Merge pull request 'zones: warn but not fail when scanning finds unknown monitors' (#13)

Reviewed-on: #13

cmd/jpictl/dump.go

@@ -52,8 +52,9 @@ const encoding = YAMLEncoding
 
 // Command
 var dumpCmd = &cobra.Command{
-	Use:   "dump",
-	Short: "generates a text representation of the config",
+	Use:    "dump",
+	Short:  "generates a text representation of the config",
+	PreRun: setVerbosity,
 	RunE: func(_ *cobra.Command, _ []string) error {
 		var buf bytes.Buffer
 		var enc Encoder

cmd/jpictl/env.go

@@ -8,8 +8,9 @@ import (
 
 // Command
 var envCmd = &cobra.Command{
-	Use:   "env",
-	Short: "generates environment variables for shell scripts",
+	Use:    "env",
+	Short:  "generates environment variables for shell scripts",
+	PreRun: setVerbosity,
 	RunE: func(_ *cobra.Command, _ []string) error {
 		m, err := cfg.LoadZones(false)
 		if err != nil {

cmd/jpictl/gateway.go

@@ -19,8 +19,9 @@ var gatewayCmd = &cobra.Command{
 
 // gateway set
 var gatewaySetCmd = &cobra.Command{
-	Use:   "set",
-	Short: "gateway set sets machines as gateways",
+	Use:    "set",
+	Short:  "gateway set sets machines as gateways",
+	PreRun: setVerbosity,
 	RunE: func(_ *cobra.Command, args []string) error {
 		m, err := cfg.LoadZones(false)
 		if err != nil {

cmd/jpictl/log.go

@@ -3,15 +3,9 @@ package main
 import (
 	"fmt"
 
-	"darvaza.org/sidecar/pkg/logger/zerolog"
 	"darvaza.org/slog"
 )
 
-// TODO: make log level configurable via flags
-var (
-	log = zerolog.New(nil, slog.Debug)
-)
-
 // fatal is a convenience wrapper for slog.Logger.Fatal().Print()
 func fatal(err error, msg string, args ...any) {
 	l := log.Fatal()

cmd/jpictl/main.go

@@ -2,6 +2,8 @@
 package main
 
 import (
+	"darvaza.org/sidecar/pkg/logger/zerolog"
+	"darvaza.org/slog"
 	"github.com/spf13/cobra"
 )
 
@@ -10,13 +12,29 @@ const (
 	CmdName = "jpictl"
 )
 
-var rootCmd = &cobra.Command{
-	Use:   CmdName,
-	Short: "control tool for jpi.cloud",
-}
+var (
+	log       = zerolog.New(nil, slog.Error)
+	verbosity int
+	rootCmd   = &cobra.Command{
+		Use:   CmdName,
+		Short: "control tool for jpi.cloud",
+	}
+)
 
 func main() {
 	if err := rootCmd.Execute(); err != nil {
 		fatal(err, "")
 	}
 }
+
+func init() {
+	rootCmd.PersistentFlags().CountVarP(&verbosity, "verbosity", "v", "increase the verbosity level to Warn, Info or Debug")
+}
+
+func setVerbosity(_ *cobra.Command, _ []string) {
+	desired := int8(slog.Error) + int8(verbosity)
+	if desired > 6 {
+		desired = 6
+	}
+	log = log.WithLevel(slog.LogLevel(desired))
+}

cmd/jpictl/write.go

@@ -6,8 +6,9 @@ import (
 
 // Command
 var writeCmd = &cobra.Command{
-	Use:   "write",
-	Short: "rewrites all config files",
+	Use:    "write",
+	Short:  "rewrites all config files",
+	PreRun: setVerbosity,
 	RunE: func(_ *cobra.Command, _ []string) error {
 		m, err := cfg.LoadZones(false)
 		if err != nil {

pkg/zones/ceph_scan.go

@@ -1,89 +1,12 @@
 package zones
 
 import (
-	"net/netip"
 	"os"
-	"strings"
 
-	"darvaza.org/core"
+	"darvaza.org/slog"
 	"git.jpi.io/amery/jpictl/pkg/ceph"
 )
 
