cluster: rename pkg/zones to pkg/cluster

Signed-off-by: Alejandro Mery <amery@jpi.io>

pkg/cluster/ceph.go

@@ -0,0 +1,122 @@
+package cluster
+
+import (
+	"bytes"
+	"net/netip"
+	"sort"
+
+	"darvaza.org/core"
+	"github.com/gofrs/uuid/v5"
+
+	"git.jpi.io/amery/jpictl/pkg/ceph"
+)
+
+// GetCephFSID returns our Ceph's FSID
+func (m *Zones) GetCephFSID() (uuid.UUID, error) {
+	if core.IsZero(m.CephFSID) {
+		// generate one
+		v, err := uuid.NewV4()
+		if err != nil {
+			return uuid.Nil, err
+		}
+		m.CephFSID = v
+	}
+	return m.CephFSID, nil
+}
+
+// GetCephConfig reads the ceph.conf file
+func (m *Zones) GetCephConfig() (*ceph.Config, error) {
+	data, err := m.ReadFile("ceph.conf")
+	if err != nil {
+		return nil, err
+	}
+
+	r := bytes.NewReader(data)
+	return ceph.NewConfigFromReader(r)
+}
+
+// WriteCephConfig writes the ceph.conf file
+func (m *Zones) WriteCephConfig(cfg *ceph.Config) error {
+	f, err := m.CreateTruncFile("ceph.conf")
+	if err != nil {
+		return err
+	}
+	defer f.Close()
+
+	_, err = cfg.WriteTo(f)
+	return err
+}
+
+// GenCephConfig prepares a ceph.Config using the cluster information
+func (m *Zones) GenCephConfig() (*ceph.Config, error) {
+	fsid, err := m.GetCephFSID()
+	if err != nil {
+		return nil, err
+	}
+
+	cfg := &ceph.Config{
+		Global: ceph.GlobalConfig{
+			FSID: fsid,
+			ClusterNetwork: netip.PrefixFrom(
+				netip.AddrFrom4([4]byte{10, 0, 0, 0}),
+				8,
+			),
+		},
+	}
+
+	m.ForEachZone(func(z *Zone) bool {
+		for _, p := range z.GetCephMonitors() {
+			addr, _ := RingOneAddress(z.ID, p.ID)
+
+			cfg.Global.Monitors = append(cfg.Global.Monitors, p.Name)
+			cfg.Global.MonitorsAddr = append(cfg.Global.MonitorsAddr, addr)
+		}
+		return false
+	})
+
+	return cfg, nil
+}
+
+// GetCephMonitors returns the set of Ceph monitors on
+// the zone
+func (z *Zone) GetCephMonitors() Machines {
+	var mons Machines
+	var first, second *Machine
+
+	z.ForEachMachine(func(p *Machine) bool {
+		switch {
+		case p.CephMonitor:
+			// it is a monitor
+			mons = append(mons, p)
+		case len(mons) > 0:
+			// zone has a monitor
+		case first == nil && !p.IsGateway():
+			// first option for monitor
+			first = p
+		case second == nil:
+			// second option for monitor
+			second = p
+		}
+
+		return false
+	})
+
+	switch {
+	case len(mons) > 0:
+		// ready
+	case first != nil:
+		// make first option our monitor
+		first.CephMonitor = true
+		mons = append(mons, first)
+	case second != nil:
+		// make second option our monitor
+		second.CephMonitor = true
+		mons = append(mons, second)
+	default:
+		// zone without machines??
+		panic("unreachable")
+	}
+
+	sort.Sort(mons)
+	return mons
+}

pkg/cluster/ceph_scan.go

@@ -0,0 +1,103 @@
+package cluster
+
+import (
+	"os"
+
+	"darvaza.org/slog"
+	"git.jpi.io/amery/jpictl/pkg/ceph"
+)
+
+type cephScanTODO struct {
+	names map[string]bool
+	addrs map[string]bool
+}
+
+func (todo *cephScanTODO) checkMachine(p *Machine) bool {
+	// on ceph all addresses are ring1
+	ring1, _ := RingOneAddress(p.Zone(), p.ID)
+	addr := ring1.String()
+
+	if _, found := todo.names[p.Name]; found {
+		// found on the TODO by name
+		todo.names[p.Name] = true
+		todo.addrs[addr] = true
+		return true
+	}
+
+	if _, found := todo.addrs[addr]; found {
+		// found on the TODO by address
+		todo.names[p.Name] = true
+		todo.addrs[addr] = true
+		return true
+	}
+
+	return false
+}
+
+func (todo *cephScanTODO) LogMissing(log slog.Logger) {
+	for name, found := range todo.names {
+		if !found {
+			log.Warn().
+				WithField("subsystem", "ceph").
+				WithField("monitor", name).
+				Print("unknown monitor")
+		}
+	}
+
+	for addr, found := range todo.addrs {
+		if !found {
+			log.Warn().
+				WithField("subsystem", "ceph").
+				WithField("monitor", addr).
+				Print("unknown monitor")
+		}
+	}
+}
+
+func newCephScanTODO(cfg *ceph.Config) *cephScanTODO {
+	todo := &cephScanTODO{
+		names: make(map[string]bool),
+		addrs: make(map[string]bool),
+	}
+
+	for _, name := range cfg.Global.Monitors {
+		todo.names[name] = false
+	}
+
+	for _, addr := range cfg.Global.MonitorsAddr {
+		todo.addrs[addr.String()] = false
+	}
+
+	return todo
+}
+
+func (m *Zones) scanCephMonitors(_ *ScanOptions) error {
+	cfg, err := m.GetCephConfig()
+	switch {
+	case os.IsNotExist(err):
+		err = nil
+	case err != nil:
+		return err
+	}
+
+	if cfg != nil {
+		// store FSID
+		m.CephFSID = cfg.Global.FSID
+
+		// flag monitors based on config
+		todo := newCephScanTODO(cfg)
+		m.ForEachMachine(func(p *Machine) bool {
+			p.CephMonitor = todo.checkMachine(p)
+			return false
+		})
+
+		todo.LogMissing(m.log)
+	}
+
+	// make sure every zone has one
+	m.ForEachZone(func(z *Zone) bool {
+		_ = z.GetCephMonitors()
+		return false
+	})
+	return nil
+}

pkg/cluster/cluster.go

@@ -0,0 +1,2 @@
+// Package cluster contains information about the cluster
+package cluster

