Merge pull request 'move pkg/zones to pkg/cluster and other rearrangements' (#18)

Reviewed-on: #18

.vscode/settings.json

@@ -0,0 +1,9 @@
+{
+    "cSpell.words": [
+        "ceph",
+        "darvaza",
+        "gofrs",
+        "jpictl",
+        "zerolog"
+    ]
+}

cmd/jpictl/config.go

@@ -1,6 +1,6 @@
 package main
 
-import "git.jpi.io/amery/jpictl/pkg/zones"
+import "git.jpi.io/amery/jpictl/pkg/cluster"
 
 // Config describes the repository
 type Config struct {
@@ -9,13 +9,14 @@ type Config struct {
 }
 
 var cfg = &Config{
-	Base:   "./m",
-	Domain: "m.jpi.cloud",
+	Base:   "m",
+	Domain: "jpi.cloud",
 }
 
 // LoadZones loads all zones and machines in the config directory
-func (cfg *Config) LoadZones(resolve bool) (*zones.Zones, error) {
-	return zones.New(cfg.Base, cfg.Domain,
-		zones.ResolvePublicAddresses(resolve),
+func (cfg *Config) LoadZones(resolve bool) (*cluster.Cluster, error) {
+	return cluster.NewFromDirectory(cfg.Base, cfg.Domain,
+		cluster.ResolvePublicAddresses(resolve),
+		cluster.WithLogger(log),
 	)
 }

cmd/jpictl/dump.go

@@ -6,7 +6,6 @@ import (
 	"io"
 	"os"
 
-	"github.com/burntSushi/toml"
 	"github.com/spf13/cobra"
 	"gopkg.in/yaml.v3"
 )
@@ -20,8 +19,8 @@ type Encoder interface {
 type Encoding int
 
 const (
-	// TOMLEncoding represents TOML encoding
-	TOMLEncoding Encoding = iota
+	// UndefinedEncoding implies the default encoding
+	UndefinedEncoding Encoding = iota
 	// JSONEncoding represents JSON encoding
 	JSONEncoding
 	// YAMLEncoding represents YAML encoding
@@ -42,18 +41,13 @@ func NewYAMLEncoder(w io.Writer) Encoder {
 	return enc
 }
 
-// NewTOMLEncoder returns a TOML [Encoder] to work on the given [io.Writer]
-func NewTOMLEncoder(w io.Writer) Encoder {
-	enc := toml.NewEncoder(w)
-	return enc
-}
-
 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
@@ -66,10 +60,8 @@ var dumpCmd = &cobra.Command{
 		switch encoding {
 		case JSONEncoding:
 			enc = NewJSONEncoder(&buf)
-		case YAMLEncoding:
-			enc = NewYAMLEncoder(&buf)
 		default:
-			enc = NewTOMLEncoder(&buf)
+			enc = NewYAMLEncoder(&buf)
 		}
 
 		if err = enc.Encode(m); err != nil {

cmd/jpictl/env.go

@@ -8,15 +8,20 @@ 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 {
 			return err
 		}
 
-		_, err = m.Env(*envExport).WriteTo(os.Stdout)
+		env, err := m.Env(*envExport)
+		if err != nil {
+			return err
+		}
+		_, err = env.WriteTo(os.Stdout)
 		return err
 	},
 }

cmd/jpictl/gateway.go

@@ -7,8 +7,9 @@ import (
 	"strconv"
 	"strings"
 
-	"git.jpi.io/amery/jpictl/pkg/zones"
 	"github.com/spf13/cobra"
+
+	"git.jpi.io/amery/jpictl/pkg/cluster"
 )
 
 // Command
@@ -19,8 +20,9 @@ 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 {
@@ -37,9 +39,9 @@ var gatewaySetCmd = &cobra.Command{
 	},
 }
 
-func gatewaySet(zi zones.ZoneIterator, gw string) error {
+func gatewaySet(zi cluster.ZoneIterator, gw string) error {
 	var err error
-	zi.ForEachZone(func(z *zones.Zone) bool {
+	zi.ForEachZone(func(z *cluster.Zone) bool {
 		for _, m := range z.Machines {
 			if m.Name == gw {
 				z.SetGateway(m.ID, true)
@@ -73,9 +75,9 @@ var gatewayUnsetCmd = &cobra.Command{
 	},
 }
 
-func gatewayUnset(zi zones.ZoneIterator, ngw string) error {
+func gatewayUnset(zi cluster.ZoneIterator, ngw string) error {
 	var err error
-	zi.ForEachZone(func(z *zones.Zone) bool {
+	zi.ForEachZone(func(z *cluster.Zone) bool {
 		for _, m := range z.Machines {
 			if m.Name == ngw && m.IsGateway() {
 				z.SetGateway(m.ID, false)
@@ -114,10 +116,10 @@ var gatewayListCmd = &cobra.Command{
 	},
 }
 
-func gatewayListAll(zi zones.ZoneIterator) error {
+func gatewayListAll(zi cluster.ZoneIterator) error {
 	var b bytes.Buffer
 	var err error
-	zi.ForEachZone(func(z *zones.Zone) bool {
+	zi.ForEachZone(func(z *cluster.Zone) bool {
 		b.WriteString(z.Name + ":")
 		var sIDs []string
 		ids, num := z.GatewayIDs()
@@ -136,10 +138,10 @@ func gatewayListAll(zi zones.ZoneIterator) error {
 	return err
 }
 
