New zone database

nameserver/addrs.go

@@ -25,7 +25,7 @@ type ZoneLookup interface {
 	LookupInaddr(inaddr string) ([]ZoneRecord, error)
 }
 
-type ZoneObserver interface {
+type ZoneObservable interface {
 	// Observe anything that affects a particular name in the zone
 	ObserveName(name string, observer ZoneRecordObserver) error
 	// Observe anything that affects a particular IP in the zone

nameserver/http_test.go

@@ -27,7 +27,11 @@ func TestHttp(t *testing.T) {
 		dockerIP        = "9.8.7.6"
 	)
 
-	zone := NewZoneDb(DefaultLocalDomain)
+	zone, err := NewZoneDb(ZoneConfig{})
+	wt.AssertNoErr(t, err)
+	err = zone.Start()
+	wt.AssertNoErr(t, err)
+	defer zone.Stop()
 
 	port := rand.Intn(10000) + 32768
 	fmt.Println("Http test on port", port)

nameserver/server.go

@@ -10,7 +10,6 @@ import (
 )
 
 const (
-	DefaultLocalDomain   = "weave.local."     // The default name used for the local domain
 	DefaultServerPort    = 53                 // The default server port
 	DefaultCLICfgFile    = "/etc/resolv.conf" // default "resolv.conf" file to try to load
 	DefaultUDPBuflen     = 4096               // bigger than the default 512

nameserver/server_test.go

@@ -44,8 +44,15 @@ func TestUDPDNSServer(t *testing.T) {
 	)
 	testCIDR1 := testAddr1 + "/24"
 
-	InitDefaultLogging(true)
-	var zone = NewZoneDb(DefaultLocalDomain)
+	InitDefaultLogging(testing.Verbose())
+	Info.Println("TestUDPDNSServer starting")
+
+	zone, err := NewZoneDb(ZoneConfig{})
+	wt.AssertNoErr(t, err)
+	err = zone.Start()
+	wt.AssertNoErr(t, err)
+	defer zone.Stop()
+
 	ip, _, _ := net.ParseCIDR(testCIDR1)
 	zone.AddRecord(containerID, successTestName, ip)
 
@@ -125,9 +132,15 @@ func TestTCPDNSServer(t *testing.T) {
 	)
 	dnsAddr := fmt.Sprintf("localhost:%d", testPort)
 
-	InitDefaultLogging(true)
-	var zone = NewZoneDb(DefaultLocalDomain)
+	InitDefaultLogging(testing.Verbose())
+	Info.Println("TestTCPDNSServer starting")
 
+	zone, err := NewZoneDb(ZoneConfig{})
+	wt.AssertNoErr(t, err)
+	err = zone.Start()
+	wt.AssertNoErr(t, err)
+	defer zone.Stop()
+	
 	// generate a list of `numAnswers` IP addresses
 	var addrs []ZoneRecord
 	bs := make([]byte, 4)

nameserver/zone.go

@@ -3,41 +3,27 @@ package nameserver
 import (
 	"bytes"
 	"fmt"
+	"github.com/benbjohnson/clock"
 	"github.com/miekg/dns"
 	. "github.com/weaveworks/weave/common"
 	"net"
 	"sync"
+	"time"
 )
 
 const (
 	RDNSDomain = "in-addr.arpa."
 )
 
+const (
+	DefaultLocalDomain     = "weave.local." // The default name used for the local domain
+
+	defaultRemoteIdent    = "weave:remote" // Ident used for info obtained from mDNS
+)
+
 // +1 to also exclude a dot
 var rdnsDomainLen = len(RDNSDomain) + 1
 
-type Zone interface {
-	AddRecord(ident string, name string, ip net.IP) error
-	DeleteRecord(ident string, ip net.IP) error
-	DeleteRecordsFor(ident string) error
-	Domain() string
-	ZoneLookup
-}
-
-type dbRecord struct {
-	Ident string
-	Name  string
-	IP    net.IP
-}
-
-// Very simple data structure for now, with linear searching.
-// TODO: make more sophisticated to improve performance.
-type ZoneDb struct {
-	mx     sync.RWMutex
-	recs   []dbRecord
-	domain string
-}
-
 type LookupError string
 
 func (ops LookupError) Error() string {
@@ -51,103 +37,609 @@ func (dup DuplicateError) Error() string {
 	return "Tried to add a duplicate entry"
 }
 
