From c038ad4431650d27797924cfb2324fdae8160075 Mon Sep 17 00:00:00 2001
From: Alejandro Mery <amery@jpi.io>
Date: Thu, 7 Sep 2023 18:32:11 +0000
Subject: [PATCH] zones: warn but not fail when scanning finds unknown monitors

Signed-off-by: Alejandro Mery <amery@jpi.io>
---
 pkg/zones/ceph_scan.go | 104 +++++------------------------------------
 1 file changed, 12 insertions(+), 92 deletions(-)

diff --git a/pkg/zones/ceph_scan.go b/pkg/zones/ceph_scan.go
index 2180c09..7cb4f37 100644
--- a/pkg/zones/ceph_scan.go
+++ b/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