-func gatewayList(zi zones.ZoneIterator, m string) error {
+func gatewayList(zi cluster.ZoneIterator, m string) error {
 	var b bytes.Buffer
 	var err error
-	zi.ForEachZone(func(z *zones.Zone) bool {
+	zi.ForEachZone(func(z *cluster.Zone) bool {
 		if z.Name == m {
 			b.WriteString(z.Name + ":")
 			ids, num := z.GatewayIDs()

cmd/jpictl/log.go

@@ -3,14 +3,9 @@ package main
 import (
 	"fmt"
 
-	"darvaza.org/sidecar/pkg/logger/zerolog"
 	"darvaza.org/slog"
 )
 
-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 +12,30 @@ 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 {

go.mod

@@ -3,11 +3,13 @@ module git.jpi.io/amery/jpictl
 go 1.19
 
 require (
-	darvaza.org/core v0.9.5
-	darvaza.org/resolver v0.5.2
-	darvaza.org/sidecar v0.0.0-20230721122716-b9c54b8adbaf
-	darvaza.org/slog v0.5.2
-	github.com/burntSushi/toml v0.3.1
+	asciigoat.org/ini v0.2.5
+	darvaza.org/core v0.9.8
+	darvaza.org/resolver v0.5.4
+	darvaza.org/sidecar v0.0.2
+	darvaza.org/slog v0.5.3
+	darvaza.org/slog/handlers/discard v0.4.5
+	github.com/gofrs/uuid/v5 v5.0.0
 	github.com/hack-pad/hackpadfs v0.2.1
 	github.com/mgechev/revive v1.3.3
 	github.com/spf13/cobra v1.7.0
@@ -17,8 +19,9 @@ require (
 )
 
 require (
-	darvaza.org/slog/handlers/filter v0.4.4 // indirect
-	darvaza.org/slog/handlers/zerolog v0.4.4 // indirect
+	asciigoat.org/core v0.3.9 // indirect
+	darvaza.org/slog/handlers/filter v0.4.5 // indirect
+	darvaza.org/slog/handlers/zerolog v0.4.5 // indirect
 	github.com/BurntSushi/toml v1.3.2 // indirect
 	github.com/chavacava/garif v0.1.0 // indirect
 	github.com/fatih/color v1.15.0 // indirect
@@ -37,7 +40,7 @@ require (
 	github.com/spf13/pflag v1.0.5 // indirect
 	golang.org/x/mod v0.12.0 // indirect
 	golang.org/x/net v0.14.0 // indirect
-	golang.org/x/sys v0.11.0 // indirect
+	golang.org/x/sys v0.12.0 // indirect
 	golang.org/x/text v0.13.0 // indirect
 	golang.org/x/tools v0.12.0 // indirect
 	gopkg.in/warnings.v0 v0.1.2 // indirect

