already been expanded, ignore it.
+	if strings.Contains(host, ".svc.") {
+		return host
+	}
+	var b strings.Builder
+	b.Grow(len(host) + len(bc.configNamespace) + len(bc.nodeDomain) + 10)
+	b.WriteString(host)
+	b.WriteString(".")
+	b.WriteString(bc.configNamespace)
+	b.WriteString(".svc.")
+	b.WriteString(bc.nodeDomain)
+	return b.String()
+}
+
 // newBootstrapConfig constructs the bootstrapConfig
 func newBootstrapConfig(config *XDSServerConfig) (*BootstrapConfig, error) {
 	// Get info from env
@@ -114,13 +137,22 @@ func newBootstrapConfig(config *XDSServerConfig) (*BootstrapConfig, error) {
 		nodeDomain = "cluster.local"
 	}
 
-	metaEnvs := os.Getenv(KitexXdsMetas)
-
-	return &BootstrapConfig{
+	bsConfig := &BootstrapConfig{
+		nodeDomain:      nodeDomain,
+		configNamespace: namespace,
 		node: &v3core.Node{
 			Id:       nodeId(podIP, podName, namespace, nodeDomain),
-			Metadata: parseMetaEnvs(metaEnvs, istioVersion),
+			Metadata: parseMetaEnvs(os.Getenv(KitexXdsMetas), istioVersion),
 		},
 		xdsSvrCfg: config,
-	}, nil
+	}
+
+	// the priority of NAMESPACE in metadata is higher than POD_NAMESPACE.
+	// ref: https://github.com/istio/istio/blob/30446a7b88aba4a0fcd5f71bae8d397a874e846f/pilot/pkg/model/context.go#L1024
+	if field, ok := bsConfig.node.Metadata.Fields[MetaNamespace]; ok {
+		if val := field.GetStringValue(); val != "" {
+			bsConfig.configNamespace = val
+		}
+	}
+	return bsConfig, nil
 }

core/manager/client.go

@@ -385,6 +385,27 @@ func (c *xdsClient) sendRequest(req *discoveryv3.DiscoveryRequest) {
 	c.reqCh <- req
 }
 
+func (c *xdsClient) resolveAddr(host string) string {
+	// In the worst case, lookupHost is called twice, try to reduce it.
+	// May exists three kind host:
+	// 1. fqdn host in Kubernetes, such as example.default.svc.cluster.local, invoke once always.
+	// 2. short name in Kubernetes, such as example, invoke once when the host exists in cipResolver, and twice when the host does not exist in cipResolver.
+	// 3. service outside Kubernetes, such as www.example.com, invoke twice always.
+	// FIXME: format as <serviceName>.<namespace> is not supported.
+	fqdn := c.config.tryExpandFQDN(host)
+	cip, ok := c.cipResolver.lookupHost(fqdn)
+	if ok && len(cip) > 0 {
+		return cip[0]
+	}
+	if fqdn != host {
+		cip, ok := c.cipResolver.lookupHost(host)
+		if ok && len(cip) > 0 {
+			return cip[0]
+		}
+	}
+	return ""
+}
+
 // getListenerName returns the listener name in this format: ${clusterIP}_${port}
 // lookup the clusterIP using the cipResolver and return the listenerName
 func (c *xdsClient) getListenerName(rName string) (string, error) {
@@ -393,10 +414,9 @@ func (c *xdsClient) getListenerName(rName string) (string, error) {
 		return "", fmt.Errorf("invalid listener name: %s", rName)
 	}
 	addr, port := tmp[0], tmp[1]
-	cip, ok := c.cipResolver.lookupHost(addr)
-	if ok && len(cip) > 0 {
-		clusterIPPort := cip[0] + "_" + port
-		return clusterIPPort, nil
+	cip := c.resolveAddr(addr)
+	if len(cip) > 0 {
+		return cip + "_" + port, nil
 	}
 	return "", fmt.Errorf("failed to convert listener name for %s", rName)
 }

core/manager/client_test.go

@@ -218,3 +218,45 @@ func TestClearCh(t *testing.T) {
 	clearRequestCh(ch, 10)
 	assert.Equal(t, 0, len(ch))
 }
+
+func TestResolveAddr(t *testing.T) {
+	resolver := newNdsResolver()
+	resolver.updateLookupTable(map[string][]string{
+		"echoa.default.svc.cluster.local": {"1.1.1.1"},
+		"echob.default.svc.cluster.local": {"1.1.1.1"},
+	})
+	cli := &xdsClient{
+		cipResolver: resolver,
+		config: &BootstrapConfig{
+			configNamespace: "default",
+			nodeDomain:      "cluster.local",
+		},
+	}
+	testCases := []struct {
+		desc string
+		host string
+		want string
+	}{
+		{
+			desc: "fqdn",
+			host: "echoa.default.svc.cluster.local",
+			want: "1.1.1.1",
+		},
+		{
+			desc: "short name",
+			host: "echoa",
+			want: "1.1.1.1",
+		},
+		{
+			desc: "not found",
+			host: "echoa.default",
+			want: "",
+		},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.desc, func(t *testing.T) {
+			got := cli.resolveAddr(tc.host)
+			assert.Equal(t, tc.want, got)
+		})
+	}
+}