-func NewZoneDb(domain string) *ZoneDb {
-	return &ZoneDb{
-		domain: domain,
+type Zone interface {
+	ZoneObservable
+	ZoneLookup
+	// The domain where we operate (eg, "weave.local.")
+	Domain() string
+	// Add a record in the local database
+	AddRecord(ident string, name string, ip net.IP) error
+	// Delete a record in the local database
+	DeleteRecord(ident string, ip net.IP) error
+	// Delete all records for an ident in the local database
+	DeleteRecordsFor(ident string) error
+	// Return a status string
+	Status() string
+}
+
+///////////////////////////////////////////////////////////////////
+//
+// Zone database:
+//
+//           idents            names             entry
+//         *-------*          *------*         *------*
+// zone -> | ident |  --*-->  | name |  --*->  | IPv4 |
+//         *-------*    |     *------*    |    *------*
+//         | ident |    *-->  | name |    *->  | IPv4 |
+//         *-------*          *------*         *------*
+//            ...               ...
+//
+// The zone database keeps an ident per container, but also a special
+// ident for names/IPs obtained from remote peers.
+//
+// Each name can have multiple IPs (so far, we only consider IPv4s), and
+// each IP is stored in a `recordEntry` in the database. Entries store
+// some additional information like priority and weight of the IP (for
+// example, for future use with SRV records)
+//
+
+// An entry in the zone database
+type recordEntry struct {
+	Record
+
+	observers []ZoneRecordObserver // the observers for this record
+	insTime   time.Time            // time when this record was inserted
+}
+
+func newRecordEntry(zr ZoneRecord, now time.Time) *recordEntry {
+	return &recordEntry{
+		Record: Record{
+			name:     zr.Name(),
+			ip:       zr.IP(),
+			priority: zr.Priority(),
+			weight:   zr.Weight(),
+			ttl:      zr.TTL(),
+		},
+		insTime:   now,
+		observers: make([]ZoneRecordObserver, 0),
 	}
 }
 
-func (zone *ZoneDb) Domain() string {
-	return zone.domain
+func (i *recordEntry) addIPObserver(zro ZoneRecordObserver) {
+	i.observers = append(i.observers, zro)
 }
 
-func (zone *ZoneDb) indexOf(match func(dbRecord) bool) int {
-	for i, r := range zone.recs {
-		if match(r) {
-			return i
+// Notify all the IP observers and then remove them all
+// IP address observers are notified when
+// - the IP is removed
+// - the name is removed
+func (re *recordEntry) notifyIPObservers() {
+	numObservers := len(re.observers)
+	if numObservers > 0 {
+		Debug.Printf("[zonedb] Notifying %d observers of '%s'", numObservers, re.ip)
+		for _, observer := range re.observers {
+			observer()
 		}
+		re.observers = make([]ZoneRecordObserver, 0)
 	}
-	return -1
 }
 
-func (zone *ZoneDb) String() string {
-	zone.mx.RLock()
-	defer zone.mx.RUnlock()
+func (re recordEntry) TTL() int {
+	// if we never set the TTL (eg, when using AddRecord()), return the standard value...
+	if re.ttl == 0 {
+		return int(localTTL)
+	}
+	return re.ttl
+}
+
+// Check if the info in this record is still valid according to the TTL
+func (re recordEntry) hasExpired(now time.Time) bool {
+	return re.ttl > 0 && now.Sub(re.insTime).Seconds() > float64(re.ttl)
+}
+
+func (re recordEntry) String() string {
 	var buf bytes.Buffer
-	for _, r := range zone.recs {
-		fmt.Fprintf(&buf, "%.12s %s %v\n", r.Ident, r.IP, r.Name)
+	fmt.Fprintf(&buf, "%s", re.Record)
+	lobs := len(re.observers)
+	if lobs > 0 {
+		fmt.Fprintf(&buf, "/OBS:%d", lobs)
 	}
 	return buf.String()
 }
 
