env: set ceph monitors variables

they indicate the ceph monitors on the specified zone

* MON{zoneID}_NAME
* MON{zoneID}_ID
* MON{zoneID}_IP

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

go.mod

@@ -3,6 +3,7 @@ module git.jpi.io/amery/jpictl
 go 1.19
 
 require (
+	asciigoat.org/ini v0.2.5
 	darvaza.org/core v0.9.8
 	darvaza.org/resolver v0.5.4
 	darvaza.org/sidecar v0.0.2
@@ -18,6 +19,7 @@ require (
 )
 
 require (
+	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

go.sum

@@ -1,3 +1,7 @@
+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=

pkg/ceph/ceph.go

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

pkg/ceph/config.go

@@ -0,0 +1,38 @@
+package ceph
+
+import (
+	"io"
+	"net/netip"
+
+	"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_host,comma"`
+	MonitorsAddr   []netip.Addr `ini:"mon_initial_members,comma"`
+	ClusterNetwork netip.Prefix `ini:"cluster_network"`
+}
+
+// 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 +1,108 @@
+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
+
+	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/zones/ceph.go

@@ -1,8 +1,14 @@
 package zones
 
 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
@@ -13,3 +19,88 @@ func (m *Zones) GetCephFSID() (uuid.UUID, error) {
 	}
 	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)
+}
+
+// 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/zones/ceph_scan.go

@@ -0,0 +1,176 @@
+package zones
+
+import (
+	"net/netip"
+	"os"
+	"strings"
+
+	"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_host:")
+		for i, name := range err.Names {
+			if i != 0 {
+				_, _ = w.WriteRune(',')
+			}
+			_, _ = w.WriteString(name)
+		}
+	}
+}
+
+func (err *CephMissingMonitorError) writeAddrs(w *strings.Builder) {
+	_, _ = w.WriteString(" mon_initial_members:")
+	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
+}
+
+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) Missing() error {
+	var check CephMissingMonitorError
+
+	for name, found := range todo.names {
+		if !found {
+			check.appendName(name)
+		}
+	}
+
+	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)
+		}
+	}
+
+	return check.AsError()
+}
+
+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
+	}
+
+	// flag monitors based on config
+	todo := newCephScanTODO(cfg)
+	m.ForEachMachine(func(p *Machine) bool {
+		p.CephMonitor = todo.checkMachine(p)
+		return false
+	})
+	if err := todo.Missing(); err != nil {
+		return core.Wrap(err, "ceph")
+	}
+
+	// make sure every zone has one
+	m.ForEachZone(func(z *Zone) bool {
+		_ = z.GetCephMonitors()
+		return false
+	})
+	return nil
+}

pkg/zones/env.go

@@ -59,6 +59,15 @@ 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 +120,44 @@ 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/zones/scan.go

@@ -12,6 +12,7 @@ func (m *Zones) scan(opts *ScanOptions) error {
 		m.scanZoneIDs,
 		m.scanSort,
 		m.scanGateways,
+		m.scanCephMonitors,
 	} {
 		if err := fn(opts); err != nil {
 			return err