pkg/cluster/env.go

@@ -0,0 +1,177 @@
+package cluster
+
+import (
+	"bytes"
+	"fmt"
+	"io"
+	"strings"
+)
+
+// Env is a shell environment factory for this cluster
+type Env struct {
+	ZoneIterator
+
+	cephFSID string
+	export   bool
+}
+
+// Env returns a shell environment factory
+func (m *Zones) Env(export bool) (*Env, error) {
+	fsid, err := m.GetCephFSID()
+	if err != nil {
+		return nil, err
+	}
+
+	env := &Env{
+		ZoneIterator: m,
+		cephFSID:     fsid.String(),
+		export:       export,
+	}
+
+	return env, nil
+}
+
+// Zones returns the list of Zone IDs
+func (m *Env) Zones() []int {
+	var zones []int
+
+	m.ForEachZone(func(z *Zone) bool {
+		zones = append(zones, z.ID)
+		return false
+	})
+
+	return zones
+}
+
+// WriteTo generates environment variables for shell scripts
+func (m *Env) WriteTo(w io.Writer) (int64, error) {
+	var buf bytes.Buffer
+
+	if m.cephFSID != "" {
+		m.writeEnvVar(&buf, m.cephFSID, "FSID")
+	}
+
+	m.writeEnvVarInts(&buf, m.Zones(), "ZONES")
+
+	m.ForEachZone(func(z *Zone) bool {
+		m.writeEnvZone(&buf, z)
+		return false
+	})
+
+	return buf.WriteTo(w)
+}
+
+func (m *Env) writeEnvZone(w io.Writer, z *Zone) {
+	zoneID := z.ID
+
+	// ZONE{zoneID}
+	m.writeEnvVar(w, genEnvZoneNodes(z), "ZONE%v", zoneID)
+
+	// ZONE{zoneID}_NAME
+	m.writeEnvVar(w, z.Name, "ZONE%v_%s", zoneID, "NAME")
+
+	// ZONE{zoneID}_GW
+	gateways, _ := z.GatewayIDs()
+	m.writeEnvVarInts(w, gateways, "ZONE%v_%s", zoneID, "GW")
+
+	// Ceph
+	monitors := z.GetCephMonitors()
+	// MON{zoneID}_NAME
+	m.writeEnvVar(w, genEnvZoneCephMonNames(monitors), "MON%v_%s", zoneID, "NAME")
+	// MON{zoneID}_IP
+	m.writeEnvVar(w, genEnvZoneCephMonIPs(monitors), "MON%v_%s", zoneID, "IP")
+	// MON{zoneID}_ID
+	m.writeEnvVar(w, genEnvZoneCephMonIDs(monitors), "MON%v_%s", zoneID, "ID")
+}
+
+func (m *Env) writeEnvVarInts(w io.Writer, value []int, name string, args ...any) {
+	var s string
+
+	if n := len(value); n > 0 {
+		var buf bytes.Buffer
+
+		for i, v := range value {
+			if i != 0 {
+				_, _ = fmt.Fprint(&buf, " ")
+			}
+			_, _ = fmt.Fprintf(&buf, "%v", v)
+		}
+
+		s = buf.String()
+	}
+
+	m.writeEnvVar(w, s, name, args...)
+}
+
+func (m *Env) writeEnvVar(w io.Writer, value string, name string, args ...any) {
+	var prefix string
+
+	if m.export {
+		prefix = "export "
+	}
+
+	if len(args) > 0 {
+		name = fmt.Sprintf(name, args...)
+	}
+
+	if name != "" {
+		value = strings.TrimSpace(value)
+
+		_, _ = fmt.Fprintf(w, "%s%s=%q\n", prefix, name, value)
+	}
+}
+
+func genEnvZoneNodes(z *Zone) string {
+	if n := z.Len(); n > 0 {
+		s := make([]string, 0, n)
+
+		z.ForEachMachine(func(p *Machine) bool {
+			s = append(s, p.Name)
+			return false
+		})
+
+		return strings.Join(s, " ")
+	}
+	return ""
+}
+
+func genEnvZoneCephMonNames(m Machines) string {
+	var buf strings.Builder
+	m.ForEachMachine(func(p *Machine) bool {
+		if buf.Len() > 0 {
+			_, _ = buf.WriteRune(' ')
+		}
+		_, _ = buf.WriteString(p.Name)
+
+		return false
+	})
+	return buf.String()
+}
+
+func genEnvZoneCephMonIPs(m Machines) string {
+	var buf strings.Builder
+	m.ForEachMachine(func(p *Machine) bool {
+		addr, _ := RingOneAddress(p.Zone(), p.ID)
+
+		if buf.Len() > 0 {
+			_, _ = buf.WriteRune(' ')
+		}
+		_, _ = buf.WriteString(addr.String())
+
+		return false
+	})
+	return buf.String()
+}
+
+func genEnvZoneCephMonIDs(m Machines) string {
+	var buf strings.Builder
+	m.ForEachMachine(func(p *Machine) bool {
+		if buf.Len() > 0 {
+			_, _ = buf.WriteRune(' ')
+		}
+		_, _ = fmt.Fprintf(&buf, "%v", p.ID)
+
+		return false
+	})
+	return buf.String()
+}

pkg/cluster/errors.go

@@ -0,0 +1,16 @@
+package cluster
+
+import "errors"
+
+var (
+	// ErrInvalidName indicates the name isn't valid
+	ErrInvalidName = errors.New("invalid name")
+
+	// ErrUnknownNode indicates there is a reference to a node
+	// we don't have on the tree
+	ErrUnknownNode = errors.New("node does not exist")
+
+	// ErrInvalidNode indicates the nodes can't be used for
+	// the intended purpose
+	ErrInvalidNode = errors.New("invalid node")
+)

pkg/cluster/log.go

@@ -0,0 +1,49 @@
+package cluster
+
+import "darvaza.org/slog"
+
+type logger interface {
+	withDebug() (slog.Logger, bool)
+	withInfo() (slog.Logger, bool)
+
+	debug() slog.Logger
+	info() slog.Logger
+	warn(error) slog.Logger
+	error(error) slog.Logger
+}
+
+var (
+	_ logger = (*Zones)(nil)
+)
+
+func (z *Zones) withDebug() (slog.Logger, bool) {
+	return z.debug().WithEnabled()
+}
+
+func (z *Zones) withInfo() (slog.Logger, bool) {
+	return z.debug().WithEnabled()
+}
+
+func (z *Zones) debug() slog.Logger {
+	return z.log.Debug()
+}
+
+func (z *Zones) info() slog.Logger {
+	return z.log.Info()
+}
+
+func (z *Zones) warn(err error) slog.Logger {
+	l := z.log.Warn()
+	if err != nil {
+		l = l.WithField(slog.ErrorFieldName, err)
+	}
+	return l
+}
+
+func (z *Zones) error(err error) slog.Logger {
+	l := z.log.Error()
+	if err != nil {
+		l = l.WithField(slog.ErrorFieldName, err)
+	}
+	return l
+}

pkg/cluster/machine.go

@@ -0,0 +1,74 @@
+package cluster
+
+import (
+	"net/netip"
+	"strings"
+)
+
+// revive:disable:line-length-limit
+
+// A Machine is a machine on a Zone
+type Machine struct {
+	zone   *Zone
+	logger `json:"-" yaml:"-"`
+
+	ID   int
+	Name string `json:"-" yaml:"-"`
+
+	CephMonitor     bool         `json:"ceph_monitor,omitempty" yaml:"ceph_monitor,omitempty"`
+	PublicAddresses []netip.Addr `json:"public,omitempty"       yaml:"public,omitempty"`
+	Rings           []*RingInfo  `json:"rings,omitempty"        yaml:"rings,omitempty"`
+}
+
+// revive:enable:line-length-limit
+
+func (m *Machine) String() string {
+	return m.Name
+}
+
+// FullName returns the Name of the machine including domain name
+func (m *Machine) FullName() string {
+	if domain := m.zone.zones.domain; domain != "" {
+		var s = []string{
+			m.Name,
+			domain,
+		}
+
+		return strings.Join(s, ".")
+	}
+
+	return m.Name
+}
+
+// IsGateway tells if the Machine is a ring0 gateway
+func (m *Machine) IsGateway() bool {
+	_, ok := m.getRingInfo(0)
+	return ok
+}
+
+// SetGateway enables/disables a Machine ring0 integration
+func (m *Machine) SetGateway(enabled bool) error {
+	ri, found := m.getRingInfo(0)
+	switch {
+	case !found && !enabled:
+		return nil
+	case !found:
+		var err error
+
+		if ri, err = m.createRingInfo(0, false); err != nil {
+			return err
+		}
+	}
+
+	ri.Enabled = enabled
+	return m.SyncWireguardConfig(0)
+}
+
+// Zone indicates the [Zone] this machine belongs to
+func (m *Machine) Zone() int {
+	return m.zone.ID
+}
+
+func (m *Machine) getPeerByName(name string) (*Machine, bool) {
+	return m.zone.zones.GetMachineByName(name)
+}