go.sum

@@ -1,19 +1,23 @@
-darvaza.org/core v0.9.5 h1:sS5pZFwicaxJIQixEiqkMr9GknVHYL+EbKDMkR/4jDM=
-darvaza.org/core v0.9.5/go.mod h1:O3tHBMlw+xB47uGh5CUx7dXAujBAMmD8BCRFPZmIw54=
-darvaza.org/resolver v0.5.2 h1:VjHhEr/MJBszeDb7tYlXQ9Bsyh4xrDR7Sd10WAmPD6k=
-darvaza.org/resolver v0.5.2/go.mod h1:fFvsVPEFeMzUIWlLG47Go/6uJYtRLb9R8HIgYg3uaxE=
-darvaza.org/sidecar v0.0.0-20230721122716-b9c54b8adbaf h1:ya5ZQicBb/GWll3rlqra8No7oJXks7y1m/cJGYBypv4=
-darvaza.org/sidecar v0.0.0-20230721122716-b9c54b8adbaf/go.mod h1:by+bPsMa7Rxc/ZYG1qBunrtKocv/DkrPBmyFlmq/j2Q=
-darvaza.org/slog v0.5.2 h1:8TG1WyHjOyh2vW6t3pjzZVaWzpko5MIIpeI7LWqHFvs=
-darvaza.org/slog v0.5.2/go.mod h1:HAkEpxTA/mkiLNUXJo5qsCh8EVCtA3evje8GAaCDWHI=
-darvaza.org/slog/handlers/filter v0.4.4 h1:b2e2T9fQzMdJ0ia+f6b7kw9/T9GFwhFCKob/2tqhGGU=
-darvaza.org/slog/handlers/filter v0.4.4/go.mod h1:cQlJWuolB6guLug09sX/8Zrzct++M6SPCGvXR37E7Cc=
-darvaza.org/slog/handlers/zerolog v0.4.4 h1:OR1ASvH1fBCq3t85t4OU6oJPPuqMB1tsDoSpsh6HVJU=
-darvaza.org/slog/handlers/zerolog v0.4.4/go.mod h1:t60TeEbFcMLo74CkXC2S0rKlnwF4ixZyBR4fqIJV1GE=
+asciigoat.org/core v0.3.9 h1:hgDDz4ecm3ZvehX++m8A/IzAt+B5oDPiRtxatzfUHPQ=
+asciigoat.org/core v0.3.9/go.mod h1:CAaHwyw8MpAq4a1MYtN2dxJrsK+hmIdW50OndaQZYPI=
+asciigoat.org/ini v0.2.5 h1:4gRIp9rU+XQt8+HMqZO5R7GavMv9Yl2+N+je6djDIAE=
+asciigoat.org/ini v0.2.5/go.mod h1:gmXzJ9XFqf1NLk5nQkj04USQ4tMtdRJHNQX6vp3DzjU=
+darvaza.org/core v0.9.8 h1:luLxgfUc2pzuusYPo/Z/dC/qr9XZPKpSQw8/kS7zNUM=
+darvaza.org/core v0.9.8/go.mod h1:Dbme64naxeshQfxcVJX9ZT7AiGyIY8kldfuELVtf8mw=
+darvaza.org/resolver v0.5.4 h1:dlSBNV14yYsp7Kg7ipwYOMNsLbrpeXa8Z0HBTa0Ryxs=
+darvaza.org/resolver v0.5.4/go.mod h1:vHMkQUmHjaetFqG2ZLZJiQHsXEMGoTOFGm+NXwfndhE=
+darvaza.org/sidecar v0.0.2 h1:4H8FUxc43kkLjxdShN1CoxLTcoHQsZjDVwm7kt6eIK0=
+darvaza.org/sidecar v0.0.2/go.mod h1:yFC3Qt3j+uS7n9CMpLxwrA68z+FNJhENoenBc9zBJJo=
+darvaza.org/slog v0.5.3 h1:sQzmZXgqRh9oFMKBwEYrEpucLvKJVZxaxa2bHIA6GJ0=
+darvaza.org/slog v0.5.3/go.mod h1:59d+yi+C7gn4pDDuwbbOKawERpdXthFFk1Yc+Sv6XB0=
+darvaza.org/slog/handlers/discard v0.4.5 h1:RRykOItNolHyiUav57lG/GFBL33rcljoa0nWTpY+T0g=
+darvaza.org/slog/handlers/discard v0.4.5/go.mod h1:HYHfISQjMqcPbPoPZ92ib/u7s9JcXvF6OaygpPFwdF8=
+darvaza.org/slog/handlers/filter v0.4.5 h1:CX1bMzldd67e3y3s3Sh4jK8Lyo0WMvTGBB2lD315jhc=
+darvaza.org/slog/handlers/filter v0.4.5/go.mod h1:OuH9rHYg9CIErTJCZliMnFexBfP/HJ9PZ1V1VwSCZ1g=
+darvaza.org/slog/handlers/zerolog v0.4.5 h1:W4cgGORx4wImr+RL96CWSQGTdkZzKX6YHXPSYJvdoB4=
+darvaza.org/slog/handlers/zerolog v0.4.5/go.mod h1:mCoh/mIl8Nsa6Yu1Um7d7cos6RuEJzgaTXaX5LDRUao=
 github.com/BurntSushi/toml v1.3.2 h1:o7IhLm0Msx3BaB+n3Ag7L8EVlByGnpq14C4YWiu/gL8=
 github.com/BurntSushi/toml v1.3.2/go.mod h1:CxXYINrC8qIiEnFrOxCa7Jy5BFHlXnUU2pbicEuybxQ=
-github.com/burntSushi/toml v0.3.1 h1:Hu1cOEC2qtKULZJCzym5tyA35bZr3HREuolgiAzMlhY=
-github.com/burntSushi/toml v0.3.1/go.mod h1:sGTquCpRYr9McuHdv0m6YKIhx8DJGJa4t04/Y9pfSio=
 github.com/chavacava/garif v0.1.0 h1:2JHa3hbYf5D9dsgseMKAmc/MZ109otzgNFk5s87H9Pc=
 github.com/chavacava/garif v0.1.0/go.mod h1:XMyYCkEL58DF0oyW4qDjjnPWONs2HBqYKI+UIPD+Gww=
 github.com/coreos/go-systemd/v22 v22.5.0/go.mod h1:Y58oyj3AT4RCenI/lSvhwexgC+NSVTIJ3seZv2GcEnc=
@@ -26,6 +30,8 @@ github.com/fatih/color v1.15.0/go.mod h1:0h5ZqXfHYED7Bhv2ZJamyIOUej9KtShiJESRwBD
 github.com/fatih/structtag v1.2.0 h1:/OdNE99OxoI/PqaW/SuSK9uxxT3f/tcSZgon/ssNSx4=
 github.com/fatih/structtag v1.2.0/go.mod h1:mBJUNpUnHmRKrKlQQlmCrh5PuhftFbNv8Ys4/aAZl94=
 github.com/godbus/dbus/v5 v5.0.4/go.mod h1:xhWf0FNVPg57R7Z0UbKHbJfkEywrmjJnf7w5xrFpKfA=
+github.com/gofrs/uuid/v5 v5.0.0 h1:p544++a97kEL+svbcFbCQVM9KFu0Yo25UoISXGNNH9M=
+github.com/gofrs/uuid/v5 v5.0.0/go.mod h1:CDOjlDMVAtN56jqyRUZh58JT31Tiw7/oQyEXZV+9bD8=
 github.com/hack-pad/hackpadfs v0.2.1 h1:FelFhIhv26gyjujoA/yeFO+6YGlqzmc9la/6iKMIxMw=
 github.com/hack-pad/hackpadfs v0.2.1/go.mod h1:khQBuCEwGXWakkmq8ZiFUvUZz84ZkJ2KNwKvChs4OrU=
 github.com/inconshreveable/mousetrap v1.1.0 h1:wN+x4NVGpMsO7ErUn/mUI3vEoE6Jt13X2s0bqwp9tc8=
@@ -84,8 +90,8 @@ golang.org/x/sys v0.0.0-20210630005230-0f9fa26af87c/go.mod h1:oPkhp1MJrh7nUepCBc
 golang.org/x/sys v0.0.0-20210927094055-39ccf1dd6fa6/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.0.0-20220811171246-fbc7d0a398ab/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.11.0 h1:eG7RXZHdqOJ1i+0lgLgCpSXAp6M3LYlAo6osgSi0xOM=
-golang.org/x/sys v0.11.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.12.0 h1:CM0HF96J0hcLAwsHPJZjfdNzs0gftsLfgKt57wWHJ0o=
+golang.org/x/sys v0.12.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/text v0.13.0 h1:ablQoSUd0tRdKxZewP80B+BaqeKJuVhuRxj/dkrun3k=
 golang.org/x/text v0.13.0/go.mod h1:TvPlkZtksWOMsz7fbANvkp4WM8x/WCo/om8BMLbz+aE=
 golang.org/x/tools v0.12.0 h1:YW6HUoUmYBpwSgyaGaZq1fHjrBjX1rlpZ54T6mu2kss=

