Merge pull request #619 from youtube/suguwork

tabletserver: verbose error for killed queries

go/mysql/mysql.go

@@ -50,6 +50,9 @@ const (
 	// ErrOptionPreventsStatement is C.ER_OPTION_PREVENTS_STATEMENT
 	ErrOptionPreventsStatement = C.ER_OPTION_PREVENTS_STATEMENT
 
+	// ErrServerLost is C.CR_SERVER_LOST (2013)
+	ErrServerLost = C.CR_SERVER_LOST
+
 	// RedactedPassword is the password value used in redacted configs
 	RedactedPassword = "****"
 )

go/mysql/vtmysql.h

@@ -3,6 +3,7 @@
 // license that can be found in the LICENSE file.
 
 #include <mysql.h>
+#include <errmsg.h>
 
 // This API provides convenient C wrapper functions for mysql client.
 

go/vt/mysqlctl/slave_connection.go

@@ -10,6 +10,7 @@ import (
 	"fmt"
 
 	log "github.com/golang/glog"
+	"github.com/youtube/vitess/go/mysql"
 	"github.com/youtube/vitess/go/pools"
 	"github.com/youtube/vitess/go/sqldb"
 	"github.com/youtube/vitess/go/sync2"
@@ -105,8 +106,8 @@ func (sc *SlaveConnection) StartBinlogDump(startPos proto.ReplicationPosition) (
 
 			buf, err = sc.Conn.ReadPacket()
 			if err != nil {
-				if sqlErr, ok := err.(*sqldb.SqlError); ok && sqlErr.Number() == 2013 {
-					// errno 2013 = Lost connection to MySQL server during query
+				if sqlErr, ok := err.(*sqldb.SqlError); ok && sqlErr.Number() == mysql.ErrServerLost {
+					// ErrServerLost = Lost connection to MySQL server during query
 					// This is not necessarily an error. It could just be that we closed
 					// the connection from outside.
 					log.Infof("connection closed during binlog stream (possibly intentional): %v", err)

go/vt/tabletserver/connpool_test.go

@@ -78,7 +78,7 @@ func TestConnectivity(t *testing.T) {
 		newctx, cancel = withTimeout(ctx, 2*time.Millisecond)
 		_, err = conn.Exec(newctx, "select sleep(1) from dual", 1000, true)
 		cancel()
-		lostConn := "error: Lost connection to MySQL server during query (errno 2013) during query: select sleep(1) from dual"
+		lostConn := "error: the query was killed either because it timed out or was canceled: Lost connection to MySQL server during query (errno 2013) during query: select sleep(1) from dual"
 		if err == nil || err.Error() != lostConn {
 			t.Errorf("got: %v, want %s", err, lostConn)
 		}

go/vt/tabletserver/tablet_error.go

@@ -111,8 +111,8 @@ func IsConnErr(err error) bool {
 			return false
 		}
 	}
-	// 2013 means that someone sniped the query.
-	if sqlError == 2013 {
+	// ErrServerLost means that someone sniped the query.
+	if sqlError == mysql.ErrServerLost {
 		return false
 	}
 	return sqlError >= 2000 && sqlError <= 2018
@@ -135,6 +135,10 @@ func (te *TabletError) Prefix() string {
 	case ErrNotInTx:
 		prefix = "not_in_tx: "
 	}
+	// Special case for killed queries.
+	if te.SqlError == mysql.ErrServerLost {
+		prefix = prefix + "the query was killed either because it timed out or was canceled: "
+	}
 	return prefix
 }
 

go/vt/tabletserver/tablet_error_test.go

@@ -46,10 +46,14 @@ func TestTabletErrorConnError(t *testing.T) {
 	if !IsConnErr(tabletErr) {
 		t.Fatalf("table error: %v is a connection error", tabletErr)
 	}
-	tabletErr = NewTabletErrorSql(ErrFatal, sqldb.NewSqlError(2013, "test"))
+	tabletErr = NewTabletErrorSql(ErrFatal, sqldb.NewSqlError(mysql.ErrServerLost, "test"))
 	if IsConnErr(tabletErr) {
 		t.Fatalf("table error: %v is not a connection error", tabletErr)
 	}
+	want := "fatal: the query was killed either because it timed out or was canceled: test (errno 2013)"
+	if tabletErr.Error() != want {
+		t.Fatalf("tablet error: %v, want %s", tabletErr, want)
+	}
 	sqlErr := sqldb.NewSqlError(1998, "test")
 	if IsConnErr(sqlErr) {
 		t.Fatalf("sql error: %v is not a connection error", sqlErr)

test/queryservice_tests/nocache_tests.py

@@ -367,7 +367,7 @@ class TestNocache(framework.TestCase):
       conn.begin()
       cu.execute("select sleep(0.5) from vtocc_test", {})
     except dbexceptions.DatabaseError as e:
-      if "error: Query" not in str(e) and "error: Lost connection" not in str(e):
+      if "error: Query" not in str(e) and "error: the query was killed" not in str(e):
         self.fail("Query not killed as expected")
     else:
       self.fail("Did not receive exception")

test/queryservice_tests/stream_tests.py

@@ -112,7 +112,7 @@ class TestStream(framework.TestCase):
       thd.join()
       with self.assertRaises(dbexceptions.DatabaseError) as cm:
         cu.fetchall()
-      errMsg1 = "error: Lost connection to MySQL server during query (errno 2013)"
+      errMsg1 = "error: the query was killed either because it timed out or was canceled: Lost connectioy to MySQL server during query (errno 2013)"
       errMsg2 = "error: Query execution was interrupted (errno 1317)"
       self.assertTrue(cm.exception not in (errMsg1, errMsg2), "did not raise interruption error: %s" % str(cm.exception))
       cu.close()