zones: set Machine.CephMonitor if its referenced as monitor on ceph.conf

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

pkg/zones/ceph_scan.go

@@ -0,0 +1,173 @@
+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
+	}
+
+	if cfg != nil {
+		// 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")
+		}
+	}
+
+	return nil
+}

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