pkg/ceph/ceph.go

@@ -0,0 +1,2 @@
+// Package ceph deals with ceph config
+package ceph

pkg/ceph/config.go

@@ -0,0 +1,67 @@
+package ceph
+
+import (
+	"bytes"
+	"fmt"
+	"io"
+	"net/netip"
+	"strings"
+
+	"github.com/gofrs/uuid/v5"
+
+	"asciigoat.org/ini/basic"
+)
+
+// Config represents a ceph.conf file
+type Config struct {
+	Global GlobalConfig `ini:"global"`
+}
+
+// GlobalConfig represents the [global] section of a ceph.conf file
+type GlobalConfig struct {
+	FSID           uuid.UUID    `ini:"fsid"`
+	Monitors       []string     `ini:"mon_initial_members,comma"`
+	MonitorsAddr   []netip.Addr `ini:"mon_host,comma"`
+	ClusterNetwork netip.Prefix `ini:"cluster_network"`
+}
+
+// WriteTo writes a Wireguard [Config] onto the provided [io.Writer]
+func (cfg *Config) WriteTo(w io.Writer) (int64, error) {
+	var buf bytes.Buffer
+
+	writeGlobalToBuffer(&buf, &cfg.Global)
+	return buf.WriteTo(w)
+}
+
+func writeGlobalToBuffer(w *bytes.Buffer, c *GlobalConfig) {
+	_, _ = w.WriteString("[global]\n")
+	_, _ = fmt.Fprintf(w, "%s = %s\n", "fsid", c.FSID.String())
+	_, _ = fmt.Fprintf(w, "%s = %s\n", "mon_initial_members", strings.Join(c.Monitors, ", "))
+	_, _ = fmt.Fprintf(w, "%s = %s\n", "mon_host", joinAddrs(c.MonitorsAddr, ", "))
+	_, _ = fmt.Fprintf(w, "%s = %s\n", "cluster_network", c.ClusterNetwork.String())
+}
+
+func joinAddrs(addrs []netip.Addr, sep string) string {
+	s := make([]string, len(addrs))
+
+	for i, addr := range addrs {
+		s[i] = addr.String()
+	}
+
+	return strings.Join(s, sep)
+}
+
+// NewConfigFromReader parses the ceph.conf file
+func NewConfigFromReader(r io.Reader) (*Config, error) {
+	doc, err := basic.Decode(r)
+	if err != nil {
+		return nil, err
+	}
+
+	cfg, err := newConfigFromDocument(doc)
+	if err != nil {
+		return nil, err
+	}
+
+	return cfg, nil
+}

