Merge #150978 #151500 #151758 #151768 #151787

150978: roachtest: add logging to jobs/stress r=kev-cao a=msbutler

job/stress recently timed out, and I couldn't figure out why. This patch adds a bit more logging for the next failure, and removes some deprecated machinery.

Informs #150801

Release note: none

151500: sql: make DeleteRange chunk size metamorphic r=yuzefovich a=yuzefovich

Previously, we used the hard-coded value of 600 for `row.TableTruncateChunkSize` which originally dictated the usage of DelRange KV requests when truncating the table and over time the constant's usage has been expanded to other places where DelRange was used. In particular, we have a fast-path deleteRange operator that uses it, and this commit makes the constant there metamorphic.

Additionally, this commit separates out the constants that were unified in 3d19b62 so that the index deletion code (that also uses DelRange requests) has its own, and in tests (which seem to have rotted a bit over time), we now hard-code the previous value.

Informs: #151294.
Epic: None
Release note: None

151758: roachtest: route schema workload flakes in pcr tests to foundations r=kev-cao a=msbutler

Informs #150548
Informs #151645

Release note: none

151768: tests,sqlsmith: skip st_snap in TestRandomSyntaxSQLSmith r=yuzefovich a=yuzefovich

There is a known issue with `st_snap` that causes a crash in geos, so we avoid generating a stmt with this builtin for now.

Additionally, I noticed some random stmts that always will hit an error because ORDER BY clause contains a type that doesn't define an ordering, so this commit also adjusts the sqlsmith generation for REFCURSOR, JSONPATH, and PGVECTOR (to match `ensureColumnOrderable`).

Informs: #151103
Fixes: #151580

Release note: None

151787: fingerprintutils: select database and tables with aost if specified r=kev-cao a=msbutler

Previously, if a user wanted to fingerprint a database(s) with an AOST, the utility mistakenly selected the database/tables without an aost. This patch fixes this.

Fixes #150548

Release note: none

Co-authored-by: Michael Butler <butler@cockroachlabs.com>
Co-authored-by: Yahor Yuzefovich <yahor@cockroachlabs.com>

Author: 3 people

pkg/cmd/roachtest/tests/backup_restore_roundtrip.go

@@ -45,6 +45,16 @@ var (
 	}
 )
 
+func handleSchemaChangeWorkloadError(err error) error {
+	// If the UNEXPECTED ERROR detail appears, the workload likely flaked.
+	// Otherwise, the workload could have failed due to other reasons like a node
+	// crash.
+	if err != nil && strings.Contains(errors.FlattenDetails(err), "UNEXPECTED ERROR") {
+		return registry.ErrorWithOwner(registry.OwnerSQLFoundations, errors.Wrapf(err, "schema change workload failed"))
+	}
+	return err
+}
+
 const numFullBackups = 3
 
 type roundTripSpecs struct {
@@ -261,18 +271,9 @@ func startBackgroundWorkloads(
 		return nil, err
 	}
 
-	handleChemaChangeError := func(err error) error {
-		// If the UNEXPECTED ERROR detail appears, the workload likely flaked.
-		// Otherwise, the workload could have failed due to other reasons like a node
-		// crash.
-		if err != nil && strings.Contains(errors.FlattenDetails(err), "UNEXPECTED ERROR") {
-			return registry.ErrorWithOwner(registry.OwnerSQLFoundations, errors.Wrapf(err, "schema change workload failed"))
-		}
-		return err
-	}
 	err = c.RunE(ctx, option.WithNodes(workloadNode), scInit.String())
 	if err != nil {
-		return nil, handleChemaChangeError(err)
+		return nil, handleSchemaChangeWorkloadError(err)
 	}
 
 	run := func() (func(), error) {
@@ -289,7 +290,7 @@ func startBackgroundWorkloads(
 		})
 		stopSC := workloadWithCancel(m, func(ctx context.Context) error {
 			if err := c.RunE(ctx, option.WithNodes(workloadNode), scRun.String()); err != nil {
-				return handleChemaChangeError(err)
+				return handleSchemaChangeWorkloadError(err)
 			}
 			return nil
 		})

pkg/cmd/roachtest/tests/cluster_to_cluster.go

@@ -1012,7 +1012,7 @@ func (rd *replicationDriver) maybeRunSchemaChangeWorkload(
 			if err != nil && ctx.Err() == nil {
 				// Implies the workload context was not cancelled and the workload cmd returned a
 				// different error.
-				return errors.Wrapf(err, `schema change workload context was not cancelled. Error returned by workload cmd`)
+				return errors.Wrapf(handleSchemaChangeWorkloadError(err), `schema change workload context was not cancelled. Error returned by workload cmd`)
 			}
 			return nil
 		})