pkg/cluster/machine_file.go

@@ -0,0 +1,87 @@
+package cluster
+
+import (
+	"bytes"
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+
+	fs "github.com/hack-pad/hackpadfs"
+)
+
+// OpenFile opens a file on the machine's config directory with the specified flags
+func (m *Machine) OpenFile(name string, flags int, args ...any) (fs.File, error) {
+	base := m.zone.zones.dir
+	fullName := m.getFilename(name, args...)
+
+	return fs.OpenFile(base, fullName, flags, 0644)
+}
+
+// CreateTruncFile creates or truncates a file on the machine's config directory
+func (m *Machine) CreateTruncFile(name string, args ...any) (io.WriteCloser, error) {
+	return m.openWriter(name, os.O_CREATE|os.O_TRUNC, args...)
+}
+
+// CreateFile creates a file on the machine's config directory
+func (m *Machine) CreateFile(name string, args ...any) (io.WriteCloser, error) {
+	return m.openWriter(name, os.O_CREATE, args...)
+}
+
+func (m *Machine) openWriter(name string, flags int, args ...any) (io.WriteCloser, error) {
+	f, err := m.OpenFile(name, os.O_WRONLY|flags, args...)
+	if err != nil {
+		return nil, err
+	}
+
+	return f.(io.WriteCloser), nil
+}
+
+// RemoveFile deletes a file from the machine's config directory
+func (m *Machine) RemoveFile(name string, args ...any) error {
+	base := m.zone.zones.dir
+	fullName := m.getFilename(name, args...)
+	err := fs.Remove(base, fullName)
+
+	switch {
+	case os.IsNotExist(err):
+		return nil
+	default:
+		return err
+	}
+}
+
+// ReadFile reads a file from the machine's config directory
+func (m *Machine) ReadFile(name string, args ...any) ([]byte, error) {
+	base := m.zone.zones.dir
+	fullName := m.getFilename(name, args...)
+
+	return fs.ReadFile(base, fullName)
+}
+
+// WriteStringFile writes the given content to a file on the machine's config directory
+func (m *Machine) WriteStringFile(value string, name string, args ...any) error {
+	f, err := m.CreateTruncFile(name, args...)
+	if err != nil {
+		return err
+	}
+	defer f.Close()
+
+	buf := bytes.NewBufferString(value)
+	_, err = buf.WriteTo(f)
+	return err
+}
+
+func (m *Machine) getFilename(name string, args ...any) string {
+	if len(args) > 0 {
+		name = fmt.Sprintf(name, args...)
+	}
+
+	s := []string{
+		m.zone.Name,
+		m.Name,
+		name,
+	}
+
+	return filepath.Join(s...)
+}

pkg/cluster/machine_rings.go

@@ -0,0 +1,262 @@
+package cluster
+
+import (
+	"bytes"
+	"errors"
+	"fmt"
+	"os"
+
+	"darvaza.org/core"
+
+	"git.jpi.io/amery/jpictl/pkg/wireguard"
+)
+
+// GetWireguardKeys reads a wgN.key/wgN.pub files
+func (m *Machine) GetWireguardKeys(ring int) (wireguard.KeyPair, error) {
+	var (
+		data []byte
+		err  error
+		out  wireguard.KeyPair
+	)
+
+	data, err = m.ReadFile("wg%v.key", ring)
+	if err != nil {
+		// failed to read
+		return out, err
+	}
+
+	out.PrivateKey, err = wireguard.PrivateKeyFromBase64(string(data))
+	if err != nil {
+		// bad key
+		err = core.Wrapf(err, "wg%v.key", ring)
+		return out, err
+	}
+
+	data, err = m.ReadFile("wg%v.pub", ring)
+	switch {
+	case os.IsNotExist(err):
+		// no wgN.pub is fine
+	case err != nil:
+		// failed to read
+		return out, err
+	default:
+		// good read
+		out.PublicKey, err = wireguard.PublicKeyFromBase64(string(data))
+		if err != nil {
+			// bad key
+			err = core.Wrapf(err, "wg%v.pub", ring)
+			return out, err
+		}
+	}
+
+	err = out.Validate()
+	return out, err
+}
+
+func (m *Machine) tryReadWireguardKeys(ring int) error {
+	kp, err := m.GetWireguardKeys(ring)
+	switch {
+	case os.IsNotExist(err):
+		// ignore
+		return nil
+	case err != nil:
+		// something went wrong
+		return err
+	default:
+		// import keys
+		ri := &RingInfo{
+			Ring: ring,
+			Keys: kp,
+		}
+
+		return m.applyRingInfo(ring, ri)
+	}
+}
+
+// RemoveWireguardKeys deletes wgN.key and wgN.pub from
+// the machine's config directory
+func (m *Machine) RemoveWireguardKeys(ring int) error {
+	var err error
+
+	err = m.RemoveFile("wg%v.pub", ring)
+	switch {
+	case os.IsNotExist(err):
+		// ignore
+	case err != nil:
+		return err
+	}
+
+	err = m.RemoveFile("wg%v.key", ring)
+	if os.IsNotExist(err) {
+		// ignore
+		err = nil
+	}
+
+	return err
+}
+
+// GetWireguardConfig reads a wgN.conf file
+func (m *Machine) GetWireguardConfig(ring int) (*wireguard.Config, error) {
+	data, err := m.ReadFile("wg%v.conf", ring)
+	if err != nil {
+		return nil, err
+	}
+
+	r := bytes.NewReader(data)
+	return wireguard.NewConfigFromReader(r)
+}
+
+func (m *Machine) tryApplyWireguardConfig(ring int) error {
+	wg, err := m.GetWireguardConfig(ring)
+	switch {
+	case os.IsNotExist(err):
+		return nil
+	case err != nil:
+		return err
+	default:
+		return m.applyWireguardConfig(ring, wg)
+	}
+}
+
+func (m *Machine) applyWireguardConfig(ring int, wg *wireguard.Config) error {
+	addr := wg.GetAddress()
+	zoneID, nodeID, ok := Rings[ring].Decode(addr)
+	if !ok {
+		return fmt.Errorf("%s: invalid address", addr)
+	}
+
+	if err := m.applyZoneNodeID(zoneID, nodeID); err != nil {
+		return core.Wrapf(err, "%s: invalid address", addr)
+	}
+
+	if err := m.applyWireguardInterfaceConfig(ring, wg.Interface); err != nil {
+		return core.Wrap(err, "interface")
+	}
+
+	for _, peer := range wg.Peer {
+		err := m.applyWireguardPeerConfig(ring, peer)
+		switch {
+		case errors.Is(err, ErrUnknownNode):
+			// ignore unknown peers
+			m.warn(nil).
+				WithField("subsystem", "wireguard").
+				WithField("node", m.Name).
+				WithField("peer", peer.Endpoint.Host).
+				WithField("ring", ring).
+				Print("ignoring unknown endpoint")
+		case err != nil:
+			return core.Wrap(err, "peer")
+		}
+	}
+
+	return nil
+}
+
+func (m *Machine) getRingInfo(ring int) (*RingInfo, bool) {
+	for _, ri := range m.Rings {
+		if ri.Ring == ring {
+			return ri, ri.Enabled
+		}
+	}
+
+	return nil, false
+}
+
+func (m *Machine) applyRingInfo(ring int, new *RingInfo) error {
+	cur, _ := m.getRingInfo(ring)
+	if cur == nil {
+		// first, append
+		m.debug().
+			WithField("node", m.Name).
+			WithField("ring", ring).
+			Print("found")
+		m.Rings = append(m.Rings, new)
+		return nil
+	}
+
+	// extra, merge
+	return cur.Merge(new)
+}
+
+func (m *Machine) applyWireguardInterfaceConfig(ring int, data wireguard.InterfaceConfig) error {
+	ri := &RingInfo{
+		Ring:    ring,
+		Enabled: true,
+		Keys: wireguard.KeyPair{
+			PrivateKey: data.PrivateKey,
+		},
+	}
+
+	return m.applyRingInfo(ring, ri)
+}
+
+func (m *Machine) applyWireguardPeerConfig(ring int, pc wireguard.PeerConfig) error {
+	peer, found := m.getPeerByName(pc.Endpoint.Name())
+	switch {
+	case !found:
+		// unknown
+		return core.Wrap(ErrUnknownNode, pc.Endpoint.Host)
+	case ring == 1 && m.zone != peer.zone:
+		// invalid zone
+		return core.Wrap(ErrInvalidNode, peer.Name)
+	default:
+		// apply RingInfo
+		ri := &RingInfo{
+			Ring:    ring,
+			Enabled: true,
+			Keys: wireguard.KeyPair{
+				PublicKey: pc.PublicKey,
+			},
+		}
+
+		return peer.applyRingInfo(ring, ri)
+	}
+}
+
+func (m *Machine) applyZoneNodeID(zoneID, nodeID int) error {
+	switch {
+	case zoneID == 0:
+		return fmt.Errorf("invalid %s", "zoneID")
+	case nodeID == 0:
+		return fmt.Errorf("invalid %s", "nodeID")
+	case m.ID != nodeID:
+		return fmt.Errorf("invalid %s: %v ≠ %v", "zoneID", m.ID, nodeID)
+	case m.zone.ID != 0 && m.zone.ID != zoneID:
+		return fmt.Errorf("invalid %s: %v ≠ %v", "zoneID", m.zone.ID, zoneID)
+	case m.zone.ID == 0:
+		m.zone.ID = zoneID
+	}
+
+	return nil
+}
+
+// RemoveWireguardConfig deletes wgN.conf from the machine's
+// config directory.
+func (m *Machine) RemoveWireguardConfig(ring int) error {
+	err := m.RemoveFile("wg%v.conf", ring)
+	if os.IsNotExist(err) {
+		err = nil
+	}
+
+	return err
+}
+
+func (m *Machine) createRingInfo(ring int, enabled bool) (*RingInfo, error) {
+	keys, err := wireguard.NewKeyPair()
+	if err != nil {
+		return nil, err
+	}
+
+	ri := &RingInfo{
+		Ring:    ring,
+		Enabled: enabled,
+		Keys:    keys,
+	}
+
+	err = m.applyRingInfo(ring, ri)
+	if err != nil {
+		return nil, err
+	}
+
+	return ri, nil
+}