pkg/ceph/config_parser.go

@@ -0,0 +1,110 @@
+package ceph
+
+import (
+	"io/fs"
+	"net/netip"
+
+	"asciigoat.org/ini/basic"
+	"asciigoat.org/ini/parser"
+
+	"darvaza.org/core"
+)
+
+var sectionMap = map[string]func(*Config, *basic.Section) error{
+	"global": loadGlobalConfSection,
+}
+
+func loadConfSection(out *Config, src *basic.Section) error {
+	h, ok := sectionMap[src.Key]
+	if !ok {
+		return core.Wrapf(fs.ErrInvalid, "unknown section %q", src.Key)
+	}
+
+	return h(out, src)
+}
+
+func loadGlobalConfSection(out *Config, src *basic.Section) error {
+	var cfg GlobalConfig
+
+	for _, field := range src.Fields {
+		if err := loadGlobalConfField(&cfg, field); err != nil {
+			return core.Wrap(err, "global")
+		}
+	}
+
+	out.Global = cfg
+	return nil
+}
+
+// revive:disable:cyclomatic
+// revive:disable:cognitive-complexity
+
+func loadGlobalConfField(cfg *GlobalConfig, field basic.Field) error {
+	// revive:enable:cyclomatic
+	// revive:enable:cognitive-complexity
+
+	// TODO: refactor when asciigoat's ini parser learns to do reflection
+
+	switch field.Key {
+	case "fsid":
+		if !core.IsZero(cfg.FSID) {
+			return core.Wrapf(fs.ErrInvalid, "duplicate field %q", field.Key)
+		}
+
+		err := cfg.FSID.UnmarshalText([]byte(field.Value))
+		switch {
+		case err != nil:
+			return core.Wrap(err, field.Key)
+		default:
+			return nil
+		}
+	case "mon_host":
+		entries, _ := parser.SplitCommaArray(field.Value)
+		for _, s := range entries {
+			var addr netip.Addr
+
+			if err := addr.UnmarshalText([]byte(s)); err != nil {
+				return core.Wrap(err, field.Key)
+			}
+
+			cfg.MonitorsAddr = append(cfg.MonitorsAddr, addr)
+		}
+		return nil
+	case "mon_initial_members":
+		entries, _ := parser.SplitCommaArray(field.Value)
+		cfg.Monitors = append(cfg.Monitors, entries...)
+		return nil
+	case "cluster_network":
+		if !core.IsZero(cfg.ClusterNetwork) {
+			err := core.Wrap(fs.ErrInvalid, "fields before the first section")
+			return err
+		}
+
+		err := cfg.ClusterNetwork.UnmarshalText([]byte(field.Value))
+		switch {
+		case err != nil:
+			return core.Wrap(err, field.Key)
+		default:
+			return nil
+		}
+	}
+	return nil
+}
+
+func newConfigFromDocument(doc *basic.Document) (*Config, error) {
+	var out Config
+
+	if len(doc.Global) > 0 {
+		err := core.Wrap(fs.ErrInvalid, "fields before the first section")
+		return nil, err
+	}
+
+	for i := range doc.Sections {
+		src := &doc.Sections[i]
+		if err := loadConfSection(&out, src); err != nil {
+			return nil, err
+		}
+	}
+
+	return &out, nil
+}

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 *Cluster) 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 *Cluster) 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 *Cluster) 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 *Cluster) 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 *Cluster) 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 +1,77 @@
+// Package cluster contains information about the cluster
+package cluster
+
+import (
+	"io/fs"
+
+	"darvaza.org/resolver"
+	"darvaza.org/slog"
+	"github.com/gofrs/uuid/v5"
+)
+
+var (
+	_ MachineIterator = (*Cluster)(nil)
+	_ ZoneIterator    = (*Cluster)(nil)
+)
+
+// revive:disable:line-length-limit
+
+// Cluster represents all zones in a cluster
+type Cluster struct {
+	dir      fs.FS
+	log      slog.Logger
+	resolver resolver.Resolver
+
+	BaseDir string `json:"dir,omitempty"    yaml:"dir,omitempty"`
+	Name    string `json:"name,omitempty"   yaml:"name,omitempty"`
+	Domain  string `json:"domain,omitempty" yaml:"domain,omitempty"`
+
+	CephFSID uuid.UUID `json:"ceph_fsid,omitempty" yaml:"ceph_fsid,omitempty"`
+	Zones    []*Zone   `json:"zones,omitempty"     yaml:"zones,omitempty"`
+}
+
+// revive:enable:line-length-limit
+
+// ForEachMachine calls a function for each Machine in the cluster
+// until instructed to terminate the loop
+func (m *Cluster) 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 *Cluster) 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 *Cluster) 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/cluster_file.go

@@ -1,4 +1,4 @@
-package zones
+package cluster
 
 import (
 	"fmt"
@@ -9,7 +9,7 @@ import (
 )
 
 // 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) {
+func (m *Cluster) OpenFile(name string, flags int, args ...any) (fs.File, error) {
 	if len(args) > 0 {
 		name = fmt.Sprintf(name, args...)
 	}
@@ -18,16 +18,16 @@ func (m *Zones) OpenFile(name string, flags int, args ...any) (fs.File, error) {
 }
 
 // CreateTruncFile creates or truncates a file on the cluster's config directory
-func (m *Zones) CreateTruncFile(name string, args ...any) (io.WriteCloser, error) {
+func (m *Cluster) 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) {
+func (m *Cluster) 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) {
+func (m *Cluster) 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
@@ -37,7 +37,7 @@ func (m *Zones) openWriter(name string, flags int, args ...any) (io.WriteCloser,
 }
 
 // ReadFile reads a file from the cluster's config directory
-func (m *Zones) ReadFile(name string, args ...any) ([]byte, error) {
+func (m *Cluster) ReadFile(name string, args ...any) ([]byte, error) {
 	if len(args) > 0 {
 		name = fmt.Sprintf(name, args...)
 	}

pkg/cluster/cluster_scan.go

@@ -1,17 +1,20 @@
-package zones
+package cluster
 
 import (
 	"io/fs"
 	"sort"
+
+	"darvaza.org/core"
 )
 