pkg/cmd/roachtest/tests/jobs.go

@@ -67,7 +67,6 @@ func runJobsStress(ctx context.Context, t test.Test, c cluster.Cluster) {
 		showJobsTimeout = 30 * time.Second
 		pollerMinFrequencySeconds = 5
 		workloadDuration = time.Minute * 5
-
 	}
 	sqlDB := sqlutils.MakeSQLRunner(conn)
 	sqlDB.Exec(t, "CREATE DATABASE d")
@@ -83,22 +82,24 @@ func runJobsStress(ctx context.Context, t test.Test, c cluster.Cluster) {
 
 	done := make(chan struct{})
 	earlyExit := make(chan struct{}, 1)
-	m := c.NewDeprecatedMonitor(ctx)
+	group := t.NewErrorGroup()
 
-	m.Go(func(ctx context.Context) error {
+	group.Go(func(ctx context.Context, l *logger.Logger) error {
 		defer close(done)
 		var testTimer timeutil.Timer
 		testTimer.Reset(workloadDuration)
 		select {
 		case <-earlyExit:
+			l.Printf("Exiting early")
 		case <-testTimer.C:
+			l.Printf("workload duration of %s elapsed", workloadDuration)
 		}
 		return nil
 	})
 
-	randomPoller := func(f func(ctx context.Context, t test.Test, c cluster.Cluster, rng *rand.Rand) error) func(ctx context.Context) error {
+	randomPoller := func(f func(ctx context.Context, t test.Test, c cluster.Cluster, rng *rand.Rand) error) func(ctx context.Context, _ *logger.Logger) error {
 
-		return func(ctx context.Context) error {
+		return func(ctx context.Context, _ *logger.Logger) error {
 			var pTimer timeutil.Timer
 			defer pTimer.Stop()
 			for {
@@ -111,6 +112,7 @@ func runJobsStress(ctx context.Context, t test.Test, c cluster.Cluster) {
 					return nil
 				case <-pTimer.C:
 					if err := f(ctx, t, c, rng); err != nil {
+						t.L().Printf("Error running periodic function: %s", err)
 						earlyExit <- struct{}{}
 						return err
 					}
@@ -119,16 +121,19 @@ func runJobsStress(ctx context.Context, t test.Test, c cluster.Cluster) {
 		}
 	}
 
-	m.Go(randomPoller(checkJobQueryLatency))
+	group.Go(randomPoller(checkJobQueryLatency))
 
-	m.Go(randomPoller(pauseResumeChangefeeds))
+	group.Go(randomPoller(pauseResumeChangefeeds))
 
-	createTablesWithChangefeeds(ctx, t, c, rng)
+	group.Go(func(ctx context.Context, _ *logger.Logger) error {
+		createTablesWithChangefeeds(ctx, t, c, rng)
+		return nil
+	})
 
 	// TODO(msbutler): consider adding a schema change workload to the existing
 	// tables to further stress the job system.
 
-	m.Wait()
+	require.NoError(t, group.WaitE())
 	checkJobSystemHealth(ctx, t, c, rng)
 }
 
@@ -150,7 +155,7 @@ func createTablesWithChangefeeds(
 		tableName := tableNamePrefix + fmt.Sprintf("%d", i)
 		sqlDB.Exec(t, fmt.Sprintf(`CREATE TABLE %s %s`, tableName, tableSchema))
 		sqlDB.Exec(t, fmt.Sprintf(`INSERT INTO %s VALUES (1, 'x'),(2,'y')`, tableName))
-		sqlDB.Exec(t, fmt.Sprintf("CREATE CHANGEFEED FOR %s INTO 'null://' WITH gc_protect_expires_after='5m', protect_data_from_gc_on_pause", tableName))
+		sqlDB.Exec(t, fmt.Sprintf("CREATE CHANGEFEED FOR %s INTO 'null://' WITH gc_protect_expires_after='2m', protect_data_from_gc_on_pause", tableName))
 		if i%(tableCount/5) == 0 {
 			t.L().Printf("Created %d tables so far", i)
 		}

pkg/internal/sqlsmith/relational.go

@@ -1790,12 +1790,19 @@ func (s *Smither) makeHaving(refs colRefs) *tree.Where {
 }
 
