add `-retry` option for `volume.fix.replication`

fix https://github.com/chrislusf/seaweedfs/issues/2191

1 files changed, 8 insertions, 7 deletions

diff --git a/weed/shell/command_volume_fix_replication.go b/weed/shell/command_volume_fix_replication.go
index b48bd3ea0..d1f830efb 100644
--- a/weed/shell/command_volume_fix_replication.go
+++ b/weed/shell/command_volume_fix_replication.go
@@ -58,6 +58,7 @@ func (c *commandVolumeFixReplication) Do(args []string, commandEnv *CommandEnv,
 	volFixReplicationCommand := flag.NewFlagSet(c.Name(), flag.ContinueOnError)
 	c.collectionPattern = volFixReplicationCommand.String("collectionPattern", "", "match with wildcard characters '*' and '?'")
 	skipChange := volFixReplicationCommand.Bool("n", false, "skip the changes")
+	retryCount := volFixReplicationCommand.Int("retry", 0, "how many times to retry")
 	if err = volFixReplicationCommand.Parse(args); err != nil {
 		return nil
 	}
@@ -100,7 +101,7 @@ func (c *commandVolumeFixReplication) Do(args []string, commandEnv *CommandEnv,
 	}
 
 	// find the most under populated data nodes
-	return c.fixUnderReplicatedVolumes(commandEnv, writer, takeAction, underReplicatedVolumeIds, volumeReplicas, allLocations)
+	return c.fixUnderReplicatedVolumes(commandEnv, writer, takeAction, underReplicatedVolumeIds, volumeReplicas, allLocations, *retryCount)
 
 }
 
@@ -154,16 +155,16 @@ func (c *commandVolumeFixReplication) fixOverReplicatedVolumes(commandEnv *Comma
 	return nil
 }
 
-func (c *commandVolumeFixReplication) fixUnderReplicatedVolumes(commandEnv *CommandEnv, writer io.Writer, takeAction bool, underReplicatedVolumeIds []uint32, volumeReplicas map[uint32][]*VolumeReplica, allLocations []location) error {
+func (c *commandVolumeFixReplication) fixUnderReplicatedVolumes(commandEnv *CommandEnv, writer io.Writer, takeAction bool, underReplicatedVolumeIds []uint32, volumeReplicas map[uint32][]*VolumeReplica, allLocations []location, retryCount int) (err error) {
 
 	for _, vid := range underReplicatedVolumeIds {
-		err := c.fixOneUnderReplicatedVolume(commandEnv, writer, takeAction, volumeReplicas, vid, allLocations)
-		if err != nil {
-			return err
+		for i:=0;i<retryCount+1;i++{
+			if err = c.fixOneUnderReplicatedVolume(commandEnv, writer, takeAction, volumeReplicas, vid, allLocations); err == nil {
+				continue
+			}
 		}
-
 	}
-	return nil
+	return
 }
 
 func (c *commandVolumeFixReplication) fixOneUnderReplicatedVolume(commandEnv *CommandEnv, writer io.Writer, takeAction bool, volumeReplicas map[uint32][]*VolumeReplica, vid uint32, allLocations []location) error {