Merge trift2.:/MySQL/M51/mysql-5.1

into  trift2.:/MySQL/M51/push-5.1

BUILD/check-cpu

@@ -105,6 +105,12 @@ check_cpu () {
     *Athlon*64*)
       cpu_arg="athlon64";
       ;;
+    *Turion*)
+      cpu_arg="athlon64";
+      ;;
+    *Opteron*)
+      cpu_arg="athlon64";
+      ;;
     *Athlon*)
       cpu_arg="athlon";
       ;;

Makefile.am

@@ -136,8 +136,10 @@ test-bt:
 	@PERL@ ./mysql-test-run.pl --force --comment=rpl --suite=rpl
 	-cd mysql-test ; MTR_BUILD_THREAD=auto \
 	@PERL@ ./mysql-test-run.pl --force --comment=partitions --suite=parts
-	-cd mysql-test ; MTR_BUILD_THREAD=auto \
+
-	@PERL@ ./mysql-test-run.pl --force --comment=rowlock --suite=row_lock
+# Re-enable the "rowlock" suite when bug#28685 is fixed
+#	-cd mysql-test ; MTR_BUILD_THREAD=auto \
+#	@PERL@ ./mysql-test-run.pl --force --comment=rowlock --suite=row_lock
 
 # Re-enable the "jp" suite when bug#28563 is fixed
 #	-cd mysql-test ; MTR_BUILD_THREAD=auto \

client/mysqlslap.c

@@ -35,7 +35,8 @@
   Supply your own create and query SQL statements, with 50 clients 
   querying (200 selects for each):
 
-    mysqlslap --create="CREATE TABLE A (a int);INSERT INTO A (23)" \
+    mysqlslap --delimiter=";" \
+              --create="CREATE TABLE A (a int);INSERT INTO A VALUES (23)" \
               --query="SELECT * FROM A" --concurrency=50 --iterations=200
 
   Let the program build the query SQL statement with a table of two int
@@ -554,7 +555,7 @@ static struct my_option my_long_options[] =
     GET_STR, REQUIRED_ARG, 0, 0, 0, 0, 0, 0},
   {"host", 'h', "Connect to host.", (uchar**) &host, (uchar**) &host, 0, GET_STR,
     REQUIRED_ARG, 0, 0, 0, 0, 0, 0},
-  {"iterations", 'i', "Number of times too run the tests.", (uchar**) &iterations,
+  {"iterations", 'i', "Number of times to run the tests.", (uchar**) &iterations,
     (uchar**) &iterations, 0, GET_UINT, REQUIRED_ARG, 1, 0, 0, 0, 0, 0},
   {"number-char-cols", 'x', 
     "Number of VARCHAR columns to create table with if specifying --auto-generate-sql ",

config/ac-macros/compiler_flag.m4

@@ -7,7 +7,7 @@ AC_DEFUN([AC_SYS_COMPILER_FLAG],
   AC_CACHE_VAL(mysql_cv_option_$2,
   [
     CFLAGS="[$]OLD_CFLAGS $1"
-    AC_TRY_RUN([int main(){exit(0);}],mysql_cv_option_$2=yes,mysql_cv_option_$2=no,mysql_cv_option_$2=no)
+    AC_TRY_LINK([int main(){exit(0);}],mysql_cv_option_$2=yes,mysql_cv_option_$2=no,mysql_cv_option_$2=no)
  ])
 
   CFLAGS="[$]OLD_CFLAGS"

config/ac-macros/misc.m4

@@ -450,29 +450,6 @@ AC_DEFINE([HAVE_BOOL], [1], [bool is not defined by all C++ compilators])
 fi
 ])dnl
 
-AC_DEFUN([MYSQL_STACK_DIRECTION],
- [AC_CACHE_CHECK(stack direction for C alloca, ac_cv_c_stack_direction,
- [AC_TRY_RUN([#include <stdlib.h>
- int find_stack_direction ()
- {
-   static char *addr = 0;
-   auto char dummy;
-   if (addr == 0)
-     {
-       addr = &dummy;
-       return find_stack_direction ();
-     }
-   else
-     return (&dummy > addr) ? 1 : -1;
- }
- int main ()
- {
-   exit (find_stack_direction() < 0);
- }], ac_cv_c_stack_direction=1, ac_cv_c_stack_direction=-1,
-   ac_cv_c_stack_direction=0)])
- AC_DEFINE_UNQUOTED(STACK_DIRECTION, $ac_cv_c_stack_direction)
-])dnl
-
 AC_DEFUN([MYSQL_CHECK_LONGLONG_TO_FLOAT],
 [
 AC_MSG_CHECKING(if conversion of longlong to float works)
@@ -488,7 +465,9 @@ int main()
   fprintf(file,"%g\n",f);
   fclose(file);
   return (0);
-}], ac_cv_conv_longlong_to_float=`cat conftestval`, ac_cv_conv_longlong_to_float=0, ifelse([$2], , , ac_cv_conv_longlong_to_float=$2))])dnl
+}], ac_cv_conv_longlong_to_float=`cat conftestval`,
+    ac_cv_conv_longlong_to_float=0,
+    ac_cv_conv_longlong_to_float="yes")])dnl  # Cross compiling, assume can convert
 if test "$ac_cv_conv_longlong_to_float" = "1" -o "$ac_cv_conv_longlong_to_float" = "yes"
 then
   ac_cv_conv_longlong_to_float=yes

