From e55bcbb69e9d15bc4556d4dcb3160ca8cd4ccc81 Mon Sep 17 00:00:00 2001
From: Matt Lord <mattalord@gmail.com>
Date: Wed, 27 Sep 2023 18:25:10 -0400
Subject: [PATCH] Stop the controller for a VDiff before deleting it

This allows you safely delete VDiffs that were still running.

Signed-off-by: Matt Lord <mattalord@gmail.com>
---
 go/test/endtoend/vreplication/vdiff2_test.go |  4 ++--
 go/vt/vttablet/tabletmanager/vdiff/action.go | 25 ++++++++++++++++++++
 2 files changed, 27 insertions(+), 2 deletions(-)

diff --git a/go/test/endtoend/vreplication/vdiff2_test.go b/go/test/endtoend/vreplication/vdiff2_test.go
index 0f34782a43c..72b09e8fede 100644
--- a/go/test/endtoend/vreplication/vdiff2_test.go
+++ b/go/test/endtoend/vreplication/vdiff2_test.go
@@ -190,11 +190,11 @@ func testWorkflow(t *testing.T, vc *VitessCluster, tc *testCase, tks *Keyspace,
 	vdiff(t, tc.targetKs, tc.workflow, allCellNames, true, true, nil)
 
 	if tc.autoRetryError {
-		testAutoRetryError(t, tc, cells[0].Name)
+		testAutoRetryError(t, tc, allCellNames)
 	}
 
 	if tc.resume {
-		testResume(t, tc, cells[0].Name)
+		testResume(t, tc, allCellNames)
 	}
 
 	// These are done here so that we have a valid workflow to test the commands against
diff --git a/go/vt/vttablet/tabletmanager/vdiff/action.go b/go/vt/vttablet/tabletmanager/vdiff/action.go
index f7d68349ca0..5f957964b3d 100644
--- a/go/vt/vttablet/tabletmanager/vdiff/action.go
+++ b/go/vt/vttablet/tabletmanager/vdiff/action.go
@@ -354,9 +354,13 @@ func (vde *Engine) handleStopAction(ctx context.Context, dbClient binlogplayer.D
 func (vde *Engine) handleDeleteAction(ctx context.Context, dbClient binlogplayer.DBClient, action VDiffAction, req *tabletmanagerdatapb.VDiffRequest, resp *tabletmanagerdatapb.VDiffResponse) error {
 	var err error
 	query := ""
+	vde.mu.Lock()
+	defer vde.mu.Unlock()
 
 	switch req.ActionArg {
 	case AllActionArg:
+		// We need to stop all running controllers.
+		vde.resetControllers()
 		query, err = sqlparser.ParseAndBind(sqlDeleteVDiffs,
 			sqltypes.StringBindVariable(req.Keyspace),
 			sqltypes.StringBindVariable(req.Workflow),
@@ -369,6 +373,27 @@ func (vde *Engine) handleDeleteAction(ctx context.Context, dbClient binlogplayer
 		if err != nil {
 			return fmt.Errorf("action argument %s not supported", req.ActionArg)
 		}
+		// We need to be sure that the controller is cleaned up, if it's
+		// still running, before we delete the vdiff record.
+		query, err = sqlparser.ParseAndBind(sqlGetVDiffID,
+			sqltypes.StringBindVariable(uuid.String()),
+		)
+		if err != nil {
+			return err
+		}
+		res, err := dbClient.ExecuteFetch(query, 1)
+		if err != nil {
+			return err
+		}
+		row := res.Named().Row() // Must only be one
+		if row == nil {
+			return vterrors.Errorf(vtrpcpb.Code_FAILED_PRECONDITION, "no vdiff found for UUID %s keyspace %s and workflow %s on tablet %v",
+				uuid, req.Keyspace, req.Workflow, vde.thisTablet.Alias)
+		}
+		controller, ok := vde.controllers[int64(row.AsInt64("id", 0))]
+		if ok {
+			controller.Stop()
+		}
 		query, err = sqlparser.ParseAndBind(sqlDeleteVDiffByUUID, sqltypes.StringBindVariable(uuid.String()))
 		if err != nil {
 			return err