VDiff: add --no-samples option to vdiff show

changelog/25.0/25.0.0/summary.md

@@ -19,6 +19,7 @@
 - **[Minor Changes](#minor-changes)**
     - **[VReplication](#minor-changes-vreplication)**
         - [Default data protection for `_reverse` workflow cancel/complete](#vreplication-reverse-workflow-data-protection)
+        - [`vdiff show --only-summary` omits the per-table row-sample report](#vreplication-vdiff-only-summary)
     - **[VTGate](#minor-changes-vtgate)**
         - [Ingress bytes in query LogStats](#vtgate-logstats-ingress-bytes)
         - [New controls for cross-keyspace reads](#vtgate-cross-keyspace-reads)
@@ -170,6 +171,16 @@ The `--keep-data` flag help text has been updated to note this default explicitl
 
 See [#19906](https://github.com/vitessio/vitess/pull/19906) for details.
 
+#### <a id="vreplication-vdiff-only-summary"/>`vdiff show --only-summary` omits the per-table row-sample report</a>
+
+`vtctldclient vdiff ... show` now accepts an `--only-summary` flag. When set, the per-table diff report has its sampled-row arrays (`MismatchedRowsSample`, `ExtraRowsSourceSample`, `ExtraRowsTargetSample`) stripped on the tablet before the response is built, while the scalar counters (processed, matching, mismatched, and extra rows) and all other summary fields are preserved.
+
+The sampled-row arrays carry actual row data, including large `BLOB`/`JSON` columns, and `vdiff show` aggregates them across every target shard. For diffs over tables with large rows this could push the aggregated response past gRPC message limits and make `vdiff show` fail outright, leaving no way to read the summary or mismatch state. `--only-summary` lets callers that only need progress and the mismatch state avoid transferring the samples while keeping the reported counts accurate.
+
+The option is exposed as `only_summary` on the `VDiffShowRequest` (vtctld) and `VDiffReportOptions` (tablet) protobuf messages. It is opt-in and backward compatible: without the flag, the full report is returned as before.
+
+See [#20870](https://github.com/vitessio/vitess/pull/20870) for details.
+
 ### <a id="minor-changes-vtgate"/>VTGate</a>
 
 #### <a id="vtgate-logstats-ingress-bytes"/>Ingress bytes in query LogStats</a>

go/cmd/vtctldclient/command/vreplication/vdiff/vdiff.go

@@ -82,8 +82,9 @@ var (
 	}{}
 
 	showOptions = struct {
-		Arg     string
-		Verbose bool
+		Arg         string
+		Verbose     bool
+		OnlySummary bool
 	}{}
 
 	stopOptions = struct {
@@ -645,6 +646,7 @@ func commandShow(cmd *cobra.Command, args []string) error {
 		Workflow:       common.BaseOptions.Workflow,
 		TargetKeyspace: common.BaseOptions.TargetKeyspace,
 		Arg:            showOptions.Arg,
+		OnlySummary:    showOptions.OnlySummary,
 	})
 	if err != nil {
 		return err
@@ -709,6 +711,7 @@ func registerCommands(root *cobra.Command) {
 	base.AddCommand(resume)
 
 	show.Flags().BoolVar(&showOptions.Verbose, "verbose", false, "Show verbose output in summaries")
+	show.Flags().BoolVar(&showOptions.OnlySummary, "only-summary", false, "Omit the per-table diff report body and return only the vdiff and per-table summary state. Useful for large diffs where the report can exceed gRPC message limits.")
 	base.AddCommand(show)
 
 	stop.Flags().StringSliceVar(&stopOptions.TargetShards, "target-shards", nil, "The target shards to stop the vdiff on; default is all shards.")

go/vt/vtctl/workflow/vdiff.go

@@ -566,6 +566,11 @@ func (s *Server) VDiffShow(ctx context.Context, req *vtctldatapb.VDiffShowReques
 		Workflow:  req.Workflow,
 		Action:    string(vdiff.ShowAction),
 		ActionArg: req.Arg,
+		Options: &tabletmanagerdatapb.VDiffOptions{
+			ReportOptions: &tabletmanagerdatapb.VDiffReportOptions{
+				OnlySummary: req.GetOnlySummary(),
+			},
+		},
 	}
 
 	ts, err := s.buildTrafficSwitcher(ctx, req.TargetKeyspace, req.Workflow)

go/vt/vttablet/tabletmanager/vdiff/action.go

@@ -115,11 +115,27 @@ func (vde *Engine) PerformVDiffAction(ctx context.Context, req *tabletmanagerdat
 	return resp, nil
 }
 
