Check aliveness of all nodes in a single etcd query

Signed-off-by: Oshank Kumar <okumar@redhat.com>

glusterd2/store/liveness.go

@@ -3,6 +3,7 @@ package store
 import (
 	"context"
 	"os"
+	"path"
 	"strconv"
 	"time"
 
@@ -56,6 +57,52 @@ func (s *GDStore) IsNodeAlive(peerID interface{}) (int, bool) {
 	return 0, false
 }
 
+// GetAliveNodes will return a map of all alive nodes. It will contain
+// peerID and corresponding process id. It uses single etcd query
+// to get all alive nodes.
+func (s *GDStore) GetAliveNodes(ctx context.Context) map[string]int {
+	peerStatus := map[string]int{}
+
+	resp, err := s.Get(ctx, LivenessKeyPrefix, clientv3.WithPrefix())
+	if err != nil {
+		return peerStatus
+	}
+
+	for _, kv := range resp.Kvs {
+		peerID := path.Base(string(kv.Key))
+		pid, err := strconv.Atoi(string(kv.Value))
+		if err == nil {
+			peerStatus[peerID] = pid
+		}
+	}
+
+	return peerStatus
+}
+
+// AreNodesAlive returns true if all given nodes are alive.
+func (s *GDStore) AreNodesAlive(ctx context.Context, peerIDs ...uuid.UUID) bool {
+	var nodeIDs []string
+
+	for _, peerID := range peerIDs {
+		nodeIDs = append(nodeIDs, peerID.String())
+
+	}
+
+	peerStatus := s.GetAliveNodes(ctx)
+
+	if len(nodeIDs) > len(peerStatus) {
+		return false
+	}
+
+	for _, nodeID := range nodeIDs {
+		if _, ok := peerStatus[nodeID]; !ok {
+			return false
+		}
+	}
+
+	return true
+}
+
 func (s *GDStore) publishLiveness() error {
 	// publish liveness of this instance into the store
 	key := LivenessKeyPrefix + gdctx.MyUUID.String()

glusterd2/transactionv2/transaction.go

@@ -124,11 +124,8 @@ func (t *Txn) removeContextData() {
 }
 
 func (t *Txn) checkAlive() error {
-	for _, node := range t.Nodes {
-		// TODO: Using prefixed query, get all alive nodes in a single etcd query
-		if _, online := store.Store.IsNodeAlive(node); !online {
-			return fmt.Errorf("node %s is probably down", node.String())
-		}
+	if online := store.Store.AreNodesAlive(context.Background(), t.Nodes...); !online {
+		return errors.New("probably some of the nodes are down")
 	}
 	return nil
 }