pkg/cluster/machine_scan.go

@@ -0,0 +1,88 @@
+package cluster
+
+import (
+	"context"
+	"net/netip"
+	"strconv"
+	"strings"
+	"time"
+
+	"darvaza.org/core"
+)
+
+// LookupNetIP uses the DNS Resolver to get the public addresses associated
+// to a Machine
+func (m *Machine) LookupNetIP(timeout time.Duration) ([]netip.Addr, error) {
+	ctx, cancel := context.WithTimeout(context.Background(), timeout)
+
+	defer cancel()
+
+	return m.zone.zones.resolver.LookupNetIP(ctx, "ip", m.FullName())
+}
+
+// UpdatePublicAddresses uses the DNS Resolver to set Machine.PublicAddresses
+func (m *Machine) UpdatePublicAddresses() error {
+	addrs, err := m.LookupNetIP(2 * time.Second)
+	if err != nil {
+		return err
+	}
+
+	m.PublicAddresses = addrs
+	return nil
+}
+
+func (m *Machine) init() error {
+	if err := m.setID(); err != nil {
+		return core.Wrap(err, m.Name)
+	}
+
+	for i := 0; i < RingsCount; i++ {
+		if err := m.tryReadWireguardKeys(i); err != nil {
+			return core.Wrap(err, m.Name)
+		}
+	}
+
+	return nil
+}
+
+func (m *Machine) setID() error {
+	zoneName := m.zone.Name
+
+	l := len(zoneName)
+	switch {
+	case len(m.Name) < l+2:
+		return ErrInvalidName
+	case !strings.HasPrefix(m.Name, zoneName):
+		return ErrInvalidName
+	case m.Name[l] != '-':
+		return ErrInvalidName
+	}
+
+	suffix := m.Name[l+1:]
+	id, err := strconv.ParseInt(suffix, 10, 8)
+	if err != nil {
+		return err
+	}
+
+	m.ID = int(id)
+	return nil
+}
+
+func (m *Machine) scan(opts *ScanOptions) error {
+	for i := 0; i < RingsCount; i++ {
+		if err := m.tryApplyWireguardConfig(i); err != nil {
+			m.error(err).
+				WithField("subsystem", "wireguard").
+				WithField("node", m.Name).
+				WithField("ring", i).
+				Print()
+			return err
+		}
+	}
+
+	if !opts.DontResolvePublicAddresses {
+		return m.UpdatePublicAddresses()
+	}
+
+	return nil
+}

pkg/cluster/options.go

@@ -0,0 +1,134 @@
+package cluster
+
+import (
+	"io/fs"
+	"path/filepath"
+
+	"darvaza.org/resolver"
+	"darvaza.org/slog"
+	"darvaza.org/slog/handlers/discard"
+	"github.com/hack-pad/hackpadfs/os"
+)
+
+// A ScanOption pre-configures the Zones before scanning
+type ScanOption func(*Zones, *ScanOptions) error
+
+// ScanOptions contains flags used by the initial scan
+type ScanOptions struct {
+	// DontResolvePublicAddresses indicates we shouldn't
+	// pre-populate Machine.PublicAddresses during the
+	// initial scan
+	DontResolvePublicAddresses bool
+
+	// Logger specifies the logger to be used. otherwise
+	// the scanner will be mute
+	slog.Logger
+}
+
+// ResolvePublicAddresses instructs the scanner to use
+// the DNS resolver to get PublicAddresses of nodes.
+// Default is true
+func ResolvePublicAddresses(resolve bool) ScanOption {
+	return func(m *Zones, opt *ScanOptions) error {
+		opt.DontResolvePublicAddresses = !resolve
+		return nil
+	}
+}
+
+// WithLookuper specifies what resolver.Lookuper to use to
+// find public addresses
+func WithLookuper(h resolver.Lookuper) ScanOption {
+	return func(m *Zones, opt *ScanOptions) error {
+		if h == nil {
+			return fs.ErrInvalid
+		}
+		m.resolver = resolver.NewResolver(h)
+		return nil
+	}
+}
+
+// WithResolver specifies what resolver to use to find
+// public addresses. if nil is passed, the [net.Resolver] will be used.
+// The default is using Cloudflare's 1.1.1.1.
+func WithResolver(h resolver.Resolver) ScanOption {
+	return func(m *Zones, opt *ScanOptions) error {
+		if h == nil {
+			h = resolver.SystemResolver(true)
+		}
+
+		m.resolver = h
+		return nil
+	}
+}
+
+// WithLogger specifies what to use for logging
+func WithLogger(log slog.Logger) ScanOption {
+	return func(m *Zones, opt *ScanOptions) error {
+		if log == nil {
+			log = discard.New()
+		}
+
+		opt.Logger = log
+		m.log = log
+		return nil
+	}
+}
+
+func (m *Zones) setDefaults(opt *ScanOptions) error {
+	if m.resolver == nil {
+		h := resolver.NewCloudflareLookuper()
+
+		if err := WithLookuper(h)(m, opt); err != nil {
+			return err
+		}
+	}
+
+	if opt.Logger == nil {
+		if err := WithLogger(nil)(m, opt); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+// NewFS builds a [Zones] tree using the given directory
+func NewFS(dir fs.FS, domain string, opts ...ScanOption) (*Zones, error) {
+	var scanOptions ScanOptions
+
+	z := &Zones{
+		dir:    dir,
+		domain: domain,
+	}
+
+	for _, opt := range opts {
+		if err := opt(z, &scanOptions); err != nil {
+			return nil, err
+		}
+	}
+
+	if err := z.setDefaults(&scanOptions); err != nil {
+		return nil, err
+	}
+
+	if err := z.scan(&scanOptions); err != nil {
+		return nil, err
+	}
+
+	return z, nil
+}
+
+// New builds a [Zones] tree using the given directory
+func New(dir, domain string, opts ...ScanOption) (*Zones, error) {
+	dir, err := filepath.Abs(dir)
+	if err != nil {
+		return nil, err
+	}
+
+	base, err := os.NewFS().Sub(dir[1:])
+	if err != nil {
+		return nil, err
+	}
+
+	return NewFS(base, domain, opts...)
+}

