sql/gcjob/gcjobnotifier: fix rare panic

If a notifier is added before the initial system config, it can result in a
crash. This may be more common in 22.1, where the initial config takes longer
to be populated.

Fixes #77425

Release note (bug fix): Fixes a rare crash which can occur when restarting a
node after dropping tables.

pkg/sql/gcjob/gcjobnotifier/notifier.go

@@ -86,8 +86,11 @@ func (n *Notifier) AddNotifyee(ctx context.Context) (onChange <-chan struct{}, c
 	if n.mu.deltaFilter == nil {
 		zoneCfgFilter := gossip.MakeSystemConfigDeltaFilter(n.prefix)
 		n.mu.deltaFilter = &zoneCfgFilter
-		// Initialize the filter with the current values.
-		n.mu.deltaFilter.ForModified(n.provider.GetSystemConfig(), func(kv roachpb.KeyValue) {})
+		// Initialize the filter with the current values, if they exist.
+		cfg := n.provider.GetSystemConfig()
+		if cfg != nil {
+			n.mu.deltaFilter.ForModified(cfg, func(kv roachpb.KeyValue) {})
+		}
 	}
 	c := make(chan struct{}, 1)
 	n.mu.notifyees[c] = struct{}{}

pkg/sql/gcjob/gcjobnotifier/notifier_test.go

@@ -80,6 +80,23 @@ func TestNotifier(t *testing.T) {
 			n.AddNotifyee(ctx)
 		})
 	})
+	t.Run("safe to on AddNotifyee after start before config", func(t *testing.T) {
+		ch := make(chan struct{}, 1)
+		cfg := mkSystemConfig(mkZoneConfigKV(1, 1, "1"))
+		p := &testingProvider{ch: ch}
+		n := New(settings, p, keys.SystemSQLCodec, stopper)
+		n.Start(ctx)
+		n1Ch, cleanup := n.AddNotifyee(ctx)
+		defer cleanup()
+		select {
+		case <-time.After(10 * time.Millisecond):
+		case <-n1Ch:
+			t.Fatal("should not have gotten notified")
+		}
+		p.setSystemConfig(cfg)
+		ch <- struct{}{}
+		<-n1Ch
+	})
 	t.Run("notifies on changed delta and cleanup", func(t *testing.T) {
 		cfg := config.NewSystemConfig(zonepb.DefaultSystemZoneConfigRef())
 		cfg.Values = []roachpb.KeyValue{