-func (zone *ZoneDb) LookupName(name string) ([]ZoneRecord, error) {
+///////////////////////////////////////////////////////////////////
+
+// a name, with group of records by IPv4
+type name struct {
+	name            string
+	ipv4            map[IPv4]*recordEntry // all the IPv4 records for this name
+	observers       []ZoneRecordObserver  // the observers for this name
+	lastAccessTime  time.Time             // last time a Lookup* used info from this name
+}
+
+func newName(n string) *name {
+	return &name{
+		name:      n,
+		ipv4:      make(map[IPv4]*recordEntry),
+		observers: make([]ZoneRecordObserver, 0),
+	}
+}
+func (n name) len() int             { return len(n.ipv4) }
+func (n name) empty() bool          { return n.len() == 0 }
+func (n name) hasIP(ip net.IP) bool { return n.hasIPv4(ipToIPv4(ip)) }
+func (n name) hasIPv4(ip IPv4) bool { _, found := n.ipv4[ip]; return found }
+
+// Get all the entries for this name
+func (n *name) getAllEntries() (res []*recordEntry) {
+	for _, entry := range n.ipv4 {
+		res = append(res, entry)
+	}
+	return
+}
+
+// Get the entry for an IP in this name
+func (n *name) getEntryForIP(ip IPv4) (res *recordEntry) {
+	if entry, found := n.ipv4[ip]; found {
+		res = entry
+	}
+	return
+}
+
+// Add a new IPv4 to this name
+func (n *name) addIP(zr ZoneRecord, now time.Time) (*recordEntry, error) {
+	ne := newRecordEntry(zr, now)
+	ipv4 := ipToIPv4(zr.IP())
+	n.ipv4[ipv4] = ne
+	n.notifyNameObservers()
+	return ne, nil
+}
+
+// Delete an IP for this name
+func (n *name) deleteIP(ip IPv4) bool {
+	if ipRecord, found := n.ipv4[ip]; found {
+		ipRecord.notifyIPObservers()
+		n.notifyNameObservers()
+		delete(n.ipv4, ip)
+		return true
+	}
+	return false
+}
+
+// Update the list of IPs for this name, adding new ones and removing old IPs...
+func (n *name) updateIPs(records []ZoneRecord, now time.Time) (added int, removed int) {
+	ipsMap := make(map[IPv4]bool)
+
+	// add the new IPs
+	for _, record := range records {
+		ipv4 := ipToIPv4(record.IP())
+		ipsMap[ipv4] = true
+		if !n.hasIPv4(ipv4) {
+			n.addIP(record, now)
+			added++
+		}
+	}
+
+	// remove the old IPs
+	for ipv4, ipRecord := range n.ipv4 {
+		if _, isNew := ipsMap[ipv4]; !isNew {
+			ipRecord.notifyIPObservers() // we must notify the observers before removing the IP record...
+			delete(n.ipv4, ipv4)
+			removed++
+		}
+	}
+
+	// notify observers and update times
+	if added > 0 || removed > 0 {
+		n.notifyNameObservers()
+	}
+	return
+}
+
+// Add a observer for this name
+func (n *name) addNameObserver(observer ZoneRecordObserver) {
+	n.observers = append(n.observers, observer)
+}
+
+// Notify and flush all the observers
+// Name observers are notified when
+// - an IP is added
+// - an IP is removed
+// - the name is removed
+func (n *name) notifyNameObservers() {
+	numObservers := len(n.observers)
+	if numObservers > 0 {
+		Debug.Printf("[zonedb] Notifying %d observers of '%s'", numObservers, n.name)
+		for _, observer := range n.observers {
+			observer()
+		}
+		n.observers = make([]ZoneRecordObserver, 0)
+	}
+}
+
+// Convert a set of IP records to a comma-separated string
+func (n name) String() string {
+	var buf bytes.Buffer
+	for _, ip := range n.ipv4 {
+		fmt.Fprintf(&buf, "%s, ", ip)
+	}
+	l := buf.Len()
+	if l > 2 {
+		buf.Truncate(l - 2)
+	}
+	return buf.String()
+}
+
+///////////////////////////////////////////////////////////////////
+
+// all the names in a ident
+type namesSet struct {
+	names map[string]*name
+}
+
+func newNamesSet() *namesSet    { return &namesSet{names: make(map[string]*name)} }
+func (ns namesSet) len() int    { return len(ns.names) }
+func (ns namesSet) empty() bool { return ns.len() == 0 }
+func (ns namesSet) hasName(name string) bool {
+	if n, found := ns.names[name]; !found {
+		return false
+	} else {
+		// the name must have some valid IPs in order to be considered as "existent"
+		return len(n.ipv4) > 0
+	}
+}
+
+// Get the zone entries for a name
+// If the name does not exist and `create` is true, a new name will be created
+func (ns *namesSet) getName(name string, create bool) (n *name) {
+	if n, found := ns.names[name]; !found && create {
+		newName := newName(name)
+		ns.names[name] = newName
+		return newName
+	} else {
+		return n
+	}
+}
+
+// Get all the entries for a name
+func (ns *namesSet) getEntriesForName(name string) (res []*recordEntry) {
+	n := ns.getName(name, false)
+	if n != nil {
+		res = n.getAllEntries()
+	}
+	return
+}
+
+// Get all the entries for an IP, for all names
+func (ns *namesSet) getEntriesForIP(ip IPv4) (res []*recordEntry) {
+	for _, name := range ns.names {
+		entry := name.getEntryForIP(ip)
+		if entry != nil {
+			res = append(res, entry)
+		}
+	}
+	return
+}
+
+// Add a new IPv4 to a name
+func (ns *namesSet) addIPToName(zr ZoneRecord, now time.Time) (*recordEntry, error) {
+	n := ns.getName(zr.Name(), true)
+	ipv4 := ipToIPv4(zr.IP())
+	if n.hasIPv4(ipv4) {
+		return n.getEntryForIP(ipv4), DuplicateError{}
+	}
+	ns.touchName(zr.Name(), now)
+	return n.addIP(zr, now)
+}
+
+// Delete a name in the names set
+func (ns *namesSet) deleteName(n string) error {
+	if name, found := ns.names[n]; found {
+		for _, ipRecord := range name.getAllEntries() {
+			ipRecord.notifyIPObservers()
+		}
+		name.notifyNameObservers()
+		delete(ns.names, n)
+		return nil
+	} else {
+		return LookupError(fmt.Sprintf("%+v", name))
+	}
+}
+
+// Delete an IPv4 from all names
+func (ns *namesSet) deleteIP(ip IPv4) error {
+	cnt := 0
+	for nkey, n := range ns.names {
+		if n.deleteIP(ip) {
+			cnt++
+		}
+		if n.empty() {
+			delete(ns.names, nkey)
+		}
+	}
+
+	// we must return an error if no records have been found and deleted
+	if cnt == 0 {
+		return LookupError(fmt.Sprintf("%+v", ip))
+	}
+	return nil
+}
+
+// Get the name query time
+func (ns namesSet) getNameLastAccess(n string) time.Time {
+	if name, found := ns.names[n]; found {
+		return name.lastAccessTime
+	}
+	return time.Time{}
+}
+
+// Touch the last access time for a name
+// The access time is saved only for locally-introduced records (otherwise it could
+// be lost when names are irrelevant...)
+func (ns *namesSet) touchName(name string, now time.Time) {
+	Debug.Printf("[zonedb] Touching name %s", name)
+	n := ns.getName(name, false)
+	if n != nil {
+		n.lastAccessTime = now
+	}
+}
+
