GetSchema: limit concurrent operations

go/vt/mysqlctl/schema.go

@@ -22,21 +22,29 @@ import (
 	"regexp"
 	"sort"
 	"strings"
-	"sync"
+
+	"golang.org/x/sync/errgroup"
 
 	"vitess.io/vitess/go/mysql"
+	"vitess.io/vitess/go/sqlescape"
 	"vitess.io/vitess/go/sqltypes"
 	"vitess.io/vitess/go/vt/concurrency"
-	"vitess.io/vitess/go/vt/vterrors"
-	"vitess.io/vitess/go/vt/vtgate/evalengine"
-
-	"vitess.io/vitess/go/sqlescape"
 	"vitess.io/vitess/go/vt/log"
 	"vitess.io/vitess/go/vt/mysqlctl/tmutils"
-
 	querypb "vitess.io/vitess/go/vt/proto/query"
 	tabletmanagerdatapb "vitess.io/vitess/go/vt/proto/tabletmanagerdata"
 	"vitess.io/vitess/go/vt/proto/vtrpc"
+	"vitess.io/vitess/go/vt/vterrors"
+	"vitess.io/vitess/go/vt/vtgate/evalengine"
+)
+
+const (
+	// In a local environment and without latency, we have seen that an unbounded concurrency still translates to less than
+	// 20 concurrent MySQL connections. Which is why placing a limit of 20 concurrent goroutines (each mapped to a MySQL connection)
+	// is unlikely to affect optimal environments.
+	// In high latency environments, unbounded concurrency can translate to a very high number of concurrent MySQL connections. This
+	// is an undesirable behavior. We prefer to push back on GetSchema and make it run over longer time, instead.
+	getSchemaConcurrency = 20
 )
 
 var autoIncr = regexp.MustCompile(` AUTO_INCREMENT=\d+`)
@@ -102,18 +110,18 @@ func (mysqld *Mysqld) GetSchema(ctx context.Context, dbName string, request *tab
 	ctx, cancel := context.WithCancel(ctx)
 	defer cancel()
 
-	var wg sync.WaitGroup
 	allErrors := &concurrency.AllErrorRecorder{}
 
+	eg, ctx := errgroup.WithContext(ctx)
+	eg.SetLimit(getSchemaConcurrency)
+
 	// Get per-table schema concurrently.
 	tableNames := make([]string, 0, len(tds))
 	for _, td := range tds {
 		tableNames = append(tableNames, td.Name)
+		td := td
 
-		wg.Add(1)
-		go func(td *tabletmanagerdatapb.TableDefinition) {
-			defer wg.Done()
-
+		eg.Go(func() error {
 			fields, columns, schema, err := mysqld.collectSchema(ctx, dbName, td.Name, td.Type, request.TableSchemaOnly)
 			if err != nil {
 				// There's a possible race condition: it could happen that a table was dropped in between reading
@@ -122,40 +130,39 @@ func (mysqld *Mysqld) GetSchema(ctx context.Context, dbName string, request *tab
 				// This is fine. We identify the situation and keep the table without any fields/columns/key information
 				sqlErr, isSQLErr := mysql.NewSQLErrorFromError(err).(*mysql.SQLError)
 				if isSQLErr && sqlErr != nil && sqlErr.Number() == mysql.ERNoSuchTable {
-					return
+					return nil
 				}
 
 				allErrors.RecordError(err)
 				cancel()
-				return
+				return err
 			}
 
 			td.Fields = fields
 			td.Columns = columns
 			td.Schema = schema
-		}(td)
+			return nil
+		})
 	}
 
+	colMap := map[string][]string{}
 	// Get primary columns concurrently.
 	// The below runs a single query on `INFORMATION_SCHEMA` and does not interact with the actual tables.
 	// It is therefore safe to run even if some tables are dropped in the interim.
-	colMap := map[string][]string{}
-	if len(tableNames) > 0 {
-		wg.Add(1)
-		go func() {
-			defer wg.Done()
-
+	if len(tableNames) > 0 && !request.TableSchemaOnly {
+		eg.Go(func() error {
 			var err error
 			colMap, err = mysqld.getPrimaryKeyColumns(ctx, dbName, tableNames...)
 			if err != nil {
 				allErrors.RecordError(err)
 				cancel()
-				return
+				return err
 			}
-		}()
+			return nil
+		})
 	}
 
-	wg.Wait()
+	eg.Wait()
 	if err := allErrors.AggrError(vterrors.Aggregate); err != nil {
 		return nil, err
 	}