-func (vde *Engine) getVDiffSummary(vdiffID int64, dbClient binlogplayer.DBClient) (*query.QueryResult, error) {
+// vdiffSummaryQuery returns the summary query to run. When onlySummary is true
+// it returns the variant that strips the row-sample arrays from the per-table
+// report, keeping the scalar counters. The samples are the part that can grow
+// very large (they carry sampled row data, including large blob/JSON columns)
+// and, fanned out across many target shards, can push the aggregated response
+// past gRPC message limits. Stripping them lets callers that only need the
+// vdiff/table state, counts and has_mismatch avoid transferring that data while
+// keeping the summary counters accurate. All other summary columns are
+// unaffected.
+func vdiffSummaryQuery(onlySummary bool) string {
+	if onlySummary {
+		return sqlVDiffSummaryOnly
+	}
+	return sqlVDiffSummary
+}
+
+func (vde *Engine) getVDiffSummary(vdiffID int64, dbClient binlogplayer.DBClient, reportOpts *tabletmanagerdatapb.VDiffReportOptions) (*query.QueryResult, error) {
 	var qr *sqltypes.Result
 	var err error
 
-	query, err := sqlparser.ParseAndBind(sqlVDiffSummary, sqltypes.Int64BindVariable(vdiffID), sqltypes.StringBindVariable(vde.dbName))
+	query, err := sqlparser.ParseAndBind(vdiffSummaryQuery(reportOpts.GetOnlySummary()), sqltypes.Int64BindVariable(vdiffID), sqltypes.StringBindVariable(vde.dbName))
 	if err != nil {
 		return nil, err
 	}
@@ -336,7 +352,7 @@ func (vde *Engine) handleShowAction(ctx context.Context, dbClient binlogplayer.D
 		case 1:
 			row := qr.Named().Row()
 			vdiffID, _ := row["id"].ToInt64()
-			summary, err := vde.getVDiffSummary(vdiffID, dbClient)
+			summary, err := vde.getVDiffSummary(vdiffID, dbClient, req.GetOptions().GetReportOptions())
 			resp.Output = summary
 			if err != nil {
 				return err

go/vt/vttablet/tabletmanager/vdiff/action_test.go

@@ -19,6 +19,7 @@ package vdiff
 import (
 	"context"
 	"fmt"
+	"strings"
 	"testing"
 	"time"
 
@@ -34,6 +35,47 @@ import (
 	vtrpcpb "vitess.io/vitess/go/vt/proto/vtrpc"
 )
 
+// TestVDiffSummaryQuery asserts that the two summary-query variants differ only
+// in how they project the report column: the full variant returns the stored
+// report as-is, while the summary-only variant strips just the row-sample
+// arrays and keeps the scalar counters. The assertions are structural (split on
+// the report select-expression and compare the surrounding column list and
+// FROM/WHERE) so harmless SQL formatting changes don't break the test.
+func TestVDiffSummaryQuery(t *testing.T) {
+	const (
+		fullReportExpr    = `vdt.report as report`
+		summaryReportExpr = `JSON_REMOVE(vdt.report, '$.MismatchedRowsSample', '$.ExtraRowsSourceSample', '$.ExtraRowsTargetSample') as report`
+	)
+
+	full := vdiffSummaryQuery(false)
+	summary := vdiffSummaryQuery(true)
+
+	// Both must be bindable (report/columns aside, the FROM/WHERE has the %a
+	// placeholders that ParseAndBind fills).
+	require.Contains(t, full, "%a", "bind placeholders must be preserved for ParseAndBind")
+	require.Contains(t, summary, "%a", "bind placeholders must be preserved for ParseAndBind")
+
+	// Each variant uses its own report select-expression and not the other's.
+	require.Contains(t, full, fullReportExpr, "full query must select the stored report as-is")
+	require.NotContains(t, full, "JSON_REMOVE", "full query must not strip the report")
+	require.Contains(t, summary, summaryReportExpr, "summary-only query must strip the row-sample arrays from the report")
+
+	// The summary-only variant must strip the sample arrays but must not touch
+	// the scalar counters, so the summary counts stay accurate.
+	for _, sample := range []string{"MismatchedRowsSample", "ExtraRowsSourceSample", "ExtraRowsTargetSample"} {
+		require.Contains(t, summaryReportExpr, sample, "expected sample array %q to be stripped", sample)
+	}
+
+	// The two variants must be identical everywhere except the report
+	// select-expression: swapping in the full expression must reproduce the
+	// full query exactly, so no other column or clause can silently differ.
+	require.Equal(t,
+		full,
+		strings.Replace(summary, summaryReportExpr, fullReportExpr, 1),
+		"summary-only must differ from the full query only in the report select-expression",
+	)
+}
+
 func TestPerformVDiffAction(t *testing.T) {
 	ctx, cancel := context.WithTimeout(t.Context(), 30*time.Second)
 	defer cancel()