+func (ns namesSet) String() string {
+	var buf bytes.Buffer
+	for _, name := range ns.names {
+		fmt.Fprintf(&buf, "%s, ", name)
+	}
+	l := buf.Len()
+	if l > 2 {
+		buf.Truncate(l - 2)
+	}
+	return buf.String()
+}
+
+///////////////////////////////////////////////////////////////////
+
+// Names sets, by ident name
+type identRecordsSet map[string]*namesSet
+
+func (irs identRecordsSet) String() string {
+	var buf bytes.Buffer
+	for ident, nameset := range irs {
+		fmt.Fprintf(&buf, "%.12s: %s\n", ident, nameset)
+	}
+	return buf.String()
+}
+
+///////////////////////////////////////////////////////////////////
+
+type ZoneConfig struct {
+	// The local domain
+	Domain string
+	// The interface where mDNS operates
+	Iface *net.Interface
+	// Force a specific mDNS client
+	MDNSClient ZoneMDNSClient
+	// Force a specific mDNS server
+	MDNSServer ZoneMDNSServer
+	// For a specific clock provider
+	Clock clock.Clock
+}
+
+// Zone database
+// The zone database contains the locally known information about the zone.
+// The zone database uses some other mechanisms (eg, mDNS) for finding unknown
+// information or for keeping current information up to date.
+type zoneDb struct {
+	mx     sync.RWMutex
+	idents identRecordsSet
+
+	mdnsCli       ZoneMDNSClient
+	mdnsSrv       ZoneMDNSServer
+	iface         *net.Interface
+	domain        string        // the local domain
+	clock         clock.Clock
+}
+
+// Create a new zone database
+func NewZoneDb(config ZoneConfig) (zone *zoneDb, err error) {
+	zone = &zoneDb{
+		domain:           config.Domain,
+		idents:           make(identRecordsSet),
+		mdnsCli:          config.MDNSClient,
+		mdnsSrv:          config.MDNSServer,
+		iface:            config.Iface,
+		clock:            config.Clock,
+	}
+
+	// fix the default configuration parameters
+	if zone.clock == nil {
+		zone.clock = clock.New()
+	}
+	if len(zone.domain) == 0 {
+		zone.domain = DefaultLocalDomain
+	}
+
+	// Create the mDNS client and server
+	if zone.mdnsCli == nil {
+		if zone.mdnsCli, err = NewMDNSClient(); err != nil {
+			return
+		}
+	}
+	if zone.mdnsSrv == nil {
+		if zone.mdnsSrv, err = NewMDNSServer(zone); err != nil {
+			return
+		}
+	}
+
+	return
+}
+
+// Start the zone database
+func (zone *zoneDb) Start() (err error) {
+	if zone.iface != nil {
+		Info.Printf("[zonedb] Using mDNS on %s", zone.iface)
+	} else {
+		Info.Printf("[zonedb] Using mDNS on all interfaces")
+	}
+
+	if err = zone.mdnsCli.Start(zone.iface); err != nil {
+		return
+	}
+	if err = zone.mdnsSrv.Start(zone.iface); err != nil {
+		return
+	}
+	return
+}
+
+// Perform a graceful shutdown of the zone database
+func (zone *zoneDb) Stop() error {
+	Debug.Printf("[zonedb] Exiting mDNS client and server...")
+	zone.mdnsCli.Stop()
+	zone.mdnsSrv.Stop()
+	return nil
+}
+
+// Obtain the domain where this database keeps information for
+func (zone *zoneDb) Domain() string {
+	return zone.domain
+}
+
+// Return the status string
+func (zone *zoneDb) Status() string {
+	var buf bytes.Buffer
+	fmt.Fprintln(&buf, "Local domain", zone.domain)
+	fmt.Fprintln(&buf, "Interface", zone.iface)
+	fmt.Fprintf(&buf, "Zone database:\n%s", zone)
+	return buf.String()
+}
+
+// Add a record with `name` pointing to `ip`
+func (zone *zoneDb) AddRecord(ident string, name string, ip net.IP) (err error) {
+	zone.mx.Lock()
+	defer zone.mx.Unlock()
+	Debug.Printf("[zonedb] Adding record: '%s'/'%s'[%s]", ident, name, ip)
+	record := Record{dns.Fqdn(name), ip, 0, 0, 0}
+	_, err = zone.getNamesSet(ident).addIPToName(record, zone.clock.Now())
+	return
+}
+
+// Delete all records for an ident
+func (zone *zoneDb) DeleteRecordsFor(ident string) (err error) {
+	zone.mx.Lock()
+	defer zone.mx.Unlock()
+	Debug.Printf("[zonedb] Deleting records for ident '%s'", ident)
+	zone.deleteIdent(ident)
+	return
+}
+
+// Delete an IP in a ident
+func (zone *zoneDb) DeleteRecord(ident string, ip net.IP) (err error) {
+	zone.mx.Lock()
+	defer zone.mx.Unlock()
+
+	Debug.Printf("[zonedb] Deleting record '%s'/[%s]", ident, ip)
+	if _, found := zone.idents[ident]; found {
+		err = zone.idents[ident].deleteIP(ipToIPv4(ip))
+		if zone.idents[ident].empty() {
+			zone.deleteIdent(ident)
+		}
+	} else {
+		err = LookupError(ident)
+	}
+	return
+}
+
+// Observe a name.
+// The name must have at least one valid IP address.
+// Name observers are notified when
+// - an IP is added
+// - an IP is removed
+// - the name is removed
+// The observer will be invoked on any change/removal that affects this name.
+// Each observer will be invoked at least once (but possibly more). After that, they will be removed.
+// The observer should not try to lock the ZoneDB (you will get a deadlock)
+func (zone *zoneDb) ObserveName(name string, observer ZoneRecordObserver) (err error) {
+	zone.mx.Lock()
+	defer zone.mx.Unlock()
+
+	cnt := 0
+	name = dns.Fqdn(name)
+	for _, nameset := range zone.idents {
+		if n := nameset.getName(name, false); n != nil {
+			n.addNameObserver(observer)
+			cnt += 1
+		}
+	}
+	if cnt == 0 {
+		err = LookupError(fmt.Sprintf("%s", name))
+	}
+	return
+}
+
+// Observe a IP address.
+// The IP address must be exit in the database.
+// IP address observers are notified when
+// - the IP is removed
+// - the name is removed
+// The observer will be invoked on any change/removal that affects this IP.
+// Each observer will be invoked at least once (but possibly more). After that, they will be removed.
+// The observer should not try to lock the ZoneDB (you will get a deadlock)
+func (zone *zoneDb) ObserveInaddr(inaddr string, observer ZoneRecordObserver) (err error) {
+	zone.mx.Lock()
+	defer zone.mx.Unlock()
+
+	cnt := 0
+	revIP, err := raddrToIPv4(inaddr)
+	if err != nil {
+		return newParseError("lookup address", inaddr)
+	}
+	for _, nameset := range zone.idents {
+		for _, entry := range nameset.getEntriesForIP(revIP) {
+			entry.addIPObserver(observer)
+			cnt += 1
+		}
+	}
+	if cnt == 0 {
+		err = LookupError(fmt.Sprintf("%+v/%+v", inaddr, revIP))
+	}
+	return
+}
+
+// Get the string representation of a zone
+func (zone zoneDb) String() string {
 	zone.mx.RLock()
 	defer zone.mx.RUnlock()
-	for _, r := range zone.recs {
-		if r.Name == name {
-			return []ZoneRecord{Record{r.Name, r.IP, 0, 0, 0}}, nil
-		}
-	}
-	return nil, LookupError(name)
+	return zone.idents.String()
 }
 