pkg/cluster/rings.go

@@ -0,0 +1,349 @@
+package cluster
+
+import (
+	"fmt"
+	"io/fs"
+	"net/netip"
+
+	"git.jpi.io/amery/jpictl/pkg/wireguard"
+)
+
+const (
+	// MaxZoneID indicates the highest ID allowed for a Zone
+	MaxZoneID = 0xf
+	// MaxNodeID indicates the highest Machine ID allowed within a Zone
+	MaxNodeID = 0xff - 1
+	// RingsCount indicates how many wireguard rings we have
+	RingsCount = 2
+	// RingZeroPort is the port wireguard uses for ring0
+	RingZeroPort = 51800
+	// RingOnePort is the port wireguard uses for ring1
+	RingOnePort = 51810
+)
+
+// RingInfo contains represents the Wireguard endpoint details
+// for a Machine on a particular ring
+type RingInfo struct {
+	Ring    int
+	Enabled bool
+	Keys    wireguard.KeyPair
+}
+
+// Merge attempts to combine two RingInfo structs
+func (ri *RingInfo) Merge(alter *RingInfo) error {
+	switch {
+	case alter == nil:
+		return nil
+	case ri.Ring != alter.Ring:
+		// different ring
+		return fmt.Errorf("invalid %s: %v ≠ %v", "ring", ri.Ring, alter.Ring)
+	case ri.Enabled && !alter.Enabled:
+		// can't disable via Merge
+		return fmt.Errorf("invalid %s: %v → %v", "enabled", ri.Enabled, alter.Enabled)
+	case !canMergeKeyPairs(ri.Keys, alter.Keys):
+		// incompatible keypairs
+		return fmt.Errorf("invalid %s: %s ≠ %s", "keys", ri.Keys, alter.Keys)
+	}
+
+	return ri.unsafeMerge(alter)
+}
+
+func (ri *RingInfo) unsafeMerge(alter *RingInfo) error {
+	// enable via Merge
+	if alter.Enabled {
+		ri.Enabled = true
+	}
+
+	// fill the gaps on our keypair
+	if ri.Keys.PrivateKey.IsZero() {
+		ri.Keys.PrivateKey = alter.Keys.PrivateKey
+	}
+	if ri.Keys.PublicKey.IsZero() {
+		ri.Keys.PublicKey = alter.Keys.PublicKey
+	}
+
+	return nil
+}
+
+func canMergeKeyPairs(p1, p2 wireguard.KeyPair) bool {
+	switch {
+	case !p1.PrivateKey.IsZero() && !p2.PrivateKey.IsZero() && !p1.PrivateKey.Equal(p2.PrivateKey):
+		return false
+	case !p1.PublicKey.IsZero() && !p2.PublicKey.IsZero() && !p1.PublicKey.Equal(p2.PublicKey):
+		return false
+	default:
+		return true
+	}
+}
+
+// RingAddressEncoder provides encoder/decoder access for a particular
+// Wireguard ring
+type RingAddressEncoder struct {
+	ID     int
+	Port   uint16
+	Encode func(zoneID, nodeID int) (netip.Addr, bool)
+	Decode func(addr netip.Addr) (zoneID, nodeID int, ok bool)
+}
+
+var (
+	// RingZero is a wg0 address encoder/decoder
+	RingZero = RingAddressEncoder{
+		ID:     0,
+		Port:   RingZeroPort,
+		Decode: ParseRingZeroAddress,
+		Encode: RingZeroAddress,
+	}
+	// RingOne is a wg1 address encoder/decoder
+	RingOne = RingAddressEncoder{
+		ID:     1,
+		Port:   RingOnePort,
+		Decode: ParseRingOneAddress,
+		Encode: RingOneAddress,
+	}
+	// Rings provides indexed access to the ring address encoders
+	Rings = [RingsCount]RingAddressEncoder{
+		RingZero,
+		RingOne,
+	}
+)
+
+// ValidZoneID checks if the given zoneID is a valid 4 bit zone number.
+//
+// 0 is reserved, and only allowed when composing CIDRs.
+func ValidZoneID(zoneID int) bool {
+	switch {
+	case zoneID < 0 || zoneID > MaxZoneID:
+		return false
+	default:
+		return true
+	}
+}
+
+// ValidNodeID checks if the given nodeID is a valid 8 bit number.
+// nodeID is unique within a Zone.
+// 0 is reserved, and only allowed when composing CIDRs.
+func ValidNodeID(nodeID int) bool {
+	switch {
+	case nodeID < 0 || nodeID > MaxNodeID:
+		return false
+	default:
+		return true
+	}
+}
+
+// ParseRingZeroAddress extracts zone and node ID from a wg0 [netip.Addr]
+// wg0 addresses are of the form `10.0.{{zoneID}}.{{nodeID}}`
+func ParseRingZeroAddress(addr netip.Addr) (zoneID int, nodeID int, ok bool) {
+	if addr.IsValid() {
+		a4 := addr.As4()
+
+		if a4[0] == 10 && a4[1] == 0 {
+			return int(a4[2]), int(a4[3]), true
+		}
+	}
+	return 0, 0, false
+}
+
+// RingZeroAddress returns a wg0 IP address
+func RingZeroAddress(zoneID, nodeID int) (netip.Addr, bool) {
+	switch {
+	case !ValidZoneID(zoneID) || !ValidNodeID(nodeID):
+		return netip.Addr{}, false
+	default:
+		a4 := [4]uint8{10, 0, uint8(zoneID), uint8(nodeID)}
+		return netip.AddrFrom4(a4), true
+	}
+}
+
+// ParseRingOneAddress extracts zone and node ID from a wg1 [netip.Addr]
+// wg1 addresses are of the form `10.{{zoneID << 4}}.{{nodeID}}`
+func ParseRingOneAddress(addr netip.Addr) (zoneID int, nodeID int, ok bool) {
+	if addr.IsValid() {
+		a4 := addr.As4()
+
+		if a4[0] == 10 && a4[2] == 0 {
+			zoneID = int(a4[1] >> 4)
+			nodeID = int(a4[3])
+			return zoneID, nodeID, true
+		}
+	}
+	return 0, 0, false
+}
+
+// RingOneAddress returns a wg1 IP address
+func RingOneAddress(zoneID, nodeID int) (netip.Addr, bool) {
+	switch {
+	case !ValidZoneID(zoneID) || !ValidNodeID(nodeID):
+		return netip.Addr{}, false
+	default:
+		a4 := [4]uint8{10, uint8(zoneID << 4), 0, uint8(nodeID)}
+		return netip.AddrFrom4(a4), true
+	}
+}
+
+var (
+	_ MachineIterator = (*Ring)(nil)
+	_ ZoneIterator    = (*Ring)(nil)
+)
+
+// A Ring describes all peers on a ring
+type Ring struct {
+	RingAddressEncoder
+	ZoneIterator
+
+	Peers []*RingPeer
+}
+
+// AddPeer adds a [Machine] to the ring
+func (r *Ring) AddPeer(p *Machine) bool {
+	ri, ok := p.getRingInfo(r.ID)
+	if !ok {
+		return false
+	}
+
+	nodeID := p.ID
+	zoneID := p.Zone()
+	addr, _ := r.Encode(zoneID, nodeID)
+
+	rp := &RingPeer{
+		Node:       p,
+		Address:    addr,
+		PrivateKey: ri.Keys.PrivateKey,
+		PeerConfig: wireguard.PeerConfig{
+			Name:      fmt.Sprintf("%s-%v", p.Name, r.ID),
+			PublicKey: ri.Keys.PublicKey,
+			Endpoint: wireguard.EndpointAddress{
+				Host: p.FullName(),
+				Port: r.Port,
+			},
+		},
+	}
+
+	switch {
+	case r.ID == 0:
+		r.setRingZeroAllowedIPs(rp)
+	case p.IsGateway():
+		r.setRingOneGatewayAllowedIPs(rp)
+	default:
+		r.setRingOneNodeAllowedIPs(rp)
+	}
+
+	r.Peers = append(r.Peers, rp)
+	return true
+}
+
+func (r *Ring) setRingZeroAllowedIPs(rp *RingPeer) {
+	zoneID, _, _ := r.Decode(rp.Address)
+
+	// everyone on ring0 is a gateway to ring1
+	addr, _ := RingOneAddress(zoneID, 0)
+	rp.AllowCIDR(addr, 12)
+
+	// peer
+	rp.AllowCIDR(rp.Address, 32)
+}
+
+func (r *Ring) setRingOneGatewayAllowedIPs(rp *RingPeer) {
+	zoneID, _, _ := r.Decode(rp.Address)
+
+	// peer
+	rp.AllowCIDR(rp.Address, 32)
+
+	// ring1 gateways connect to all other ring1 networks
+	r.ForEachZone(func(z *Zone) bool {
+		if z.ID != zoneID {
+			addr, _ := r.Encode(z.ID, 0)
+			rp.AllowCIDR(addr, 12)
+		}
+		return false
+	})
+
+	// ring1 gateways also connect to all ring0 addresses
+	r.ForEachZone(func(z *Zone) bool {
+		z.ForEachMachine(func(p *Machine) bool {
+			if p.IsGateway() {
+				addr, _ := RingZeroAddress(z.ID, p.ID)
+				rp.AllowCIDR(addr, 32)
+			}
+			return false
+		})
+		return false
+	})
+}
+
+func (*Ring) setRingOneNodeAllowedIPs(rp *RingPeer) {
+	// only to the peer itself
+	rp.AllowCIDR(rp.Address, 32)
+}
+
+// ForEachMachine calls a function for each Machine in the ring
+// until instructed to terminate the loop
+func (r *Ring) ForEachMachine(fn func(*Machine) bool) {
+	for _, pp := range r.Peers {
+		if fn(pp.Node) {
+			return
+		}
+	}
+}
+
+// ExportConfig builds a wgN.conf for the specified machine on the ring
+func (r *Ring) ExportConfig(p *Machine) (*wireguard.Config, error) {
+	var found bool
+
+	out := &wireguard.Config{
+		Interface: wireguard.InterfaceConfig{
+			ListenPort: r.Port,
+		},
+	}
+
+	for _, pp := range r.Peers {
+		switch {
+		case pp.Node == p:
+			// current
+			found = true
+			out.Interface.Name = pp.PeerConfig.Name
+			out.Interface.Address = pp.Address
+			out.Interface.PrivateKey = pp.PrivateKey
+		default:
+			// peer
+			pc := pp.PeerConfig
+			out.Peer = append(out.Peer, pc)
+		}
+	}
+
+	if !found {
+		return nil, fs.ErrNotExist
+	}
+
+	return out, nil
+}
+
+// A RingPeer is a node on a [Ring]
+type RingPeer struct {
+	Node *Machine
+
+	Address    netip.Addr
+	PrivateKey wireguard.PrivateKey
+	PeerConfig wireguard.PeerConfig
+}
+
+// AllowCIDR allows an IP range via this peer
+func (rp *RingPeer) AllowCIDR(addr netip.Addr, bits int) {
+	cidr := netip.PrefixFrom(addr, bits)
+	rp.PeerConfig.AllowedIPs = append(rp.PeerConfig.AllowedIPs, cidr)
+}
+
+// NewRing composes a new Ring for Wireguard setup
+func NewRing(z ZoneIterator, m MachineIterator, ring int) (*Ring, error) {
+	r := &Ring{
+		RingAddressEncoder: Rings[ring],
+		ZoneIterator:       z,
+	}
+
+	m.ForEachMachine(func(p *Machine) bool {
+		r.AddPeer(p)
+		return false
+	})
+
+	return r, nil
+}