-func (m *Zones) scan(opts *ScanOptions) error {
+func (m *Cluster) 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
@@ -21,7 +24,7 @@ func (m *Zones) scan(opts *ScanOptions) error {
 	return nil
 }
 
-func (m *Zones) scanDirectory(_ *ScanOptions) error {
+func (m *Cluster) scanDirectory(_ *ScanOptions) error {
 	// each directory is a zone
 	entries, err := fs.ReadDir(m.dir, ".")
 	if err != nil {
@@ -30,23 +33,41 @@ func (m *Zones) scanDirectory(_ *ScanOptions) error {
 
 	for _, e := range entries {
 		if e.IsDir() {
-			z := &Zone{
-				zones: 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) scanMachines(opts *ScanOptions) error {
+func (m *Cluster) 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 *Cluster) scanMachines(opts *ScanOptions) error {
 	var err error
 	m.ForEachMachine(func(p *Machine) bool {
 		err = p.scan(opts)
@@ -55,7 +76,7 @@ func (m *Zones) scanMachines(opts *ScanOptions) error {
 	return err
 }
 
-func (m *Zones) scanZoneIDs(_ *ScanOptions) error {
+func (m *Cluster) scanZoneIDs(_ *ScanOptions) error {
 	var hasMissing bool
 	var lastZoneID int
 
@@ -85,7 +106,7 @@ func (m *Zones) scanZoneIDs(_ *ScanOptions) error {
 	return nil
 }
 
-func (m *Zones) scanSort(_ *ScanOptions) error {
+func (m *Cluster) scanSort(_ *ScanOptions) error {
 	sort.SliceStable(m.Zones, func(i, j int) bool {
 		id1 := m.Zones[i].ID
 		id2 := m.Zones[j].ID
@@ -111,7 +132,7 @@ func (m *Zones) scanSort(_ *ScanOptions) error {
 	return nil
 }
 
-func (m *Zones) scanGateways(_ *ScanOptions) error {
+func (m *Cluster) scanGateways(_ *ScanOptions) error {
 	var err error
 
 	m.ForEachZone(func(z *Zone) bool {
@@ -131,11 +152,22 @@ func (z *Zone) scan() error {
 	for _, e := range entries {
 		if e.IsDir() {
 			m := &Machine{
-				zone: z,
-				Name: e.Name(),
+				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
 			}
 

pkg/cluster/cluster_scan_options.go

@@ -1,15 +1,16 @@
-package zones
+package cluster
 
 import (
 	"io/fs"
 	"path/filepath"
 
 	"darvaza.org/resolver"
+	"darvaza.org/slog"
 	"github.com/hack-pad/hackpadfs/os"
 )
 
-// A ScanOption preconfigures the Zones before scanning
-type ScanOption func(*Zones, *ScanOptions) error
+// A ScanOption pre-configures the Zones before scanning
+type ScanOption func(*Cluster, *ScanOptions) error
 
 // ScanOptions contains flags used by the initial scan
 type ScanOptions struct {
@@ -17,13 +18,17 @@ type ScanOptions struct {
 	// 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 {
+	return func(m *Cluster, opt *ScanOptions) error {
 		opt.DontResolvePublicAddresses = !resolve
 		return nil
 	}
@@ -32,7 +37,7 @@ func ResolvePublicAddresses(resolve bool) ScanOption {
 // WithLookuper specifies what resolver.Lookuper to use to
 // find public addresses
 func WithLookuper(h resolver.Lookuper) ScanOption {
-	return func(m *Zones, opt *ScanOptions) error {
+	return func(m *Cluster, opt *ScanOptions) error {
 		if h == nil {
 			return fs.ErrInvalid
 		}
@@ -45,7 +50,7 @@ func WithLookuper(h resolver.Lookuper) ScanOption {
 // 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 {
+	return func(m *Cluster, opt *ScanOptions) error {
 		if h == nil {
 			h = resolver.SystemResolver(true)
 		}
@@ -55,55 +60,72 @@ func WithResolver(h resolver.Resolver) ScanOption {
 	}
 }
 
-func (m *Zones) setDefaults(opt *ScanOptions) error {
+// WithLogger specifies what to use for logging
+func WithLogger(log slog.Logger) ScanOption {
+	return func(m *Cluster, opt *ScanOptions) error {
+		if log == nil {
+			log = DefaultLogger()
+		}
+
+		opt.Logger = log
+		m.log = log
+		return nil
+	}
+}
+
+func (m *Cluster) setScanDefaults(opt *ScanOptions) error {
 	if m.resolver == nil {
-		h := resolver.NewCloudflareLookuper()
+		h := DefaultLookuper()
 
 		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) {
+// NewFromDirectory builds a [Cluster] tree using the given directory
+func NewFromDirectory(dir, domain string, opts ...ScanOption) (*Cluster, error) {
 	var scanOptions ScanOptions
 
-	z := &Zones{
-		dir:    dir,
-		domain: domain,
+	dir = filepath.Clean(dir)
+	fullPath, err := filepath.Abs(dir)
+	if err != nil {
+		return nil, err
+	}
+
+	sub, err := os.NewFS().Sub(fullPath[1:])
+	if err != nil {
+		return nil, err
+	}
+
+	m := &Cluster{
+		dir:     sub,
+		BaseDir: dir,
+		Name:    filepath.Base(fullPath),
+		Domain:  domain,
 	}
 
 	for _, opt := range opts {
-		if err := opt(z, &scanOptions); err != nil {
+		if err := opt(m, &scanOptions); err != nil {
 			return nil, err
 		}
 	}
 