-func (zone *ZoneDb) LookupInaddr(inaddr string) ([]ZoneRecord, error) {
-	if revIP := net.ParseIP(inaddr[:len(inaddr)-rdnsDomainLen]); revIP != nil {
-		revIP4 := revIP.To4()
-		ip := []byte{revIP4[3], revIP4[2], revIP4[1], revIP4[0]}
-		Debug.Printf("[zonedb] Looking for address: %+v", ip)
-		zone.mx.RLock()
-		defer zone.mx.RUnlock()
-		for _, r := range zone.recs {
-			if r.IP.Equal(ip) {
-				return []ZoneRecord{Record{r.Name, r.IP, 0, 0, 0}}, nil
-			}
-		}
-		return nil, LookupError(inaddr)
-	}
-	Warning.Printf("[zonedb] Asked to reverse lookup %s", inaddr)
-	return nil, LookupError(inaddr)
-}
-
-func (zone *ZoneDb) AddRecord(ident string, name string, ip net.IP) error {
-	zone.mx.Lock()
-	defer zone.mx.Unlock()
-	fqdn := dns.Fqdn(name)
-	pred := func(r dbRecord) bool { return r.Name == fqdn && r.IP.Equal(ip) && r.Ident == ident }
-	if index := zone.indexOf(pred); index != -1 {
-		return DuplicateError{}
-	}
-	zone.recs = append(zone.recs, dbRecord{ident, fqdn, ip})
-	return nil
-}
-
-func (zone *ZoneDb) DeleteRecord(ident string, ip net.IP) error {
-	zone.mx.Lock()
-	defer zone.mx.Unlock()
-	pred := func(r dbRecord) bool { return r.Ident == ident && r.IP.Equal(ip) }
-	if index := zone.indexOf(pred); index != -1 {
-		zone.recs = append(zone.recs[:index], zone.recs[index+1:]...)
-		return nil
-	}
-	return LookupError(ident)
-}
-
-func (zone *ZoneDb) DeleteRecordsFor(ident string) error {
-	zone.mx.Lock()
-	defer zone.mx.Unlock()
-	w := 0 // write index
-
-	for _, r := range zone.recs {
-		if r.Ident != ident {
-			zone.recs[w] = r
-			w++
-		}
-	}
-	zone.recs = zone.recs[:w]
-	return nil
-}
-
-func (zone *ZoneDb) ContainerDied(ident string) error {
+// Notify that a container has died
+func (zone *zoneDb) ContainerDied(ident string) error {
 	Info.Printf("[zonedb] Container %s down. Removing records", ident)
 	return zone.DeleteRecordsFor(ident)
 }
+
+// Return true if the we have obtained some information for a name from remote peers
+// We can have both local (eg, introduced through the HTTP API) and remote (eg, mDNS)
+// information for a name. This method only checks if there is remote information...
+func (zone *zoneDb) HasNameRemoteInfo(n string) bool {
+	zone.mx.Lock()
+	defer zone.mx.Unlock()
+
+	remoteNamesSet := zone.getNamesSet(defaultRemoteIdent)
+	if remoteNamesSet != nil {
+		return remoteNamesSet.hasName(n)
+	}
+	return false
+}
+
+// Get or create a names set
+func (zone *zoneDb) getNamesSet(ident string) *namesSet {
+	ns, found := zone.idents[ident]
+	if !found {
+		ns = newNamesSet()
+		zone.idents[ident] = ns
+	}
+	return ns
+}
+
+// Delete a ident
+func (zone *zoneDb) deleteIdent(ident string) {
+	Info.Printf("[zonedb] Removing everything for ident '%s'", ident)
+	if ns, found := zone.idents[ident]; found {
+		for _, n := range ns.names {
+			for _, entry := range n.getAllEntries() {
+				entry.notifyIPObservers()
+			}
+			n.notifyNameObservers()
+		}
+		delete(zone.idents, ident)
+	}
+}

