diff --git a/ddl/ddl.go b/ddl/ddl.go
index c484c7b23e..3c5d41b3c2 100644
--- a/ddl/ddl.go
+++ b/ddl/ddl.go
@@ -76,6 +76,10 @@ type ddl struct {
 	// use a persistent job list.
 	reOrgDoneCh chan error
 
+	// reOrgHandle is used for adding data reorgnization, after every batch,
+	// we will update for later job update.
+	reOrgHandle int64
+
 	quitCh chan struct{}
 	wait   sync.WaitGroup
 }
diff --git a/ddl/index.go b/ddl/index.go
index 0670a7f3cc..cfdc2aee02 100644
--- a/ddl/index.go
+++ b/ddl/index.go
@@ -16,6 +16,7 @@ package ddl
 import (
 	"bytes"
 	"strings"
+	"sync/atomic"
 
 	"github.com/juju/errors"
 	"github.com/ngaut/log"
@@ -193,6 +194,9 @@ func (d *ddl) onIndexCreate(t *meta.Meta, job *model.Job) error {
 		indexInfo.State = model.StateReorgnization
 		// initialize SnapshotVer to 0 for later reorgnization check.
 		job.SnapshotVer = 0
+		// initialize reorg handle to 0
+		job.ReOrgHandle = 0
+		atomic.StoreInt64(&d.reOrgHandle, 0)
 		err = t.UpdateTable(schemaID, tblInfo)
 		return errors.Trace(err)
 	case model.StateReorgnization:
@@ -215,9 +219,13 @@ func (d *ddl) onIndexCreate(t *meta.Meta, job *model.Job) error {
 		}
 
 		err = d.runReorgJob(func() error {
-			return d.addTableIndex(tbl, indexInfo, job.SnapshotVer)
+			return d.addTableIndex(tbl, indexInfo, job.SnapshotVer, job.ReOrgHandle)
 		})
 
+		// addTableIndex updates ReOrgHandle after one batch.
+		// so we update the job ReOrgHandle here.
+		job.ReOrgHandle = atomic.LoadInt64(&d.reOrgHandle)
+
 		if errors2.ErrorEqual(err, errWaitReorgTimeout) {
 			// if timeout, we should return, check for the owner and re-wait job done.
 			return nil
@@ -377,8 +385,7 @@ const maxBatchSize = 1024
 //  3, For one row, if the row has been already deleted, skip to next row.
 //  4, If not deleted, check whether index has existed, if existed, skip to next row.
 //  5, If index doesn't exist, create the index and then continue to handle next row.
-func (d *ddl) addTableIndex(t table.Table, indexInfo *model.IndexInfo, version uint64) error {
-	seekHandle := int64(0)
+func (d *ddl) addTableIndex(t table.Table, indexInfo *model.IndexInfo, version uint64, seekHandle int64) error {
 	for {
 		handles, err := d.getSnapshotRows(t, version, seekHandle)
 		if err != nil {
@@ -388,12 +395,14 @@ func (d *ddl) addTableIndex(t table.Table, indexInfo *model.IndexInfo, version u
 		}
 
 		seekHandle = handles[len(handles)-1] + 1
-		// TODO: save seekHandle in reorgnization job, so we can resume this job later from this handle.
 
 		err = d.backfillTableIndex(t, indexInfo, handles)
 		if err != nil {
 			return errors.Trace(err)
 		}
+
+		// update reOrgHandle here after every successful batch.
+		atomic.StoreInt64(&d.reOrgHandle, seekHandle)
 	}
 }
 
diff --git a/ddl/reorg.go b/ddl/reorg.go
index e17237b7ab..19c6e0f4b3 100644
--- a/ddl/reorg.go
+++ b/ddl/reorg.go
@@ -98,7 +98,6 @@ const waitReorgTimeout = 10 * time.Second
 var errWaitReorgTimeout = errors.New("wait for reorgnization timeout")
 
 func (d *ddl) runReorgJob(f func() error) error {
-	// TODO use persistent reorgnization job list.
 	if d.reOrgDoneCh == nil {
 		// start a reorgnization job
 		d.reOrgDoneCh = make(chan error, 1)
diff --git a/model/ddl.go b/model/ddl.go
index 33ee20554f..9f1f37e8ab 100644
--- a/model/ddl.go
+++ b/model/ddl.go
@@ -80,6 +80,11 @@ type Job struct {
 	// snapshot version for this job.
 	SnapshotVer  uint64 `json:"snapshot_ver"`
 	LastUpdateTS int64  `json:"last_update_ts"`
+	// For reorgnization adding data job like add index, add column,
+	// we will traverse the huge snapshot and add the data in batches.
+	// After some successful batches, we will update the ReorgHandle for
+	// later resuming if currect server crashed.
+	ReOrgHandle int64 `json:"reorg_handle"`
 }
 
 // Encode encodes job with json format.