 func (s *Smither) isOrderable(typ *types.T) bool {
+	if typ.Family() == types.ArrayFamily {
+		typ = typ.ArrayContents()
+	}
+	if typ.Family() == types.RefCursorFamily || typ.Family() == types.JsonpathFamily {
+		// These types don't define an ordering function in PG.
+		return false
+	}
 	if s.postgres {
 		// PostGIS cannot order box2d types.
 		return typ.Family() != types.Box2DFamily
 	}
 	switch typ.Family() {
-	case types.TSQueryFamily, types.TSVectorFamily:
+	case types.TSQueryFamily, types.TSVectorFamily, types.PGVectorFamily:
 		// We can't order by these types - see #92165.
 		return false
 	default:

pkg/sql/backfill.go

@@ -57,7 +57,7 @@ const (
 	// per chunk during an index truncation. This value is larger than the
 	// other chunk constants because the operation involves only running a
 	// DeleteRange().
-	indexTruncateChunkSize = row.TableTruncateChunkSize
+	indexTruncateChunkSize = 600
 
 	// indexTxnBackfillChunkSize is the maximum number index entries backfilled
 	// per chunk during an index backfill done in a txn. The index backfill

pkg/sql/delete_range.go

@@ -120,7 +120,7 @@ func (d *deleteRangeNode) startExec(params runParams) error {
 		// hits the key limit).
 		for len(spans) != 0 {
 			b := params.p.txn.NewBatch()
-			b.Header.MaxSpanRequestKeys = row.TableTruncateChunkSize
+			b.Header.MaxSpanRequestKeys = int64(row.DeleteRangeChunkSize(params.extendedEvalCtx.TestingKnobs.ForceProductionValues))
 			b.Header.LockTimeout = params.SessionData().LockTimeout
 			b.Header.DeadlockTimeout = params.SessionData().DeadlockTimeout
 			d.deleteSpans(params, b, spans)

pkg/sql/drop_test.go

@@ -33,7 +33,6 @@ import (
 	"github.com/cockroachdb/cockroach/pkg/sql/catalog/desctestutils"
 	"github.com/cockroachdb/cockroach/pkg/sql/catalog/tabledesc"
 	"github.com/cockroachdb/cockroach/pkg/sql/isql"
-	"github.com/cockroachdb/cockroach/pkg/sql/row"
 	"github.com/cockroachdb/cockroach/pkg/sql/sqltestutils"
 	"github.com/cockroachdb/cockroach/pkg/sql/tests"
 	"github.com/cockroachdb/cockroach/pkg/testutils"
@@ -640,7 +639,8 @@ func TestDropTable(t *testing.T) {
 	s := srv.ApplicationLayer()
 	codec := s.Codec()
 
-	numRows := 2*row.TableTruncateChunkSize + 1
+	const deprecatedTableTruncateChunkSize = 600
+	numRows := 2*deprecatedTableTruncateChunkSize + 1
 	if err := tests.CreateKVTable(sqlDB, "kv", numRows); err != nil {
 		t.Fatal(err)
 	}
@@ -754,7 +754,8 @@ func TestDropTableDeleteData(t *testing.T) {
 	// TTL into the system with AddImmediateGCZoneConfig.
 	defer sqltestutils.DisableGCTTLStrictEnforcement(t, systemDB)()
 
-	const numRows = 2*row.TableTruncateChunkSize + 1
+	const deprecatedTableTruncateChunkSize = 600
+	const numRows = 2*deprecatedTableTruncateChunkSize + 1
 	const numKeys = 3 * numRows
 	const numTables = 5
 	var descs []catalog.TableDescriptor

pkg/sql/opt/exec/execbuilder/mutation.go

@@ -742,7 +742,7 @@ func (b *Builder) buildDeleteRange(del *memo.DeleteExpr) (execPlan, error) {
 		// Mutations only allow auto-commit if there are no FK checks or cascades.
 
 		if maxRows, ok := b.indexConstraintMaxResults(&scan.ScanPrivate, scan.Relational()); ok {
-			if maxKeys := maxRows * uint64(tab.FamilyCount()); maxKeys <= row.TableTruncateChunkSize {
+			if maxKeys := maxRows * uint64(tab.FamilyCount()); maxKeys <= uint64(row.DeleteRangeChunkSize(b.evalCtx.TestingKnobs.ForceProductionValues)) {
 				autoCommit = true
 			}
 		}

pkg/sql/opt/exec/execbuilder/testdata/delete

@@ -1,4 +1,4 @@
-# LogicTest: local
+# LogicTest: local !metamorphic-batch-sizes
 
 statement ok
 CREATE TABLE kv (

pkg/sql/opt/exec/execbuilder/testdata/select_index

@@ -1,4 +1,4 @@
-# LogicTest: local
+# LogicTest: local !metamorphic-batch-sizes
 
 statement ok
 CREATE TABLE t (