workload/schemachange: add INSPECT operation to random schema workload

Adds a new inspect operation to the schema change workload, enabling random
generation of INSPECT TABLE and INSPECT DATABASE statements.

Features:
- Support for TABLE/DB targets, AS OF SYSTEM TIME
- Always runs in DETACHED mode so that it can be run inside a transaciton
- Results checked post-run via SHOW INSPECT ERRORS

Errors reported in JSON, consistent with existing workload logs

Closes #155483
Epic: CRDB-55075
Release note: none

Author: spilchen

pkg/workload/schemachange/operation_generator.go

@@ -3456,6 +3456,71 @@ func (og *operationGenerator) validate(ctx context.Context, tx pgx.Tx) (*opStmt,
 	return validateStmt, errors.Errorf("Validation FAIL:\n%s", strings.Join(errs, "\n"))
 }
 
+func (og *operationGenerator) inspect(ctx context.Context, tx pgx.Tx) (*opStmt, error) {
+	stmt := makeOpStmt(OpStmtDML)
+
+	var sb strings.Builder
+	sb.WriteString("INSPECT ")
+
+	if og.randIntn(2) == 0 {
+		tableName, err := og.randTable(ctx, tx, og.pctExisting(true), "")
+		if err != nil {
+			return nil, err
+		}
+		tableExists, err := og.tableExists(ctx, tx, tableName)
+		if err != nil {
+			return nil, err
+		}
+		sb.WriteString("TABLE ")
+		sb.WriteString(tableName.String())
+		if !tableExists {
+			stmt.expectedExecErrors.add(pgcode.UndefinedTable)
+		}
+	} else {
+		database, err := og.getDatabase(ctx, tx)
+		if err != nil {
+			return nil, err
+		}
+		useExisting := og.randIntn(100) < og.pctExisting(true)
+		var databaseName string
+		if useExisting {
+			databaseName = database
+		} else {
+			databaseName = fmt.Sprintf("inspect_db_%s", og.newUniqueSeqNumSuffix())
+			stmt.expectedExecErrors.add(pgcode.InvalidCatalogName)
+		}
+		sb.WriteString("DATABASE ")
+		sb.WriteString(databaseName)
+	}
+
+	asof := og.randomInspectAsOfClause()
+	sb.WriteString(asof)
+	// If we use an ASOF time with inspect, chances are it will conflict with
+	// the timestamp chosen for the transaction, and return an "inconsistent AS
+	// OF SYSTEM TIME timestamp" error.
+	stmt.potentialExecErrors.addAll(codesWithConditions{
+		{pgcode.FeatureNotSupported, asof != ""},
+	})
+
+	// Always run DETACHED as this allows us to use INSPECT inside of a
+	// transaction. We have post-processing at the end of the run to verify
+	// INSPECT didn't find any issues.
+	sb.WriteString(" WITH OPTIONS DETACHED")
+	stmt.sql = sb.String()
+
+	return stmt, nil
+}
+
+func (og *operationGenerator) randomInspectAsOfClause() string {
+	// Use AS OF SYSTEM TIME infrequently (~10% of the time) because transactions
+	// are never started with AS OF SYSTEM TIME, and INSPECT with AS OF inside a
+	// transaction without AS OF can cause conflicts.
+	if og.randIntn(10) == 0 {
+		return fmt.Sprintf(" AS OF SYSTEM TIME '-%ds'", og.randIntn(30)+1)
+	}
+	return ""
+}
+
 type column struct {
 	name                tree.Name
 	typ                 *types.T

pkg/workload/schemachange/optype.go

@@ -145,6 +145,9 @@ const (
 	dropView     // DROP VIEW <view>
 	truncateTable
 
+	// INSPECT ...
+	inspect // INSPECT {TABLE|DATABASE} ...
+
 	// Unimplemented operations. TODO(sql-foundations): Audit and/or implement these operations.
 	// alterDatabaseOwner
 	// alterDatabasePlacement
@@ -206,14 +209,15 @@ const (
 )
 
 func isDMLOpType(t opType) bool {
-	return t == insertRow || t == selectStmt || t == validate
+	return t == insertRow || t == selectStmt || t == validate || t == inspect
 }
 
 var opFuncs = []func(*operationGenerator, context.Context, pgx.Tx) (*opStmt, error){
 	// Non-DDL
 	insertRow:  (*operationGenerator).insertRow,
 	selectStmt: (*operationGenerator).selectStmt,
 	validate:   (*operationGenerator).validate,
+	inspect:    (*operationGenerator).inspect,
 
 	// DDL Operations
 	alterDatabaseAddRegion:            (*operationGenerator).addRegion,
@@ -273,6 +277,7 @@ var opWeights = []int{
 	insertRow:  10,
 	selectStmt: 10,
 	validate:   2, // validate twice more often
+	inspect:    1,
 
