.

go/neo/master.go

@@ -388,9 +388,8 @@ func (m *Master) recovery(ctx context.Context) (err error) {
 	wg := &sync.WaitGroup{}
 
 //trace:event traceMasterStartReady(m *Master, ready bool)
-	readyToStart := false
+	readyToStart := false // whether cluster currently can be operational or not
 	updateReadyToStart := func() {
-		// update indicator whether cluster currently can be operational or not
 		var ready bool
 		if m.node.State.PartTab.PTid == 0 {
 			// new cluster - allow startup if we have some storages passed
@@ -416,16 +415,35 @@ func (m *Master) recovery(ctx context.Context) (err error) {
 	// XXX set cluster state = RECOVERY
 	// XXX down clients
 
+	// goStorCtlRecovery spawns recovery task on a storage.
+	goStorCtlRecovery := func(stor *_MasteredPeer) {
+		inprogress++
+		// XXX wg.Add(1) + defer wg.Done() ?
+		stor.wg.Go(func(peerCtx context.Context) error {
+			ctx, cancel := xxcontext.Merge/*Cancel*/(ctx, peerCtx)
+			defer cancel()
+
+			var pt *xneo.PartitionTable
+			err := stor.run(ctx, func(...) {
+				pt, err = storCtlRecovery(...)
+			})
+
+			ack := make(chan struct{})
+			recoveryq <- storRecovery{stor: stor, partTab: pt, err: err, ack: ack}
+			<-ack
+
+			// canceled recovery does not mean we should down the storage node
+			if xcontext.Canceled(err) {
+				err = nil
+			}
+			return err
+		})
+	}
+
 	// start recovery on all storages we are currently in touch with
-	for _, stor := range m.node.State.NodeTab.StorageList() {
-		if stor.State > proto.DOWN {	// XXX state cmp ok ? XXX or stor.Link != nil ?
-			inprogress++
-			wg.Add(1)
-			go func() {
-				defer wg.Done()
-				storCtlRecovery(ctx, stor, recoveryq)
-				// XXX acj
-			}()
+	for _, peer := range m.peerTab {
+		if peer.node.Type == proto.STORAGE {
+			goStorCtlRecovery(peer)
 		}
 	}
 
@@ -486,31 +504,10 @@ loop:
 			}
 
 			// S -> start recovery
-			if peer.node.Type != proto.STORAGE {
-				break
+			if peer.node.Type == proto.STORAGE {
+				goStorCtlRecovery(peer)
 			}
 
-			// XXX wg.Add(1) + defer wg.Done() ?
-			peer.wg.Go(func(peerCtx context.Context) error {
-				ctx, cancel := xxcontext.Merge/*Cancel*/(ctx, peerCtx)
-				defer cancel()
-
-				var pt *xneo.PartitionTable
-				err := peer.run(ctx, func(...) {
-					pt, err = storCtlRecovery(...)
-				})
-
-				ack := make(chan struct{})
-				recoveryq <- storRecovery{stor: peer, partTab: pt, err: err, ack: ack}
-				<-ack
-
-				// canceled recovery does not mean we should down the peer
-				if xcontext.Canceled(err) {
-					err = nil
-				}
-				return err
-			})
-
 		// a storage node came through recovery - let's see whether
 		// ptid ↑ and if so we should take partition table from there
 		case r := <-recoveryq:
@@ -530,8 +527,6 @@ loop:
 			}
 
 			updateReadyToStart()
-
-
 		}
 	}