pkg/cluster/scan.go

@@ -0,0 +1,217 @@
+package cluster
+
+import (
+	"io/fs"
+	"sort"
+
+	"darvaza.org/core"
+)
+
+func (m *Zones) scan(opts *ScanOptions) error {
+	for _, fn := range []func(*ScanOptions) error{
+		m.scanDirectory,
+		m.scanMachines,
+		m.scanZoneIDs,
+		m.scanSort,
+		m.scanGateways,
+		m.scanCephMonitors,
+	} {
+		if err := fn(opts); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+func (m *Zones) scanDirectory(_ *ScanOptions) error {
+	// each directory is a zone
+	entries, err := fs.ReadDir(m.dir, ".")
+	if err != nil {
+		return err
+	}
+
+	for _, e := range entries {
+		if e.IsDir() {
+			z, err := m.newZone(e.Name())
+			switch {
+			case err != nil:
+				return core.Wrap(err, e.Name())
+			case z.Machines.Len() == 0:
+				z.warn(nil).
+					WithField("zone", z.Name).
+					Print("empty")
+			default:
+				m.Zones = append(m.Zones, z)
+			}
+		}
+	}
+
+	return nil
+}
+
+func (m *Zones) newZone(name string) (*Zone, error) {
+	z := &Zone{
+		zones:  m,
+		logger: m,
+		Name:   name,
+	}
+
+	z.debug().
+		WithField("zone", z.Name).
+		Print("found")
+
+	if err := z.scan(); err != nil {
+		return nil, err
+	}
+	return z, nil
+}
+
+func (m *Zones) scanMachines(opts *ScanOptions) error {
+	var err error
+	m.ForEachMachine(func(p *Machine) bool {
+		err = p.scan(opts)
+		return err != nil
+	})
+	return err
+}
+
+func (m *Zones) scanZoneIDs(_ *ScanOptions) error {
+	var hasMissing bool
+	var lastZoneID int
+
+	m.ForEachZone(func(z *Zone) bool {
+		switch {
+		case z.ID == 0:
+			hasMissing = true
+		case z.ID > lastZoneID:
+			lastZoneID = z.ID
+		}
+
+		return false
+	})
+
+	if hasMissing {
+		next := lastZoneID + 1
+
+		m.ForEachZone(func(z *Zone) bool {
+			if z.ID == 0 {
+				z.ID, next = next, next+1
+			}
+
+			return false
+		})
+	}
+
+	return nil
+}
+
+func (m *Zones) scanSort(_ *ScanOptions) error {
+	sort.SliceStable(m.Zones, func(i, j int) bool {
+		id1 := m.Zones[i].ID
+		id2 := m.Zones[j].ID
+		return id1 < id2
+	})
+
+	m.ForEachZone(func(z *Zone) bool {
+		sort.Sort(z)
+		return false
+	})
+
+	m.ForEachMachine(func(p *Machine) bool {
+		sort.SliceStable(p.Rings, func(i, j int) bool {
+			ri1 := p.Rings[i]
+			ri2 := p.Rings[j]
+
+			return ri1.Ring < ri2.Ring
+		})
+
+		return false
+	})
+
+	return nil
+}
+
+func (m *Zones) scanGateways(_ *ScanOptions) error {
+	var err error
+
+	m.ForEachZone(func(z *Zone) bool {
+		_, _, err = z.GetGateway()
+		return err != nil
+	})
+	return err
+}
+
+func (z *Zone) scan() error {
+	// each directory is a machine
+	entries, err := fs.ReadDir(z.zones.dir, z.Name)
+	if err != nil {
+		return err
+	}
+
+	for _, e := range entries {
+		if e.IsDir() {
+			m := &Machine{
+				zone:   z,
+				logger: z,
+				Name:   e.Name(),
+			}
+
+			m.debug().
+				WithField("node", m.Name).
+				WithField("zone", z.Name).
+				Print("found")
+
+			if err := m.init(); err != nil {
+				m.error(err).
+					WithField("node", m.Name).
+					WithField("zone", z.Name).
+					Print()
+
+				return err
+			}
+
+			z.Machines = append(z.Machines, m)
+		}
+	}
+
+	return nil
+}
+
+// GetGateway returns the first gateway found, if none
+// files will be created to enable the first [Machine] to
+// be one
+func (z *Zone) GetGateway() (*Machine, bool, error) {
+	var first *Machine
+	var gateway *Machine
+
+	z.zones.ForEachMachine(func(p *Machine) bool {
+		switch {
+		case p.IsGateway():
+			// found
+			gateway = p
+		case first == nil:
+			// remember
+			first = p
+		default:
+			// keep looking
+		}
+
+		return gateway != nil
+	})
+
+	switch {
+	case gateway != nil:
+		// found one
+		return gateway, false, nil
+	case first != nil:
+		// make one
+		if err := first.SetGateway(true); err != nil {
+			return first, false, err
+		}
+		return first, true, nil
+	default:
+		// Zone without nodes?
+		panic("unreachable")
+	}
+}

