Fix(Dgraph): Add a lock to backups to process one request at a time. (#…

…6316)

It's possible that two requests reach the server around the same time
and send a requests to the alphas with the same backupNum. This could
lead to issues further down the line.

Related to DGRAPH-2295

worker/backup_ee.go

@@ -16,6 +16,7 @@ import (
 	"context"
 	"net/url"
 	"sort"
+	"sync"
 	"time"
 
 	"github.com/dgraph-io/dgraph/posting"
@@ -80,6 +81,11 @@ func BackupGroup(ctx context.Context, in *pb.BackupRequest) (*pb.Status, error)
 	return res, nil
 }
 
+// backupLock is used to synchronize backups to avoid more than one backup request
+// to be processed at the same time. Multiple requests could lead to multiple
+// backups with the same backupNum in their manifest.
+var backupLock sync.Mutex
+
 func ProcessBackupRequest(ctx context.Context, req *pb.BackupRequest, forceFull bool) error {
 	if !EnterpriseEnabled() {
 		return errors.New("you must enable enterprise features first. " +
@@ -95,6 +101,10 @@ func ProcessBackupRequest(ctx context.Context, req *pb.BackupRequest, forceFull
 		return err
 	}
 
+	// Grab the lock here to avoid more than one request to be processed at the same time.
+	backupLock.Lock()
+	defer backupLock.Unlock()
+
 	ts, err := Timestamps(ctx, &pb.Num{ReadOnly: true})
 	if err != nil {
 		glog.Errorf("Unable to retrieve readonly timestamp for backup: %s", err)