nameserver/zone_lookup.go

@@ -0,0 +1,55 @@
+package nameserver
+
+import (
+	"github.com/miekg/dns"
+	. "github.com/weaveworks/weave/common"
+)
+
+// Perform a lookup for a name in the zone
+// The name can be resolved locally with the local database
+func (zone *zoneDb) LookupName(name string) (res []ZoneRecord, err error) {
+	zone.mx.RLock()
+	defer zone.mx.RUnlock()
+
+	// note: LookupName() is usually called from the mDNS server, so we do not touch the name
+	name = dns.Fqdn(name)
+	Debug.Printf("[zonedb] Looking for name '%s' in local database", name)
+	for identName, nameset := range zone.idents {
+		if identName != defaultRemoteIdent {
+			for _, ze := range nameset.getEntriesForName(name) {
+				res = append(res, ze)
+			}
+		}
+	}
+
+	if len(res) == 0 {
+		err = LookupError(name)
+	}
+	return
+}
+
+// Perform a lookup for a IP address in the zone
+// The address can be resolved locally with the local database
+func (zone *zoneDb) LookupInaddr(inaddr string) (res []ZoneRecord, err error) {
+	zone.mx.RLock()
+	defer zone.mx.RUnlock()
+
+	// note: LookupInaddr() is usually called from the mDNS server, so we do not touch the name
+
+	revIPv4, err := raddrToIPv4(inaddr)
+	if err != nil {
+		return nil, newParseError("lookup address", inaddr)
+	}
+	Debug.Printf("[zonedb] Looking for address in local database: '%s' (%s)", revIPv4, inaddr)
+	for identName, nameset := range zone.idents {
+		if identName != defaultRemoteIdent {
+			for _, ze := range nameset.getEntriesForIP(revIPv4) {
+				res = append(res, ZoneRecord(ze))
+			}
+		}
+	}
+	if len(res) == 0 {
+		err = LookupError(inaddr)
+	}
+	return
+}

nameserver/zone_test.go

@@ -1,6 +1,8 @@
 package nameserver
 
 import (
+	"fmt"
+	. "github.com/weaveworks/weave/common"
 	wt "github.com/weaveworks/weave/testing"
 	"net"
 	"testing"
@@ -8,60 +10,118 @@ import (
 
 func TestZone(t *testing.T) {
 	var (
-		containerID      = "deadbeef"
-		otherContainerID = "cowjuice"
-		successTestName  = "test1.weave."
-		testAddr1        = "10.2.2.1/24"
+		container1    = "deadbeef"
+		container2    = "cowjuice"
+		name1         = "test1.weave."
+		name1Addr1    = "10.9.2.1/24"
+		name1Addr2    = "10.9.2.2/24"
+		revName1Addr1 = "1.2.9.10.in-addr.arpa."
 	)
 