pkg/cluster/sync.go

@@ -0,0 +1,44 @@
+package cluster
+
+// SyncAll updates all config files
+func (m *Zones) SyncAll() error {
+	for _, fn := range []func() error{
+		m.SyncAllWireguard,
+		m.SyncAllCeph,
+	} {
+		if err := fn(); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+// SyncAllWireguard updates all wireguard config files
+func (m *Zones) SyncAllWireguard() error {
+	var err error
+
+	for ring := 0; ring < RingsCount; ring++ {
+		err = m.WriteWireguardKeys(ring)
+		if err != nil {
+			return err
+		}
+
+		err = m.SyncWireguardConfig(ring)
+		if err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+// SyncAllCeph updates the ceph.conf file
+func (m *Zones) SyncAllCeph() error {
+	cfg, err := m.GenCephConfig()
+	if err != nil {
+		return err
+	}
+
+	return m.WriteCephConfig(cfg)
+}

pkg/cluster/wireguard.go

@@ -0,0 +1,272 @@
+package cluster
+
+import (
+	"io/fs"
+	"os"
+)
+
+var (
+	_ WireguardConfigPruner = (*Zones)(nil)
+	_ WireguardConfigPruner = (*Zone)(nil)
+	_ WireguardConfigPruner = (*Machine)(nil)
+
+	_ WireguardConfigWriter = (*Zones)(nil)
+	_ WireguardConfigWriter = (*Zone)(nil)
+	_ WireguardConfigWriter = (*Machine)(nil)
+
+	_ WireguardConfigSyncer = (*Zones)(nil)
+	_ WireguardConfigSyncer = (*Zone)(nil)
+	_ WireguardConfigSyncer = (*Machine)(nil)
+
+	_ WireguardKeysWriter = (*Zones)(nil)
+	_ WireguardKeysWriter = (*Zone)(nil)
+	_ WireguardKeysWriter = (*Machine)(nil)
+)
+
+// A WireguardConfigPruner deletes wgN.conf on all machines under
+// its scope with the specified ring disabled
+type WireguardConfigPruner interface {
+	PruneWireguardConfig(ring int) error
+}
+
+// PruneWireguardConfig removes wgN.conf files of machines with
+// the corresponding ring disabled on all zones
+func (m *Zones) PruneWireguardConfig(ring int) error {
+	return pruneWireguardConfig(m, ring)
+}
+
+// PruneWireguardConfig removes wgN.conf files of machines with
+// the corresponding ring disabled.
+func (z *Zone) PruneWireguardConfig(ring int) error {
+	return pruneWireguardConfig(z, ring)
+}
+
+func pruneWireguardConfig(m MachineIterator, ring int) error {
+	var err error
+
+	m.ForEachMachine(func(p *Machine) bool {
+		err = p.PruneWireguardConfig(ring)
+		if os.IsNotExist(err) {
+			// ignore
+			err = nil
+		}
+
+		return err != nil
+	})
+
+	return err
+}
+
+// PruneWireguardConfig deletes the wgN.conf file if its
+// presence on the ring is disabled
+func (m *Machine) PruneWireguardConfig(ring int) error {
+	_, ok := m.getRingInfo(ring)
+	if !ok {
+		return m.RemoveWireguardConfig(ring)
+	}
+
+	return nil
+}
+
+// A WireguardConfigWriter rewrites all wgN.conf on all machines under
+// its scope attached to that ring
+type WireguardConfigWriter interface {
+	WriteWireguardConfig(ring int) error
+}
+
+// WriteWireguardConfig rewrites all wgN.conf on all machines
+// attached to that ring
+func (m *Zones) WriteWireguardConfig(ring int) error {
+	switch ring {
+	case 0:
+		return writeWireguardConfig(m, m, ring)
+	case 1:
+		var err error
+		m.ForEachZone(func(z *Zone) bool {
+			err = writeWireguardConfig(m, z, ring)
+			return err != nil
+		})
+		return err
+	default:
+		return fs.ErrInvalid
+	}
+}
+
+// WriteWireguardConfig rewrites all wgN.conf on all machines
+// on the Zone attached to that ring
+func (z *Zone) WriteWireguardConfig(ring int) error {
+	switch ring {
+	case 0:
+		return writeWireguardConfig(z.zones, z.zones, ring)
+	case 1:
+		return writeWireguardConfig(z.zones, z, ring)
+	default:
+		return fs.ErrInvalid
+	}
+}
+
+func writeWireguardConfig(z ZoneIterator, m MachineIterator, ring int) error {
+	r, err := NewRing(z, m, ring)
+	if err != nil {
+		return err
+	}
+
+	r.ForEachMachine(func(p *Machine) bool {
+		err = p.writeWireguardRingConfig(r)
+		return err != nil
+	})
+
+	return err
+}
+
+// WriteWireguardConfig rewrites the wgN.conf file of this Machine
+// if enabled
+func (m *Machine) WriteWireguardConfig(ring int) error {
+	r, err := NewRing(m.zone.zones, m.zone, ring)
+	if err != nil {
+		return err
+	}
+
+	return m.writeWireguardRingConfig(r)
+}
+
+func (m *Machine) writeWireguardRingConfig(r *Ring) error {
+	wg, err := r.ExportConfig(m)
+	if err != nil {
+		return nil
+	}
+
+	f, err := m.CreateTruncFile("wg%v.conf", r.ID)
+	if err != nil {
+		return err
+	}
+	defer f.Close()
+
+	_, err = wg.WriteTo(f)
+	return err
+}
+
+// A WireguardConfigSyncer updates all wgN.conf on all machines under
+// its scope reflecting the state of the ring
+type WireguardConfigSyncer interface {
+	SyncWireguardConfig(ring int) error
+}
+
+// SyncWireguardConfig updates all wgN.conf files for the specified
+// ring
+func (m *Zones) SyncWireguardConfig(ring int) error {
+	switch ring {
+	case 0:
+		return syncWireguardConfig(m, m, ring)
+	case 1:
+		var err error
+		m.ForEachZone(func(z *Zone) bool {
+			err = syncWireguardConfig(m, z, ring)
+			return err != nil
+		})
+		return err
+	default:
+		return fs.ErrInvalid
+	}
+}
+
+// SyncWireguardConfig updates all wgN.conf files for the specified
+// ring
+func (z *Zone) SyncWireguardConfig(ring int) error {
+	switch ring {
+	case 0:
+		return syncWireguardConfig(z.zones, z.zones, ring)
+	case 1:
+		return syncWireguardConfig(z.zones, z, ring)
+	default:
+		return fs.ErrInvalid
+	}
+}
+
+func syncWireguardConfig(z ZoneIterator, m MachineIterator, ring int) error {
+	r, err := NewRing(z, m, ring)
+	if err != nil {
+		return err
+	}
+
+	r.ForEachMachine(func(p *Machine) bool {
+		if _, ok := p.getRingInfo(ring); ok {
+			err = p.writeWireguardRingConfig(r)
+		} else {
+			err = p.RemoveWireguardConfig(ring)
+		}
+		return err != nil
+	})
+
+	return err
+}
+
+// SyncWireguardConfig updates all wgN.conf files for the specified
+// ring
+func (m *Machine) SyncWireguardConfig(ring int) error {
+	return m.zone.SyncWireguardConfig(ring)
+}
+
+// A WireguardKeysWriter writes the Wireguard Keys for all machines
+// under its scope for the specified ring
+type WireguardKeysWriter interface {
+	WriteWireguardKeys(ring int) error
+}
+
+// WriteWireguardKeys rewrites all wgN.{key,pub} files
+func (m *Zones) WriteWireguardKeys(ring int) error {
+	return writeWireguardKeys(m, ring)
+}
+
+// WriteWireguardKeys rewrites all wgN.{key,pub} files on this zone
+func (z *Zone) WriteWireguardKeys(ring int) error {
+	return writeWireguardKeys(z, ring)
+}
+
+func writeWireguardKeys(m MachineIterator, ring int) error {
+	var err error
+
+	m.ForEachMachine(func(p *Machine) bool {
+		err = p.WriteWireguardKeys(ring)
+		if os.IsNotExist(err) {
+			// ignore
+			err = nil
+		}
+
+		return err != nil
+	})
+
+	return err
+}
+
+// WriteWireguardKeys writes the wgN.key/wgN.pub files
+func (m *Machine) WriteWireguardKeys(ring int) error {
+	var err error
+	var key, pub string
+	var ri *RingInfo
+
+	ri, _ = m.getRingInfo(ring)
+	if ri != nil {
+		key = ri.Keys.PrivateKey.String()
+		pub = ri.Keys.PublicKey.String()
+	}
+
+	switch {
+	case key == "":
+		return fs.ErrNotExist
+	case pub == "":
+		pub = ri.Keys.PrivateKey.Public().String()
+	}
+
+	err = m.WriteStringFile(key+"\n", "wg%v.key", ring)
+	if err != nil {
+		return err
+	}
+
+	err = m.WriteStringFile(pub+"\n", "wg%v.pub", ring)
+	if err != nil {
+		return err
+	}
+
+	return nil
+}