 	// DDL Operations
 	alterDatabaseAddRegion:            1,
@@ -334,6 +339,7 @@ var opDeclarativeVersion = map[opType]clusterversion.Key{
 	insertRow:  clusterversion.MinSupported,
 	selectStmt: clusterversion.MinSupported,
 	validate:   clusterversion.MinSupported,
+	inspect:    clusterversion.MinSupported,
 
 	alterPolicy:                       clusterversion.V25_2,
 	alterTableAddColumn:               clusterversion.MinSupported,

pkg/workload/schemachange/optype_string.go

@@ -163,6 +163,92 @@ func setupSchemaChangePromCounter(reg prometheus.Registerer) schemaChangeCounter
 	}
 }
 
+func (s *schemaChange) logInspectErrors(
+	ctx context.Context, pool *workload.MultiConnPool, log *atomicLog,
+) error {
+	connPool := pool.Get()
+	conn, err := connPool.Acquire(ctx)
+	if err != nil {
+		log.printLn(fmt.Sprintf("unable to acquire connection for SHOW INSPECT ERRORS: %v", err))
+		return err
+	}
+	defer conn.Release()
+
+	rows, err := conn.Query(ctx, `SELECT job_id FROM [SHOW JOBS] WHERE job_type = 'INSPECT' ORDER BY created DESC`)
+	if err != nil {
+		log.printLn(fmt.Sprintf("fetching INSPECT jobs failed: %v", err))
+		return err
+	}
+	jobIDs, err := pgx.CollectRows(rows, pgx.RowTo[int64])
+	rows.Close()
+	if err != nil {
+		log.printLn(fmt.Sprintf("collecting INSPECT job IDs failed: %v", err))
+		return err
+	}
+	if len(jobIDs) == 0 {
+		return nil
+	}
+
+	type InspectJobResult struct {
+		JobID  int64            `json:"jobId"`
+		Status string           `json:"status"`
+		Errors []map[string]any `json:"errors,omitempty"`
+	}
+
+	type InspectErrorSummary struct {
+		Message string             `json:"message"`
+		Jobs    []InspectJobResult `json:"jobs"`
+	}
+
+	summary := InspectErrorSummary{
+		Message: "Inspect Job Errors",
+		Jobs:    make([]InspectJobResult, 0, len(jobIDs)),
+	}
+
+	var totalErrors int
+	for _, jobID := range jobIDs {
+		query := fmt.Sprintf("SHOW INSPECT ERRORS FOR JOB %d WITH DETAILS", jobID)
+		rows, err := conn.Query(ctx, query)
+		if err != nil {
+			log.printLn(fmt.Sprintf("%s failed: %v", query, err))
+			continue
+		}
+		results, err := pgx.CollectRows(rows, pgx.RowToMap)
+		rows.Close()
+		if err != nil {
+			log.printLn(fmt.Sprintf("collecting inspect errors for job %d failed: %v", jobID, err))
+			continue
+		}
+
+		jobResult := InspectJobResult{
+			JobID: jobID,
+		}
+
+		if len(results) == 0 {
+			jobResult.Status = "no errors reported"
+		} else {
+			jobResult.Status = fmt.Sprintf("%d error rows", len(results))
+			jobResult.Errors = results
+			totalErrors += len(results)
+		}
+
+		summary.Jobs = append(summary.Jobs, jobResult)
+	}
+
+	// Output as JSON.
+	jsonBytes, err := json.MarshalIndent(summary, "", " ")
+	if err != nil {
+		log.printLn(fmt.Sprintf("failed to marshal inspect errors to JSON: %v", err))
+		return err
+	}
+	log.printLn(string(jsonBytes))
+
+	if totalErrors > 0 {
+		return errors.Newf("found %d inspect errors across %d jobs", totalErrors, len(jobIDs))
+	}
+	return nil
+}
+
 // Meta implements the workload.Generator interface.
 func (s *schemaChange) Meta() workload.Meta { return schemaChangeMeta }
 
@@ -246,6 +332,8 @@ func (s *schemaChange) Ops(
 
 	ql := workload.QueryLoad{
 		Close: func(_ context.Context) error {
+			inspectErr := s.logInspectErrors(ctx, pool, stdoutLog)
+
 			// Create a new context for shutting down the tracer provider. The
 			// provided context may be cancelled depending on why the workload is
 			// shutting down and we always want to provide a period of time to flush
@@ -259,7 +347,7 @@ func (s *schemaChange) Ops(
 			closeErr := s.closeJSONLogFile()
 			shutdownErr := tracerProvider.Shutdown(ctx)
 			s.schemaWorkloadResultAnnotator.logWorkloadStats(stdoutLog)
-			return errors.CombineErrors(closeErr, shutdownErr)
+			return errors.CombineErrors(inspectErr, errors.CombineErrors(closeErr, shutdownErr))
 		},
 	}
 
@@ -564,6 +652,9 @@ func (w *schemaChangeWorker) run(ctx context.Context) error {
 			return err
 		}
 	}
+	if _, err := conn.Exec(ctx, "SET enable_inspect_command = true;"); err != nil {
+		return err
+	}
 
 	tx, err := conn.Begin(ctx)
 	if err != nil {