feat(dns): implement DNS names for machines in m.internal (#359)

* feat(dns): implement extra dns names for machine in m.internal

This implements <machinename>.m.internal and <machineid>.m.internal as
names that can be queried in the cluster.

Listing all machines/services is not implemented, this may come later.

Fixes: #340

Signed-off-by: Miek Gieben <miek@miek.nl>

* Implement getting all machine address

A query for m.internal returns all ip address

There is no overflow check

Signed-off-by: Miek Gieben <miek@miek.nl>

---------

Signed-off-by: Miek Gieben <miek@miek.nl>

internal/machine/dns/resolver.go

@@ -10,6 +10,8 @@ import (
 	"sync"
 	"time"
 
+	"github.com/psviderski/uncloud/internal/machine/api/pb"
+	"github.com/psviderski/uncloud/internal/machine/network"
 	"github.com/psviderski/uncloud/internal/machine/store"
 )
 
@@ -19,7 +21,9 @@ type ClusterResolver struct {
 	store *store.Store
 	// serviceIPs maps service names to container IPs.
 	serviceIPs map[string][]netip.Addr
-	// mu protects the serviceIPs map.
+	// machineIPs maps machine IDs and names to their IP.
+	machineIPs map[string]netip.Addr
+	// mu protects the serviceIPs and machineIPs map.
 	mu sync.RWMutex
 	// lastUpdate tracks when records were last updated.
 	lastUpdate time.Time
@@ -46,6 +50,13 @@ func (r *ClusterResolver) Run(ctx context.Context) error {
 	// TODO: implement machine membership check using Corrossion Admin client to filter available containers.
 	r.updateServiceIPs(containers)
 
+	machines, mchanges, err := r.store.SubscribeMachines(ctx)
+	if err != nil {
+		return fmt.Errorf("subscribe to machine changes: %w", err)
+	}
+	r.log.Info("Subscribed to machine changes in the clsuter to keep machine DNS records updated.")
+	r.updateMachineIPs(machines)
+
 	for {
 		select {
 		case _, ok := <-changes:
@@ -59,9 +70,22 @@ func (r *ClusterResolver) Run(ctx context.Context) error {
 				r.log.Error("Failed to list containers.", "err", err)
 				continue
 			}
-
 			// TODO: implement machine membership check using Corrossion Admin client to filter available containers.
 			r.updateServiceIPs(containers)
+
+		case _, ok := <-mchanges:
+			if !ok {
+				return fmt.Errorf("machine subscription failed")
+			}
+			r.log.Debug("Cluster machines changed, updating DNS records.")
+
+			machines, err := r.store.ListMachines(ctx)
+			if err != nil {
+				r.log.Error("Failed to list machines.", "err", err)
+				continue
+			}
+			r.updateMachineIPs(machines)
+
 		case <-ctx.Done():
 			return nil
 		}
@@ -121,19 +145,49 @@ func (r *ClusterResolver) updateServiceIPs(containers []store.ContainerRecord) {
 	r.log.Info("DNS records updated.", "services", len(newServiceIPs)/3, "containers", containersCount)
 }
 
-// Resolve returns IP addresses of the service containers.
-func (r *ClusterResolver) Resolve(serviceName string) []netip.Addr {
+func (r *ClusterResolver) updateMachineIPs(machines []*pb.MachineInfo) {
+	newMachineIPs := make(map[string]netip.Addr, len(machines))
+
+	for _, machine := range machines {
+		subnet, err := machine.Network.Subnet.ToPrefix()
+		if err != nil {
+			continue
+		}
+		addr := network.MachineIP(subnet)
+		newMachineIPs[machine.Name+".m"] = addr
+		newMachineIPs[machine.Id+".m"] = addr
+	}
+	r.mu.Lock()
+	r.machineIPs = newMachineIPs
+	r.mu.Unlock()
+
+	r.log.Info("DNS records updated.", "machines", len(machines))
+}
+
+// Resolve returns IP addresses of the service containers or machines.
+func (r *ClusterResolver) Resolve(name string) []netip.Addr {
 	r.mu.RLock()
 	defer r.mu.RUnlock()
 
-	ips, ok := r.serviceIPs[serviceName]
-	if !ok || len(ips) == 0 {
-		return nil
+	// Return a copy of the IPs slice to prevent modification of the original.
+	ips, ok := r.serviceIPs[name]
+	if ok && len(ips) > 0 {
+		return slices.Clone(ips)
 	}
 
-	// Return a copy of the IPs slice to prevent modification of the original.
-	ipsCopy := make([]netip.Addr, len(ips))
-	copy(ipsCopy, ips)
+	ip, ok := r.machineIPs[name]
+	if ok {
+		return slices.Clone([]netip.Addr{ip})
+	}
 
-	return ipsCopy
+	if name == "m" { // all machines subdomain
+		// collect all unique IP address
+		uniq := map[string]netip.Addr{}
+		for _, addr := range r.machineIPs {
+			uniq[addr.String()] = addr
+		}
+		return slices.Clone(slices.Collect(maps.Values(uniq)))
+	}
+
+	return nil
 }

internal/machine/dns/resolver_test.go

@@ -1,11 +1,13 @@
 package dns
 
 import (
+	"net/netip"
 	"reflect"
 	"testing"
 