pkg/cluster/zones.go

@@ -0,0 +1,197 @@
+package cluster
+
+import (
+	"io/fs"
+	"sort"
+
+	"darvaza.org/resolver"
+	"darvaza.org/slog"
+	"github.com/gofrs/uuid/v5"
+)
+
+var (
+	_ MachineIterator = Machines(nil)
+	_ sort.Interface  = Machines(nil)
+
+	_ MachineIterator = (*Zone)(nil)
+	_ MachineIterator = (*Zones)(nil)
+	_ ZoneIterator    = (*Zones)(nil)
+)
+
+// A MachineIterator is a set of Machines we can iterate on
+type MachineIterator interface {
+	ForEachMachine(func(*Machine) bool)
+}
+
+// A ZoneIterator is a set of Zones we can iterate on
+type ZoneIterator interface {
+	ForEachZone(func(*Zone) bool)
+}
+
+// Machines is a list of Machine objects
+type Machines []*Machine
+
+// ForEachMachine calls a function for each Machine in the list
+// until instructed to terminate the loop
+func (m Machines) ForEachMachine(fn func(*Machine) bool) {
+	for _, p := range m {
+		if fn(p) {
+			return
+		}
+	}
+}
+
+// Len returns the number of machines in the list
+func (m Machines) Len() int {
+	return len(m)
+}
+
+// Less implements sort.Interface to sort the list
+func (m Machines) Less(i, j int) bool {
+	a, b := m[i], m[j]
+	za, zb := a.Zone(), b.Zone()
+
+	switch {
+	case za == zb:
+		return a.ID < b.ID
+	default:
+		return za < zb
+	}
+}
+
+// Swap implements sort.Interface to sort the list
+func (m Machines) Swap(i, j int) {
+	m[i], m[j] = m[j], m[i]
+}
+
+// FilterMachines produces a subset of the machines offered by the given
+// iterator fulfilling a condition
+func FilterMachines(m MachineIterator, cond func(*Machine) bool) (Machines, int) {
+	var out []*Machine
+
+	if cond == nil {
+		// unconditional
+		cond = func(*Machine) bool { return true }
+	}
+
+	m.ForEachMachine(func(p *Machine) bool {
+		if cond(p) {
+			out = append(out, p)
+		}
+		return false
+	})
+
+	return out, len(out)
+}
+
+// Zone represents one zone in a cluster
+type Zone struct {
+	zones  *Zones
+	logger `json:"-" yaml:"-"`
+
+	ID   int
+	Name string
+
+	Machines
+}
+
+func (z *Zone) String() string {
+	return z.Name
+}
+
+// SetGateway configures a machine to be the zone's ring0 gateway
+func (z *Zone) SetGateway(gatewayID int, enabled bool) error {
+	var err error
+	var found bool
+
+	z.ForEachMachine(func(p *Machine) bool {
+		if p.ID == gatewayID {
+			found = true
+			err = p.SetGateway(enabled)
+
+			return true
+		}
+		return false
+	})
+
+	switch {
+	case err != nil:
+		return err
+	case !found:
+		return fs.ErrNotExist
+	default:
+		return nil
+	}
+}
+
+// GatewayIDs returns the list of IDs of machines that act as ring0 gateways
+func (z *Zone) GatewayIDs() ([]int, int) {
+	var out []int
+	z.ForEachMachine(func(p *Machine) bool {
+		if p.IsGateway() {
+			out = append(out, p.ID)
+		}
+		return false
+	})
+
+	return out, len(out)
+}
+
+// revive:disable:line-length-limit
+
+// Zones represents all zones in a cluster
+type Zones struct {
+	dir      fs.FS
+	log      slog.Logger
+	resolver resolver.Resolver
+	domain   string
+
+	CephFSID uuid.UUID `json:"ceph_fsid,omitempty" yaml:"ceph_fsid,omitempty"`
+	Zones    []*Zone
+}
+
+// revive:enable:line-length-limit
+
+// ForEachMachine calls a function for each Machine in the cluster
+// until instructed to terminate the loop
+func (m *Zones) ForEachMachine(fn func(*Machine) bool) {
+	m.ForEachZone(func(z *Zone) bool {
+		var term bool
+
+		z.ForEachMachine(func(p *Machine) bool {
+			term = fn(p)
+			return term
+		})
+
+		return term
+	})
+}
+
+// ForEachZone calls a function for each Zone in the cluster
+// until instructed to terminate the loop
+func (m *Zones) ForEachZone(fn func(*Zone) bool) {
+	for _, p := range m.Zones {
+		if fn(p) {
+			// terminate
+			return
+		}
+	}
+}
+
+// GetMachineByName looks for a machine with the specified
+// name on any zone
+func (m *Zones) GetMachineByName(name string) (*Machine, bool) {
+	var out *Machine
+
+	if name != "" {
+		m.ForEachMachine(func(p *Machine) bool {
+			if p.Name == name {
+				out = p
+			}
+
+			return out != nil
+		})
+	}
+
+	return out, out != nil
+}

pkg/cluster/zones_file.go

@@ -0,0 +1,46 @@
+package cluster
+
+import (
+	"fmt"
+	"io"
+	"os"
+
+	fs "github.com/hack-pad/hackpadfs"
+)
+
+// OpenFile opens a file on the cluster's config directory with the specified flags
+func (m *Zones) OpenFile(name string, flags int, args ...any) (fs.File, error) {
+	if len(args) > 0 {
+		name = fmt.Sprintf(name, args...)
+	}
+
+	return fs.OpenFile(m.dir, name, flags, 0644)
+}
+
+// CreateTruncFile creates or truncates a file on the cluster's config directory
+func (m *Zones) CreateTruncFile(name string, args ...any) (io.WriteCloser, error) {
+	return m.openWriter(name, os.O_CREATE|os.O_TRUNC, args...)
+}
+
+// CreateFile creates a file on the cluster's config directory
+func (m *Zones) CreateFile(name string, args ...any) (io.WriteCloser, error) {
+	return m.openWriter(name, os.O_CREATE, args...)
+}
+
+func (m *Zones) openWriter(name string, flags int, args ...any) (io.WriteCloser, error) {
+	f, err := m.OpenFile(name, os.O_WRONLY|flags, args...)
+	if err != nil {
+		return nil, err
+	}
+
+	return f.(io.WriteCloser), nil
+}
+
+// ReadFile reads a file from the cluster's config directory
+func (m *Zones) ReadFile(name string, args ...any) ([]byte, error) {
+	if len(args) > 0 {
+		name = fmt.Sprintf(name, args...)
+	}
+
+	return fs.ReadFile(m.dir, name)
+}