-	if err := z.setDefaults(&scanOptions); err != nil {
+	if err := m.setScanDefaults(&scanOptions); err != nil {
 		return nil, err
 	}
 
-	if err := z.scan(&scanOptions); err != nil {
+	if err := m.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...)
+	return m, nil
 }

pkg/cluster/defaults.go

@@ -0,0 +1,17 @@
+package cluster
+
+import (
+	"darvaza.org/resolver"
+	"darvaza.org/slog"
+	"darvaza.org/slog/handlers/discard"
+)
+
+// DefaultLogger returns a logger that doesn't log anything
+func DefaultLogger() slog.Logger {
+	return discard.New()
+}
+
+// DefaultLookuper returns a [resolver.Lookuper] using Cloudflare's 1.1.1.1
+func DefaultLookuper() resolver.Lookuper {
+	return resolver.NewCloudflareLookuper()
+}

pkg/cluster/env.go

@@ -1,4 +1,4 @@
-package zones
+package cluster
 
 import (
 	"bytes"
@@ -11,15 +11,24 @@ import (
 type Env struct {
 	ZoneIterator
 
-	export bool
+	cephFSID string
+	export   bool
 }
 
 // Env returns a shell environment factory
-func (m *Zones) Env(export bool) *Env {
-	return &Env{
+func (m *Cluster) 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
@@ -38,7 +47,12 @@ func (m *Env) Zones() []int {
 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
@@ -59,6 +73,15 @@ func (m *Env) writeEnvZone(w io.Writer, z *Zone) {
 	// 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) {
@@ -111,3 +134,44 @@ func genEnvZoneNodes(z *Zone) string {
 	}
 	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 = (*Cluster)(nil)
+)
+
+func (z *Cluster) withDebug() (slog.Logger, bool) {
+	return z.debug().WithEnabled()
+}
+
+func (z *Cluster) withInfo() (slog.Logger, bool) {
+	return z.debug().WithEnabled()
+}
+
+func (z *Cluster) debug() slog.Logger {
+	return z.log.Debug()
+}
+
+func (z *Cluster) info() slog.Logger {
+	return z.log.Info()
+}
+
+func (z *Cluster) warn(err error) slog.Logger {
+	l := z.log.Warn()
+	if err != nil {
+		l = l.WithField(slog.ErrorFieldName, err)
+	}
+	return l
+}
+
+func (z *Cluster) error(err error) slog.Logger {
+	l := z.log.Error()
+	if err != nil {
+		l = l.WithField(slog.ErrorFieldName, err)
+	}
+	return l
+}

pkg/cluster/machine.go

@@ -1,4 +1,4 @@
-package zones
+package cluster
 
 import (
 	"net/netip"
@@ -9,12 +9,15 @@ import (
 
 // A Machine is a machine on a Zone
 type Machine struct {
-	zone *Zone
-	ID   int    `toml:"id"`
-	Name string `toml:"-" json:"-" yaml:"-"`
+	zone   *Zone
+	logger `json:"-" yaml:"-"`
 
-	PublicAddresses []netip.Addr `toml:"public,omitempty" json:"public,omitempty" yaml:"public,omitempty"`
-	Rings           []*RingInfo  `toml:"rings,omitempty"  json:"rings,omitempty"  yaml:"rings,omitempty"`
+	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
@@ -25,16 +28,19 @@ func (m *Machine) String() string {
 
 // 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,
-		}
+	var name []string
 
-		return strings.Join(s, ".")
+	for _, s := range []string{
+		m.Name,
+		m.zone.zones.Name,
+		m.zone.zones.Domain,
+	} {
+		if s != "" {
+			name = append(name, s)
+		}
 	}
 
-	return m.Name
+	return strings.Join(name, ".")
 }
 
 // IsGateway tells if the Machine is a ring0 gateway

pkg/cluster/machine_file.go

@@ -1,4 +1,4 @@
-package zones
+package cluster
 
 import (
 	"bytes"

pkg/cluster/machine_rings.go

@@ -1,7 +1,8 @@
-package zones
+package cluster
 
 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/cluster/machine_scan.go

@@ -1,10 +1,13 @@
-package zones
+package cluster
 
 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/cluster/machines.go

@@ -0,0 +1,69 @@
+package cluster
+
+import "sort"
+
+var (
+	_ MachineIterator = Machines(nil)
+	_ sort.Interface  = Machines(nil)
+)
+
+// A MachineIterator is a set of Machines we can iterate on
+type MachineIterator interface {
+	ForEachMachine(func(*Machine) 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)
+}

pkg/cluster/rings.go

@@ -1,4 +1,4 @@
-package zones
+package cluster
 
 import (
 	"fmt"
@@ -24,9 +24,9 @@ const (
 // RingInfo contains represents the Wireguard endpoint details
 // for a Machine on a particular ring
 type RingInfo struct {
-	Ring    int               `toml:"ring"`
-	Enabled bool              `toml:"enabled,omitempty"`
-	Keys    wireguard.KeyPair `toml:"keys,omitempty"`
+	Ring    int
+	Enabled bool
+	Keys    wireguard.KeyPair
 }
 
