Merge pull request #91522 from cockroachdb/blathers/backport-release-…

…22.2-91163

release-22.2: workload/schemachange: fixes to improve stability of the test

pkg/workload/schemachange/operation_generator.go

@@ -246,7 +246,7 @@ func init() {
 var opWeights = []int{
 	addColumn:               1,
 	addConstraint:           0, // TODO(spaskob): unimplemented
-	addForeignKeyConstraint: 1,
+	addForeignKeyConstraint: 0, // Disabled and tracked with #91195
 	addRegion:               1,
 	addUniqueConstraint:     0,
 	alterTableLocality:      1,
@@ -2992,41 +2992,23 @@ func (og *operationGenerator) randParentColumnForFkRelation(
 	var typName string
 	var nullable string
 
-	for {
-		nestedTxn, err := tx.Begin(ctx)
-		if err != nil {
-			return nil, nil, err
-		}
-		err = nestedTxn.QueryRow(ctx, fmt.Sprintf(`
+	nestedTxn, err := tx.Begin(ctx)
+	if err != nil {
+		return nil, nil, err
+	}
+	err = nestedTxn.QueryRow(ctx, fmt.Sprintf(`
 	SELECT table_schema, table_name, column_name, crdb_sql_type, is_nullable FROM (
 		%s
 	)`, subQuery.String())).Scan(&tableSchema, &tableName, &columnName, &typName, &nullable)
-		if err == nil {
-			err := nestedTxn.Commit(ctx)
-			if err != nil {
-				return nil, nil, err
-			}
-			break
-		}
-		pgErr := (*pgconn.PgError)(nil)
-		if !errors.As(err, &pgErr) {
-			return nil, nil, err
-		}
-		// Intermittent undefined table errors are valid for the query above, since
-		// we are not leasing out the tables, when picking our random table.
-		if code := pgcode.MakeCode(pgErr.Code); code == pgcode.UndefinedTable ||
-			code == pgcode.UndefinedSchema {
-			err := nestedTxn.Rollback(ctx)
-			if err != nil {
-				return nil, nil, err
-			}
-			continue
-		}
+	if err != nil {
 		if rbErr := nestedTxn.Rollback(ctx); rbErr != nil {
 			err = errors.CombineErrors(err, rbErr)
 		}
 		return nil, nil, err
 	}
+	if err = nestedTxn.Commit(ctx); err != nil {
+		return nil, nil, err
+	}
 
 	columnToReturn := column{
 		name:     columnName,
@@ -3037,7 +3019,6 @@ func (og *operationGenerator) randParentColumnForFkRelation(
 		ExplicitSchema: true,
 	}, tree.Name(tableName))
 
-	var err error
 	columnToReturn.typ, err = og.typeFromTypeName(ctx, tx, typName)
 	if err != nil {
 		return nil, nil, err
@@ -3566,6 +3547,10 @@ func (og *operationGenerator) selectStmt(ctx context.Context, tx pgx.Tx) (stmt *
 	// TODO(fqazi): Temporarily allow out of memory errors on select queries. Not
 	// sure where we are hitting these, need to investigate further.
 	stmt.potentialExecErrors.add(pgcode.OutOfMemory)
+	// Disk errors can happen since there are limits on spill, and cross
+	// joins which are deep cannot do much of the FETCH FIRST X ROWS ONLY
+	// limit
+	stmt.potentialExecErrors.add(pgcode.DiskFull)
 	return stmt, nil
 }
 

pkg/workload/schemachange/watch_dog.go

@@ -12,6 +12,7 @@ package schemachange
 
 import (
 	"context"
+	gosql "database/sql"
 	"fmt"
 	"time"
 
@@ -31,9 +32,9 @@ type schemaChangeWatchDog struct {
 	// cmdChannel used to communicate from thread executing commands on the session.
 	cmdChannel chan chan struct{}
 	// activeQuery last active query observed on the monitored connection.
-	activeQuery string
+	activeQuery gosql.NullString
 	// txnID last active transaction ID observed on the target connection.
-	txnID string
+	txnID gosql.NullString
 	// numRetries number of transaction retries observed.
 	numRetries int
 }
@@ -137,8 +138,8 @@ func (w *schemaChangeWatchDog) Start(ctx context.Context, tx pgx.Tx) error {
 // reset prepares the watch dog for re-use.
 func (w *schemaChangeWatchDog) reset() {
 	w.sessionID = ""
-	w.activeQuery = ""
-	w.txnID = ""
+	w.activeQuery = gosql.NullString{}
+	w.txnID = gosql.NullString{}
 }
 
 // Stop stops monitoring the connection and waits for the watch dog thread to