feat: add allow cross cell promotion flag to prs

Signed-off-by: Manan Gupta <manan@planetscale.com>

go/cmd/vtctldclient/command/reparents.go

@@ -187,6 +187,7 @@ var plannedReparentShardOptions = struct {
 	AvoidPrimaryAliasStr    string
 	WaitReplicasTimeout     time.Duration
 	TolerableReplicationLag time.Duration
+	AllowCrossCellPromotion bool
 }{}
 
 func commandPlannedReparentShard(cmd *cobra.Command, args []string) error {
@@ -223,6 +224,7 @@ func commandPlannedReparentShard(cmd *cobra.Command, args []string) error {
 		AvoidPrimary:            avoidPrimaryAlias,
 		WaitReplicasTimeout:     protoutil.DurationToProto(plannedReparentShardOptions.WaitReplicasTimeout),
 		TolerableReplicationLag: protoutil.DurationToProto(plannedReparentShardOptions.TolerableReplicationLag),
+		AllowCrossCellPromotion: plannedReparentShardOptions.AllowCrossCellPromotion,
 	})
 	if err != nil {
 		return err
@@ -297,6 +299,7 @@ func init() {
 	PlannedReparentShard.Flags().DurationVar(&plannedReparentShardOptions.TolerableReplicationLag, "tolerable-replication-lag", 0, "Amount of replication lag that is considered acceptable for a tablet to be eligible for promotion when Vitess makes the choice of a new primary.")
 	PlannedReparentShard.Flags().StringVar(&plannedReparentShardOptions.NewPrimaryAliasStr, "new-primary", "", "Alias of a tablet that should be the new primary.")
 	PlannedReparentShard.Flags().StringVar(&plannedReparentShardOptions.AvoidPrimaryAliasStr, "avoid-primary", "", "Alias of a tablet that should not be the primary; i.e. \"reparent to any other tablet if this one is the primary\".")
+	PlannedReparentShard.Flags().BoolVar(&plannedReparentShardOptions.AllowCrossCellPromotion, "allow-cross-cell-promotion", false, "Allow cross cell promotion")
 	Root.AddCommand(PlannedReparentShard)
 
 	Root.AddCommand(ReparentTablet)

go/vt/vtctl/grpcvtctldserver/server.go

@@ -2976,10 +2976,11 @@ func (s *VtctldServer) PlannedReparentShard(ctx context.Context, req *vtctldatap
 		req.Keyspace,
 		req.Shard,
 		reparentutil.PlannedReparentOptions{
-			AvoidPrimaryAlias:   req.AvoidPrimary,
-			NewPrimaryAlias:     req.NewPrimary,
-			WaitReplicasTimeout: waitReplicasTimeout,
-			TolerableReplLag:    tolerableReplLag,
+			AvoidPrimaryAlias:       req.AvoidPrimary,
+			NewPrimaryAlias:         req.NewPrimary,
+			WaitReplicasTimeout:     waitReplicasTimeout,
+			TolerableReplLag:        tolerableReplLag,
+			AllowCrossCellPromotion: req.AllowCrossCellPromotion,
 		},
 	)
 

go/vt/vtctl/reparent.go

@@ -119,6 +119,7 @@ func commandPlannedReparentShard(ctx context.Context, wr *wrangler.Wrangler, sub
 	keyspaceShard := subFlags.String("keyspace_shard", "", "keyspace/shard of the shard that needs to be reparented")
 	newPrimary := subFlags.String("new_primary", "", "alias of a tablet that should be the new primary")
 	avoidTablet := subFlags.String("avoid_tablet", "", "alias of a tablet that should not be the primary, i.e. reparent to any other tablet if this one is the primary")
+	allowCrossCellPromotion := subFlags.Bool("allow-cross-cell-promotion", false, "allow cross cell promotions")
 
 	if err := subFlags.Parse(args); err != nil {
 		return err
@@ -153,10 +154,11 @@ func commandPlannedReparentShard(ctx context.Context, wr *wrangler.Wrangler, sub
 	}
 
