zones: introduce (private) logger interface

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

cmd/jpictl/dump.go

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

cmd/jpictl/env.go

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

cmd/jpictl/gateway.go

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

cmd/jpictl/log.go

@@ -3,9 +3,15 @@ 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

@@ -2,8 +2,6 @@
 package main
 
 import (
-	"darvaza.org/sidecar/pkg/logger/zerolog"
-	"darvaza.org/slog"
 	"github.com/spf13/cobra"
 )
 
@@ -12,29 +10,13 @@ const (
 	CmdName = "jpictl"
 )
 
-var (
-	log       = zerolog.New(nil, slog.Error)
-	verbosity int
-	rootCmd   = &cobra.Command{
-		Use:   CmdName,
-		Short: "control tool for jpi.cloud",
-	}
-)
+var 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

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

pkg/zones/ceph_scan.go

@@ -1,12 +1,89 @@
 package zones
 
 import (
+	"net/netip"
 	"os"
+	"strings"
 
-	"darvaza.org/slog"
+	"darvaza.org/core"
 	"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
@@ -34,24 +111,26 @@ func (todo *cephScanTODO) checkMachine(p *Machine) bool {
 	return false
 }
 
-func (todo *cephScanTODO) LogMissing(log slog.Logger) {
+func (todo *cephScanTODO) Missing() error {
+	var check CephMissingMonitorError
+
 	for name, found := range todo.names {
 		if !found {
-			log.Warn().
-				WithField("subsystem", "ceph").
-				WithField("monitor", name).
-				Print("unknown monitor")
+			check.appendName(name)
 		}
 	}
 
 	for addr, found := range todo.addrs {
 		if !found {
-			log.Warn().
-				WithField("subsystem", "ceph").
-				WithField("monitor", addr).
-				Print("unknown monitor")
+			var a netip.Addr
+			// it wouldn't be on the map if it wasn't valid
+			_ = a.UnmarshalText([]byte(addr))
+
+			check.appendAddr(a)
 		}
 	}
+
+	return check.AsError()
 }
 
 func newCephScanTODO(cfg *ceph.Config) *cephScanTODO {
@@ -90,8 +169,9 @@ func (m *Zones) scanCephMonitors(_ *ScanOptions) error {
 			p.CephMonitor = todo.checkMachine(p)
 			return false
 		})
-
-		todo.LogMissing(m.log)
+		if err := todo.Missing(); err != nil {
+			return core.Wrap(err, "ceph")
+		}
 	}
 
 	// make sure every zone has one

pkg/zones/errors.go

@@ -1,16 +0,0 @@
-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

@@ -9,9 +9,7 @@ import (
 
 // A Machine is a machine on a Zone
 type Machine struct {
-	zone   *Zone
-	logger `toml:"-" json:"-" yaml:"-"`
-
+	zone *Zone
 	ID   int    `toml:"id"`
 	Name string `toml:"-" json:"-" yaml:"-"`
 

pkg/zones/machine_rings.go

@@ -2,7 +2,6 @@ package zones
 
 import (
 	"bytes"
-	"errors"
 	"fmt"
 	"os"
 
@@ -122,30 +121,23 @@ 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)
+		return fmt.Errorf("%s: invalid wg%v address: %s", m.Name, ring, addr)
 	}
 
 	if err := m.applyZoneNodeID(zoneID, nodeID); err != nil {
-		return core.Wrapf(err, "%s: invalid address", addr)
+		err = core.Wrapf(err, "%s: wg%v:%s", m.Name, ring, addr)
+		return err
 	}
 
 	if err := m.applyWireguardInterfaceConfig(ring, wg.Interface); err != nil {
-		return core.Wrap(err, "interface")
+		err = core.Wrapf(err, "%s: wg%v:%s", m.Name, ring, addr)
+		return err
 	}
 
 	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")
+		if err := m.applyWireguardPeerConfig(ring, peer); err != nil {
+			err = core.Wrapf(err, "%s: wg%v:%s", m.Name, ring, addr)
+			return err
 		}
 	}
 
@@ -166,10 +158,6 @@ 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
 	}
@@ -195,10 +183,8 @@ 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{
@@ -211,6 +197,8 @@ 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

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

@@ -22,7 +22,7 @@ type ScanOptions struct {
 
 	// Logger specifies the logger to be used. otherwise
 	// the scanner will be mute
-	slog.Logger
+	Logger slog.Logger
 }
 
 // ResolvePublicAddresses instructs the scanner to use

pkg/zones/scan.go

@@ -3,8 +3,6 @@ package zones
 import (
 	"io/fs"
 	"sort"
-
-	"darvaza.org/core"
 )
 
 func (m *Zones) scan(opts *ScanOptions) error {
@@ -33,40 +31,22 @@ func (m *Zones) scanDirectory(_ *ScanOptions) error {
 
 	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)
+			z := &Zone{
+				zones: m,
+				Name:  e.Name(),
 			}
+
+			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 {
@@ -152,22 +132,11 @@ func (z *Zone) scan() error {
 	for _, e := range entries {
 		if e.IsDir() {
 			m := &Machine{
-				zone:   z,
-				logger: z,
-				Name:   e.Name(),
+				zone: 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
 			}
 

pkg/zones/zones.go

@@ -87,8 +87,7 @@ func FilterMachines(m MachineIterator, cond func(*Machine) bool) (Machines, int)
 
 // Zone represents one zone in a cluster
 type Zone struct {
-	zones  *Zones
-	logger `toml:"-" json:"-" yaml:"-"`
+	zones *Zones
 
 	ID   int    `toml:"id"`
 	Name string `toml:"name"`