 	"github.com/docker/docker/api/types/container"
 	"github.com/docker/docker/api/types/network"
+	"github.com/psviderski/uncloud/internal/machine/api/pb"
 	"github.com/psviderski/uncloud/internal/machine/store"
 	"github.com/psviderski/uncloud/pkg/api"
 	"github.com/stretchr/testify/assert"
@@ -75,3 +77,32 @@ func newRecord(serviceID, serviceName, ip, machineID string) store.ContainerReco
 		MachineID: machineID,
 	}
 }
+
+func TestClusterResolver_UpdateMachineIPs(t *testing.T) {
+	t.Parallel()
+
+	machines := []*pb.MachineInfo{
+		newMachineRecord("x0y0z0", "mach-1", "10.210.0.0/24"),
+		newMachineRecord("x1y1z1", "mach-2", "10.210.1.0/24"),
+		newMachineRecord("x2y2z2", "mach-3", "10.210.2.0/24"),
+	}
+
+	r := NewClusterResolver(nil)
+	r.updateMachineIPs(machines)
+
+	assert.NotEmpty(t, r.Resolve("mach-1.m"))
+	assert.NotEmpty(t, r.Resolve("mach-3.m"))
+	assert.NotEmpty(t, r.Resolve("x0y0z0.m"))
+
+	assert.Equal(t, 3, len(r.Resolve("m")))
+}
+
+func newMachineRecord(machineID, machineName, prefix string) *pb.MachineInfo {
+	return &pb.MachineInfo{
+		Id:   machineID,
+		Name: machineName,
+		Network: &pb.NetworkConfig{
+			Subnet: pb.NewIPPrefix(netip.MustParsePrefix(prefix)),
+		},
+	}
+}

internal/machine/dns/server.go

@@ -293,13 +293,13 @@ func (s *Server) forwardRequest(req *dns.Msg, proto string) (*dns.Msg, error) {
 // handleAQuery processes an A query for the internal domain and returns A records for the requested name.
 // The internal domain suffix is already stripped from the name. An empty list is returned if no records are found.
 func (s *Server) handleAQuery(name string) []dns.RR {
-	serviceName, mode := extractModeFromDomain(trimInternalDomain(name))
-	ips := s.resolver.Resolve(serviceName)
+	unname, mode := extractModeFromDomain(trimInternalDomain(name))
+	ips := s.resolver.Resolve(unname)
 	if len(ips) == 0 {
-		s.log.Debug("Failed to resolve service name.", "service", serviceName)
+		s.log.Debug("Failed to resolve internal name.", "name", name)
 		return nil
 	}
-	s.log.Debug("Resolved service name.", "service", serviceName, "ips", ips)
+	s.log.Debug("Resolved service internal name.", "name", name, "ips", ips)
 
 	if len(ips) > 1 {
 		// Shuffle the IPs to approximate round-robin.

website/docs/3-concepts/6-services/1-internal-dns.md

@@ -3,6 +3,7 @@
 Services can be addressed on the internal WireGuard network by service name, service ID, or a machine-scoped service name:
 
 ## Service name
+
 ```
 $ nslookup nats.internal
 Server:         127.0.0.11
@@ -30,6 +31,7 @@ Address: 10.210.1.4
 ```
 
 ## Service ID
+
 ```
 $ nslookup 3ecb3a8bbec5fd3f46efb056a934714a.internal
 Server:         127.0.0.11
@@ -40,6 +42,7 @@ Address: 10.210.0.4
 ```
 
 ## Machine ID scoped service name
+
 ```
 $ nslookup 0903f0ee483aa97d559eeeaac5e22283.m.nats.internal
 Server:         127.0.0.11
@@ -64,7 +67,8 @@ Address: 10.210.1.4
 
 Additionally, the IP ordering preference can be specified with a `rr` (round-robin) or `nearest` subdomain prefix.
 
-### `rr` (round-robin) *current default*
+### `rr` (round-robin) _current default_
+
 Randomly shuffled order on each lookup.
 
 ```
@@ -96,9 +100,11 @@ Address: 10.210.0.3
 ```
 
 ## Nearest scope
+
 Returns machine-local instances first.
 
 `machine-a`:
+
 ```
 $ nslookup nearest.worker.internal
 Server:         127.0.0.11
@@ -115,6 +121,7 @@ Address: 10.210.1.4
 ```
 
 `machine-b`:
+
 ```
 $ nslookup nearest.worker.internal
 Server:         127.0.0.11
@@ -131,3 +138,28 @@ Address: 10.210.0.4
 ```
 
 The prefixes can be used with service ID and machine-scoped service names, as well (e.g. `nearest.3ecb3a8bbec5fd3f46efb056a934714a.internal` or `rr.0903f0ee483aa97d559eeeaac5e22283.m.worker.internal`).
+
+## Machine name
+
+Machines can be addressed too on the internal WireGuard network, either by name of by ID, these names are
+available from the `m.internal` zone.
+
+```
+$ nslookup machine-1.m.internal
+Server:         127.0.0.11
+Address:        127.0.0.11#53
+
+Name:   machine-1.m.internal
+Address: 10.210.0.1
+```
+
+## Machine ID
+
+```
+$ nslookup c337f00600de51ef4375c9a9a267dba5.m.internal
+Server:         127.0.0.11
+Address:        127.0.0.11#53
+
+Name:   c337f00600de51ef4375c9a9a267dba5.m.internal
+Address: 10.210.0.1
+```