 	return wr.PlannedReparentShard(ctx, keyspace, shard, reparentutil.PlannedReparentOptions{
-		NewPrimaryAlias:     newPrimaryAlias,
-		AvoidPrimaryAlias:   avoidTabletAlias,
-		WaitReplicasTimeout: *waitReplicasTimeout,
-		TolerableReplLag:    *tolerableReplicationLag,
+		NewPrimaryAlias:         newPrimaryAlias,
+		AvoidPrimaryAlias:       avoidTabletAlias,
+		WaitReplicasTimeout:     *waitReplicasTimeout,
+		TolerableReplLag:        *tolerableReplicationLag,
+		AllowCrossCellPromotion: *allowCrossCellPromotion,
 	})
 }
 

go/vt/vtctl/reparentutil/planned_reparenter.go

@@ -57,10 +57,11 @@ type PlannedReparenter struct {
 // operations. Options are passed by value, so it is safe for callers to mutate
 // resue options structs for multiple calls.
 type PlannedReparentOptions struct {
-	NewPrimaryAlias     *topodatapb.TabletAlias
-	AvoidPrimaryAlias   *topodatapb.TabletAlias
-	WaitReplicasTimeout time.Duration
-	TolerableReplLag    time.Duration
+	NewPrimaryAlias         *topodatapb.TabletAlias
+	AvoidPrimaryAlias       *topodatapb.TabletAlias
+	WaitReplicasTimeout     time.Duration
+	TolerableReplLag        time.Duration
+	AllowCrossCellPromotion bool
 
 	// Private options managed internally. We use value-passing semantics to
 	// set these options inside a PlannedReparent without leaking these details
@@ -178,7 +179,7 @@ func (pr *PlannedReparenter) preflightChecks(
 	}
 
 	event.DispatchUpdate(ev, "electing a primary candidate")
-	opts.NewPrimaryAlias, err = ElectNewPrimary(ctx, pr.tmc, &ev.ShardInfo, tabletMap, opts.NewPrimaryAlias, opts.AvoidPrimaryAlias, opts.WaitReplicasTimeout, opts.TolerableReplLag, opts.durability, pr.logger)
+	opts.NewPrimaryAlias, err = ElectNewPrimary(ctx, pr.tmc, &ev.ShardInfo, tabletMap, opts.NewPrimaryAlias, opts.AvoidPrimaryAlias, opts.WaitReplicasTimeout, opts.TolerableReplLag, opts.durability, opts.AllowCrossCellPromotion, pr.logger)
 	if err != nil {
 		return true, err
 	}

go/vt/vtctl/reparentutil/util.go

@@ -73,6 +73,7 @@ func ElectNewPrimary(
 	waitReplicasTimeout time.Duration,
 	tolerableReplLag time.Duration,
 	durability Durabler,
+	allowCrossCellPromotion bool,
 	// (TODO:@ajm188) it's a little gross we need to pass this, maybe embed in the context?
 	logger logutil.Logger,
 ) (*topodatapb.TabletAlias, error) {
@@ -102,7 +103,7 @@ func ElectNewPrimary(
 				reasonsToInvalidate.WriteString(fmt.Sprintf("\n%v does not match the new primary alias provided", topoproto.TabletAliasString(tablet.Alias)))
 				continue
 			}
-		case primaryCell != "" && tablet.Alias.Cell != primaryCell:
+		case !allowCrossCellPromotion && primaryCell != "" && tablet.Alias.Cell != primaryCell:
 			reasonsToInvalidate.WriteString(fmt.Sprintf("\n%v is not in the same cell as the previous primary", topoproto.TabletAliasString(tablet.Alias)))
 			continue
 		case avoidPrimaryAlias != nil && topoproto.TabletAliasEqual(tablet.Alias, avoidPrimaryAlias):