-	var zone = NewZoneDb(DefaultLocalDomain)
+	InitDefaultLogging(testing.Verbose())
 
-	ip, _, _ := net.ParseCIDR(testAddr1)
-	err := zone.AddRecord(containerID, successTestName, ip)
+	zone, err := NewZoneDb(ZoneConfig{})
+	wt.AssertNoErr(t, err)
+	err = zone.Start()
+	wt.AssertNoErr(t, err)
+	defer zone.Stop()
+
+	ip1, _, _ := net.ParseCIDR(name1Addr1)
+	t.Logf("Adding '%s'/%s to '%s'", name1, ip1, container1)
+	err = zone.AddRecord(container1, name1, ip1)
 	wt.AssertNoErr(t, err)
 
 	// Add a few more records to make the job harder
-	err = zone.AddRecord("abcdef0123", "adummy.weave.", net.ParseIP("10.2.0.1"))
+	t.Logf("Adding 'adummy.weave.' to 'abcdef0123'")
+	err = zone.AddRecord("abcdef0123", "adummy.weave.", net.ParseIP("10.9.0.1"))
 	wt.AssertNoErr(t, err)
-	err = zone.AddRecord("0123abcdef", "zdummy.weave.", net.ParseIP("10.2.0.2"))
+	t.Logf("Adding 'zdummy.weave.' to '0123abcdef'")
+	err = zone.AddRecord("0123abcdef", "zdummy.weave.", net.ParseIP("10.9.0.2"))
+	wt.AssertNoErr(t, err)
+	t.Logf("Zone database:\n%s", zone)
+
+	t.Logf("Checking if we can find the name '%s'", name1)
+	foundIPs, err := zone.LookupName(name1)
 	wt.AssertNoErr(t, err)
 
-	// Check that the address is now there.
-	foundIP, err := zone.LookupName(successTestName)
-	wt.AssertNoErr(t, err)
-
-	if !foundIP[0].IP().Equal(ip) {
-		t.Fatal("Unexpected result for", successTestName, foundIP)
+	if !foundIPs[0].IP().Equal(ip1) {
+		t.Fatal("Unexpected result for", name1, foundIPs)
 	}
 
-	// See if we can find the address by IP.
-	foundName, err := zone.LookupInaddr("1.2.2.10.in-addr.arpa.")
+	t.Logf("Checking if we cannot find some silly name like 'something.wrong'")
+	foundIPs, err = zone.LookupName("something.wrong")
+	wt.AssertErrorType(t, err, (*LookupError)(nil), fmt.Sprintf("unknown name: %+v", foundIPs))
+
+	ip2, _, _ := net.ParseCIDR(name1Addr2)
+	t.Logf("Adding a second IP for '%s'/%s to '%s'", name1, ip2, container1)
+	err = zone.AddRecord(container1, name1, ip2)
 	wt.AssertNoErr(t, err)
 
-	if foundName[0].Name() != successTestName {
-		t.Fatal("Unexpected result for", ip, foundName)
+	t.Logf("Checking if we can find both the old IP and the new IP for '%s'", name1)
+	foundIPs, err = zone.LookupName(name1)
+	wt.AssertNoErr(t, err)
+	if !(foundIPs[0].IP().Equal(ip1) || foundIPs[1].IP().Equal(ip1)) {
+		t.Fatal("Unexpected result for", name1, foundIPs)
+	}
+	if !(foundIPs[0].IP().Equal(ip2) || foundIPs[1].IP().Equal(ip2)) {
+		t.Fatal("Unexpected result for", name1, foundIPs)
 	}
 
-	err = zone.AddRecord(containerID, successTestName, ip)
+	t.Logf("Checking if we can find the address by IP '1.2.9.10.in-addr.arpa.'")
+	foundNames, err := zone.LookupInaddr("1.2.9.10.in-addr.arpa.")
+	wt.AssertNoErr(t, err)
+
+	if foundNames[0].Name() != name1 {
+		t.Fatal("Unexpected result for", ip1, foundNames)
+	}
+
+	t.Logf("Checking we can not find an unknown address '30.20.10.1.in-addr.arpa.'")
+	foundNames, err = zone.LookupInaddr("30.20.10.1.in-addr.arpa.")
+	wt.AssertErrorType(t, err, (*LookupError)(nil), fmt.Sprintf("unknown IP: %+v", foundNames))
+
+	t.Logf("Checking if adding again '%s'/%s in %s results in an error", name1, ip1, container1)
+	err = zone.AddRecord(container1, name1, ip1)
 	wt.AssertErrorType(t, err, (*DuplicateError)(nil), "duplicate add")
 