-// CephMissingMonitorError is an error that contains ceph
-// monitors present in ceph.conf but not found on the cluster
-type CephMissingMonitorError struct {
-	Names []string
-	Addrs []netip.Addr
-}
-
-func (err *CephMissingMonitorError) appendName(name string) {
-	err.Names = append(err.Names, name)
-}
-
-func (err *CephMissingMonitorError) appendAddr(addr netip.Addr) {
-	err.Addrs = append(err.Addrs, addr)
-}
-
-// OK tells if this instance actual shouldn't be treated as an error
-func (err CephMissingMonitorError) OK() bool {
-	switch {
-	case len(err.Names) > 0:
-		return false
-	case len(err.Addrs) > 0:
-		return false
-	default:
-		return true
-	}
-}
-
-func (err CephMissingMonitorError) Error() string {
-	if !err.OK() {
-		var buf strings.Builder
-
-		_, _ = buf.WriteString("missing:")
-		err.writeNames(&buf)
-		err.writeAddrs(&buf)
-
-		return buf.String()
-	}
-
-	// no error
-	return ""
-}
-
-func (err *CephMissingMonitorError) writeNames(w *strings.Builder) {
-	if len(err.Names) > 0 {
-		_, _ = w.WriteString(" mon_initial_members:")
-		for i, name := range err.Names {
-			if i != 0 {
-				_, _ = w.WriteRune(',')
-			}
-			_, _ = w.WriteString(name)
-		}
-	}
-}
-
-func (err *CephMissingMonitorError) writeAddrs(w *strings.Builder) {
-	if len(err.Addrs) > 0 {
-		_, _ = w.WriteString(" mon_host:")
-		for i, addr := range err.Addrs {
-			if i != 0 {
-				_, _ = w.WriteRune(',')
-			}
-			_, _ = w.WriteString(addr.String())
-		}
-	}
-}
-
-// AsError returns nil if the instance is actually OK
-func (err *CephMissingMonitorError) AsError() error {
-	if err == nil || err.OK() {
-		return nil
-	}
-
-	return err
-}
-
 type cephScanTODO struct {
 	names map[string]bool
 	addrs map[string]bool
@@ -111,26 +34,24 @@ func (todo *cephScanTODO) checkMachine(p *Machine) bool {
 	return false
 }
 
-func (todo *cephScanTODO) Missing() error {
-	var check CephMissingMonitorError
-
+func (todo *cephScanTODO) LogMissing(log slog.Logger) {
 	for name, found := range todo.names {
 		if !found {
-			check.appendName(name)
+			log.Warn().
+				WithField("subsystem", "ceph").
+				WithField("monitor", name).
+				Print("unknown monitor")
 		}
 	}
 
 	for addr, found := range todo.addrs {
 		if !found {
-			var a netip.Addr
-			// it wouldn't be on the map if it wasn't valid
-			_ = a.UnmarshalText([]byte(addr))
-
-			check.appendAddr(a)
+			log.Warn().
+				WithField("subsystem", "ceph").
+				WithField("monitor", addr).
+				Print("unknown monitor")
 		}
 	}
-
-	return check.AsError()
 }
 
 func newCephScanTODO(cfg *ceph.Config) *cephScanTODO {
@@ -169,9 +90,8 @@ func (m *Zones) scanCephMonitors(_ *ScanOptions) error {
 			p.CephMonitor = todo.checkMachine(p)
 			return false
 		})
-		if err := todo.Missing(); err != nil {
-			return core.Wrap(err, "ceph")
-		}
+
+		todo.LogMissing(m.log)
 	}
 
 	// make sure every zone has one

pkg/zones/errors.go

@@ -0,0 +1,16 @@
+package zones
+
+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/zones/machine.go

@@ -9,8 +9,8 @@ import (
 
 // A Machine is a machine on a Zone
 type Machine struct {
-	zone *Zone
-	logger
+	zone   *Zone
+	logger `toml:"-" json:"-" yaml:"-"`
 