go/vt/vtctl/reparentutil/util_test.go

@@ -67,15 +67,16 @@ func TestElectNewPrimary(t *testing.T) {
 	ctx := context.Background()
 	logger := logutil.NewMemoryLogger()
 	tests := []struct {
-		name              string
-		tmc               *chooseNewPrimaryTestTMClient
-		shardInfo         *topo.ShardInfo
-		tabletMap         map[string]*topo.TabletInfo
-		newPrimaryAlias   *topodatapb.TabletAlias
-		avoidPrimaryAlias *topodatapb.TabletAlias
-		tolerableReplLag  time.Duration
-		expected          *topodatapb.TabletAlias
-		errContains       []string
+		name                    string
+		tmc                     *chooseNewPrimaryTestTMClient
+		shardInfo               *topo.ShardInfo
+		tabletMap               map[string]*topo.TabletInfo
+		newPrimaryAlias         *topodatapb.TabletAlias
+		avoidPrimaryAlias       *topodatapb.TabletAlias
+		tolerableReplLag        time.Duration
+		allowCrossCellPromotion bool
+		expected                *topodatapb.TabletAlias
+		errContains             []string
 	}{
 		{
 			name: "found a replica",
@@ -653,6 +654,64 @@ func TestElectNewPrimary(t *testing.T) {
 				`zone1-0000000102 is not in the same cell as the previous primary`,
 			},
 		},
+		{
+			name: "no replicas in primary cell but cross cell allowed",
+			tmc: &chooseNewPrimaryTestTMClient{
+				// zone1-101 is behind zone1-102
+				replicationStatuses: map[string]*replicationdatapb.Status{
+					"zone1-0000000101": {
+						Position: "MySQL56/3E11FA47-71CA-11E1-9E33-C80AA9429562:1",
+					},
+					"zone1-0000000102": {
+						Position: "MySQL56/3E11FA47-71CA-11E1-9E33-C80AA9429562:1-5",
+					},
+				},
+			},
+			allowCrossCellPromotion: true,
+			shardInfo: topo.NewShardInfo("testkeyspace", "-", &topodatapb.Shard{
+				PrimaryAlias: &topodatapb.TabletAlias{
+					Cell: "zone2",
+					Uid:  200,
+				},
+			}, nil),
+			tabletMap: map[string]*topo.TabletInfo{
+				"primary": {
+					Tablet: &topodatapb.Tablet{
+						Alias: &topodatapb.TabletAlias{
+							Cell: "zone2",
+							Uid:  200,
+						},
+						Type: topodatapb.TabletType_PRIMARY,
+					},
+				},
+				"replica1": {
+					Tablet: &topodatapb.Tablet{
+						Alias: &topodatapb.TabletAlias{
+							Cell: "zone1",
+							Uid:  101,
+						},
+						Type: topodatapb.TabletType_REPLICA,
+					},
+				},
+				"replica2": {
+					Tablet: &topodatapb.Tablet{
+						Alias: &topodatapb.TabletAlias{
+							Cell: "zone1",
+							Uid:  102,
+						},
+						Type: topodatapb.TabletType_REPLICA,
+					},
+				},
+			},
+			avoidPrimaryAlias: &topodatapb.TabletAlias{
+				Cell: "zone1",
+				Uid:  0,
+			},
+			expected: &topodatapb.TabletAlias{
+				Cell: "zone1",
+				Uid:  102,
+			},
+		},
 		{
 			name: "only available tablet is AvoidPrimary",
 			tmc: &chooseNewPrimaryTestTMClient{
@@ -731,7 +790,7 @@ zone1-0000000100 is not a replica`,
 		t.Run(tt.name, func(t *testing.T) {
 			t.Parallel()
 
-			actual, err := ElectNewPrimary(ctx, tt.tmc, tt.shardInfo, tt.tabletMap, tt.newPrimaryAlias, tt.avoidPrimaryAlias, time.Millisecond*50, tt.tolerableReplLag, durability, logger)
+			actual, err := ElectNewPrimary(ctx, tt.tmc, tt.shardInfo, tt.tabletMap, tt.newPrimaryAlias, tt.avoidPrimaryAlias, time.Millisecond*50, tt.tolerableReplLag, durability, tt.allowCrossCellPromotion, logger)
 			if len(tt.errContains) > 0 {
 				for _, errC := range tt.errContains {
 					assert.ErrorContains(t, err, errC)

proto/vtctldata.proto

@@ -1357,6 +1357,8 @@ message PlannedReparentShardRequest {
   // acceptable for a tablet to be eligible for promotion when Vitess makes the choice of a new primary.
   // A value of 0 indicates that Vitess shouldn't consider the replication lag at all.
   vttime.Duration tolerable_replication_lag = 6;
+  // AllowCrossCellPromotion allows cross cell promotion,
+  bool allow_cross_cell_promotion = 7;
 }
 
 message PlannedReparentShardResponse {