-	err = zone.AddRecord(otherContainerID, successTestName, ip)
-	// Delete the record for the original container
-	err = zone.DeleteRecord(containerID, ip)
+	t.Logf("Adding '%s'/%s in %s too", name1, ip1, container2)
+	err = zone.AddRecord(container2, name1, ip1)
 	wt.AssertNoErr(t, err)
 
-	_, err = zone.LookupName(successTestName)
+	name1Removed := 0
+	err = zone.ObserveName(name1, func() { t.Logf("Observer #1 for '%s' notified.", name1); name1Removed++ })
+	wt.AssertNoErr(t, err)
+	err = zone.ObserveName(name1, func() { t.Logf("Observer #2 for '%s' notified.", name1); name1Removed++ })
+	wt.AssertNoErr(t, err)
+	err = zone.ObserveInaddr(revName1Addr1, func() { t.Logf("Observer #1 for '%s' notified.", revName1Addr1); name1Removed++ })
 	wt.AssertNoErr(t, err)
 
-	err = zone.DeleteRecord(otherContainerID, ip)
+	t.Logf("Zone database:\n%s", zone)
+	t.Logf("Deleting the %s in %s", ip1, container1)
+	err = zone.DeleteRecord(container1, ip1)
+	wt.AssertNoErr(t, err)
+	t.Logf("Zone database:\n%s", zone)
+
+	t.Logf("Checking %s's observers have been notified on removal", name1)
+	if name1Removed < 3 {
+		t.Logf("Zone database:\n%s", zone)
+		t.Fatalf("Unexpected number (%d) of calls to observers", name1Removed)
+	}
+
+	t.Logf("Checking %s can be found", name1)
+	_, err = zone.LookupName(name1)
 	wt.AssertNoErr(t, err)
 
-	// Check that the address is not there now.
-	_, err = zone.LookupName(successTestName)
+	t.Logf("Checking %s is not found after removing %s it in %s and %s in %s",
+		name1, ip1, container2, ip2, container1)
+	err = zone.DeleteRecord(container1, ip2)
+	err = zone.DeleteRecord(container2, ip1)
+	wt.AssertNoErr(t, err)
+	t.Logf("Zone database:\n%s", zone)
+	_, err = zone.LookupName(name1)
 	wt.AssertErrorType(t, err, (*LookupError)(nil), "after deleting record")
 
-	// Delete a record that isn't there
-	err = zone.DeleteRecord(containerID, net.ParseIP("0.0.0.0"))
+	t.Logf("Checking if removing an unknown record results in an error")
+	err = zone.DeleteRecord(container1, net.ParseIP("0.0.0.0"))
 	wt.AssertErrorType(t, err, (*LookupError)(nil), "when deleting record that doesn't exist")
 }
 
@@ -72,14 +132,22 @@ func TestDeleteFor(t *testing.T) {
 		addr1 = "10.2.2.3/24"
 		addr2 = "10.2.7.8/24"
 	)
-	zone := NewZoneDb(DefaultLocalDomain)
+
+	InitDefaultLogging(testing.Verbose())
+
+	zone, err := NewZoneDb(ZoneConfig{})
+	wt.AssertNoErr(t, err)
+	err = zone.Start()
+	wt.AssertNoErr(t, err)
+	defer zone.Stop()
+
 	for _, addr := range []string{addr1, addr2} {
 		ip, _, _ := net.ParseCIDR(addr)
 		err := zone.AddRecord(id, name, ip)
 		wt.AssertNoErr(t, err)
 	}
 
-	_, err := zone.LookupName(name)
+	_, err = zone.LookupName(name)
 	wt.AssertNoErr(t, err)
 
 	err = zone.DeleteRecordsFor(id)

weavedns/main.go

@@ -52,15 +52,6 @@ func main() {
 	InitDefaultLogging(debug)
 	Info.Printf("WeaveDNS version %s\n", version) // first thing in log: the version
 
-	var zone = weavedns.NewZoneDb(domain)
-
-	if watch {
-		err := updater.Start(apiPath, zone)
-		if err != nil {
-			Error.Fatal("Unable to start watcher", err)
-		}
-	}
-
 	var iface *net.Interface
 	if ifaceName != "" {
 		var err error
@@ -73,6 +64,27 @@ func main() {
 		}
 	}
 
+	zoneConfig := weavedns.ZoneConfig{
+		Domain:          domain,
+		Iface:           iface,
+	}
+	zone, err := weavedns.NewZoneDb(zoneConfig)
+	if err != nil {
+		Error.Fatal("[main] Unable to initialize the Zone database", err)
+	}
+	err = zone.Start()
+	if err != nil {
+		Error.Fatal("[main] Unable to start the Zone database", err)
+	}
+	defer zone.Stop()
+
+	if watch {
+		err := updater.Start(apiPath, zone)
+		if err != nil {
+			Error.Fatal("[main] Unable to start watcher", err)
+		}
+	}
+
 	srvConfig := weavedns.DNSServerConfig{
 		Port:        dnsPort,
 		CacheLen:    cacheLen,