 // Merge attempts to combine two RingInfo structs

pkg/cluster/sync.go

@@ -1,9 +1,10 @@
-package zones
+package cluster
 
 // SyncAll updates all config files
-func (m *Zones) SyncAll() error {
+func (m *Cluster) SyncAll() error {
 	for _, fn := range []func() error{
 		m.SyncAllWireguard,
+		m.SyncAllCeph,
 	} {
 		if err := fn(); err != nil {
 			return err
@@ -14,7 +15,7 @@ func (m *Zones) SyncAll() error {
 }
 
 // SyncAllWireguard updates all wireguard config files
-func (m *Zones) SyncAllWireguard() error {
+func (m *Cluster) SyncAllWireguard() error {
 	var err error
 
 	for ring := 0; ring < RingsCount; ring++ {
@@ -31,3 +32,13 @@ func (m *Zones) SyncAllWireguard() error {
 
 	return nil
 }
+
+// SyncAllCeph updates the ceph.conf file
+func (m *Cluster) SyncAllCeph() error {
+	cfg, err := m.GenCephConfig()
+	if err != nil {
+		return err
+	}
+
+	return m.WriteCephConfig(cfg)
+}

pkg/cluster/wireguard.go

@@ -1,4 +1,4 @@
-package zones
+package cluster
 
 import (
 	"io/fs"
@@ -6,19 +6,19 @@ import (
 )
 
 var (
-	_ WireguardConfigPruner = (*Zones)(nil)
+	_ WireguardConfigPruner = (*Cluster)(nil)
 	_ WireguardConfigPruner = (*Zone)(nil)
 	_ WireguardConfigPruner = (*Machine)(nil)
 
-	_ WireguardConfigWriter = (*Zones)(nil)
+	_ WireguardConfigWriter = (*Cluster)(nil)
 	_ WireguardConfigWriter = (*Zone)(nil)
 	_ WireguardConfigWriter = (*Machine)(nil)
 
-	_ WireguardConfigSyncer = (*Zones)(nil)
+	_ WireguardConfigSyncer = (*Cluster)(nil)
 	_ WireguardConfigSyncer = (*Zone)(nil)
 	_ WireguardConfigSyncer = (*Machine)(nil)
 
-	_ WireguardKeysWriter = (*Zones)(nil)
+	_ WireguardKeysWriter = (*Cluster)(nil)
 	_ WireguardKeysWriter = (*Zone)(nil)
 	_ WireguardKeysWriter = (*Machine)(nil)
 )
@@ -31,7 +31,7 @@ type WireguardConfigPruner interface {
 
 // PruneWireguardConfig removes wgN.conf files of machines with
 // the corresponding ring disabled on all zones
-func (m *Zones) PruneWireguardConfig(ring int) error {
+func (m *Cluster) PruneWireguardConfig(ring int) error {
 	return pruneWireguardConfig(m, ring)
 }
 
@@ -76,7 +76,7 @@ type WireguardConfigWriter interface {
 
 // WriteWireguardConfig rewrites all wgN.conf on all machines
 // attached to that ring
-func (m *Zones) WriteWireguardConfig(ring int) error {
+func (m *Cluster) WriteWireguardConfig(ring int) error {
 	switch ring {
 	case 0:
 		return writeWireguardConfig(m, m, ring)
@@ -154,7 +154,7 @@ type WireguardConfigSyncer interface {
 
 // SyncWireguardConfig updates all wgN.conf files for the specified
 // ring
-func (m *Zones) SyncWireguardConfig(ring int) error {
+func (m *Cluster) SyncWireguardConfig(ring int) error {
 	switch ring {
 	case 0:
 		return syncWireguardConfig(m, m, ring)
@@ -214,7 +214,7 @@ type WireguardKeysWriter interface {
 }
 
 // WriteWireguardKeys rewrites all wgN.{key,pub} files
-func (m *Zones) WriteWireguardKeys(ring int) error {
+func (m *Cluster) WriteWireguardKeys(ring int) error {
 	return writeWireguardKeys(m, ring)
 }
 

pkg/cluster/zones.go

@@ -0,0 +1,68 @@
+package cluster
+
+import (
+	"io/fs"
+)
+
+var (
+	_ MachineIterator = (*Zone)(nil)
+)
+
+// A ZoneIterator is a set of Zones we can iterate on
+type ZoneIterator interface {
+	ForEachZone(func(*Zone) bool)
+}
+
+// A Zone is a set of machines in close proximity and strong
+// affinity.
+type Zone struct {
+	zones  *Cluster
+	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)
+}

pkg/zones/zones.go

@@ -1,189 +0,0 @@
-// Package zones contains information about the cluster
-package zones
-
-import (
-	"io/fs"
-	"sort"
-
-	"darvaza.org/resolver"
-)
-
-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
-
-	ID   int    `toml:"id"`
-	Name string `toml:"name"`
-
-	Machines `toml:"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)
-}
-
-// Zones represents all zones in a cluster
-type Zones struct {
-	dir      fs.FS
-	resolver resolver.Resolver
-	domain   string
-
-	Zones []*Zone `toml:"zones"`
-}
-
-// 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
-}