 	ID   int    `toml:"id"`
 	Name string `toml:"-" json:"-" yaml:"-"`

pkg/zones/machine_rings.go

@@ -2,6 +2,7 @@ package zones
 
 import (
 	"bytes"
+	"errors"
 	"fmt"
 	"os"
 
@@ -121,23 +122,30 @@ 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 wg%v address: %s", m.Name, ring, addr)
+		return fmt.Errorf("%s: invalid address", addr)
 	}
 
 	if err := m.applyZoneNodeID(zoneID, nodeID); err != nil {
-		err = core.Wrapf(err, "%s: wg%v:%s", m.Name, ring, addr)
-		return err
+		return core.Wrapf(err, "%s: invalid address", addr)
 	}
 
 	if err := m.applyWireguardInterfaceConfig(ring, wg.Interface); err != nil {
-		err = core.Wrapf(err, "%s: wg%v:%s", m.Name, ring, addr)
-		return err
+		return core.Wrap(err, "interface")
 	}
 
 	for _, peer := range wg.Peer {
-		if err := m.applyWireguardPeerConfig(ring, peer); err != nil {
-			err = core.Wrapf(err, "%s: wg%v:%s", m.Name, ring, addr)
-			return err
+		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")
 		}
 	}
 
@@ -158,6 +166,10 @@ 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
 	}
@@ -183,8 +195,10 @@ func (m *Machine) applyWireguardPeerConfig(ring int, pc wireguard.PeerConfig) er
 	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{
@@ -197,8 +211,6 @@ func (m *Machine) applyWireguardPeerConfig(ring int, pc wireguard.PeerConfig) er
 
 		return peer.applyRingInfo(ring, ri)
 	}
-
-	return fmt.Errorf("%q: invalid peer endpoint", pc.Endpoint.Host)
 }
 
 func (m *Machine) applyZoneNodeID(zoneID, nodeID int) error {

pkg/zones/machine_scan.go

@@ -4,7 +4,10 @@ import (
 	"context"
 	"net/netip"
 	"strconv"
+	"strings"
 	"time"
+
+	"darvaza.org/core"
 )
 
 // LookupNetIP uses the DNS Resolver to get the public addresses associated
@@ -30,21 +33,32 @@ func (m *Machine) UpdatePublicAddresses() error {
 
 func (m *Machine) init() error {
 	if err := m.setID(); err != nil {
-		return err
+		return core.Wrap(err, m.Name)
 	}
 
 	for i := 0; i < RingsCount; i++ {
 		if err := m.tryReadWireguardKeys(i); err != nil {
-			return err
+			return core.Wrap(err, m.Name)
 		}
 	}
+
 	return nil
 }
 
 func (m *Machine) setID() error {
 	zoneName := m.zone.Name
-	suffix := m.Name[len(zoneName)+1:]
 
+	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
@@ -57,6 +71,11 @@ func (m *Machine) setID() error {
 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
 		}
 	}

pkg/zones/scan.go

@@ -3,6 +3,8 @@ package zones
 import (
 	"io/fs"
 	"sort"
+
+	"darvaza.org/core"
 )
 
 func (m *Zones) scan(opts *ScanOptions) error {
@@ -31,23 +33,40 @@ func (m *Zones) scanDirectory(_ *ScanOptions) error {
 
 	for _, e := range entries {
 		if e.IsDir() {
-			z := &Zone{
-				zones:  m,
-				logger: m,
-				Name:   e.Name(),
+			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)
 			}
-
-			if err := z.scan(); err != nil {
-				return err
-			}
-
-			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 {
@@ -138,7 +157,17 @@ func (z *Zone) scan() error {
 				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
 			}
 

pkg/zones/zones.go

@@ -87,8 +87,8 @@ func FilterMachines(m MachineIterator, cond func(*Machine) bool) (Machines, int)
 
 // Zone represents one zone in a cluster
 type Zone struct {
-	zones *Zones
-	logger
+	zones  *Zones
+	logger `toml:"-" json:"-" yaml:"-"`
 
 	ID   int    `toml:"id"`
 	Name string `toml:"name"`