configure.in

@@ -230,14 +230,8 @@ AC_CHECK_PROGS(YACC, ['bison -y -p MYSQL'])
 AC_CHECK_PROG(PDFMANUAL, pdftex, manual.pdf)
 AC_CHECK_PROG(DVIS,      tex,    manual.dvi)
 
-AC_MSG_CHECKING("return type of sprintf")
-
 #check the return type of sprintf
-case $SYSTEM_TYPE in
+AC_MSG_CHECKING("return type of sprintf")
-  *netware*)
-    AC_DEFINE(SPRINTF_RETURNS_INT, [1]) AC_MSG_RESULT("int")
-    ;;
-  *)
 AC_TRY_RUN([
   int main()
     {
@@ -263,10 +257,12 @@ AC_TRY_RUN([
                [AC_DEFINE(SPRINTF_RETURNS_PTR, [1], [Broken sprintf])
                 AC_MSG_RESULT("ptr")],
                [AC_DEFINE(SPRINTF_RETURNS_GARBAGE, [1], [Broken sprintf])
-                AC_MSG_RESULT("garbage")])
+                AC_MSG_RESULT("garbage")]
-   ])
+   )],
-    ;;
+   # Cross compile, assume POSIX
-esac
+   [AC_DEFINE(SPRINTF_RETURNS_INT, [1], [POSIX sprintf])
+    AC_MSG_RESULT("int (we assume)")]
+)
 
 AC_PATH_PROG(uname_prog, uname, no)
 
@@ -1667,6 +1663,12 @@ AC_ARG_WITH(client-ldflags,
     [CLIENT_EXTRA_LDFLAGS=])
 AC_SUBST(CLIENT_EXTRA_LDFLAGS)
 
+AC_ARG_WITH(mysqld-libs,
+    [  --with-mysqld-libs   Extra libraries to link with for mysqld],
+    [MYSQLD_EXTRA_LIBS=$withval],
+    [MYSQLD_EXTRA_LIBS=])
+AC_SUBST(MYSQLD_EXTRA_LIBS)
+
 AC_ARG_WITH(lib-ccflags,
     [  --with-lib-ccflags      Extra CC options for libraries],
     [LIB_EXTRA_CCFLAGS=$withval],
@@ -1784,8 +1786,6 @@ MYSQL_TYPE_ACCEPT
 #---END:
 # Figure out what type of struct rlimit to use with setrlimit
 MYSQL_TYPE_STRUCT_RLIMIT
-# Find where the stack goes
-MYSQL_STACK_DIRECTION
 # We want to skip alloca on irix unconditionally. It may work on some version..
 MYSQL_FUNC_ALLOCA
 # Do struct timespec have members tv_sec or ts_sec

include/config-netware.h

@@ -112,9 +112,6 @@ extern "C" {
 /* signal by closing the sockets */
 #define SIGNAL_WITH_VIO_CLOSE 1
 
-/* On NetWare, stack grows towards lower address*/
-#define STACK_DIRECTION -1
-
 /* On NetWare, we need to set stack size for threads, otherwise default 16K is used */
 #define NW_THD_STACKSIZE 65536
 

include/config-win.h

@@ -250,8 +250,6 @@ inline double ulonglong2double(ulonglong value)
 #endif
 
 
-#define STACK_DIRECTION -1
-
 /* Optimized store functions for Intel x86 */
 
 #ifndef _WIN64

mysql-test/Makefile.am

@@ -71,7 +71,7 @@ dist-hook:
 	$(INSTALL_DATA) $(srcdir)/std_data/ndb_backup50/BACKUP* $(distdir)/std_data/ndb_backup50
 	$(INSTALL_DATA) $(srcdir)/std_data/ndb_backup51/BACKUP* $(distdir)/std_data/ndb_backup51
 	$(INSTALL_DATA) $(srcdir)/lib/*.pl $(distdir)/lib
-	-rm -rf `find $(distdir)/suite -type d -name SCCS`
+	-rm -rf `find $(distdir)/suite -type d -name SCCS` $(distdir)/suite/row_lock
 
 install-data-local:
 	$(mkinstalldirs) \
@@ -113,7 +113,7 @@ install-data-local:
 	$(INSTALL_DATA) $(srcdir)/std_data/ndb_backup50/BACKUP* $(DESTDIR)$(testdir)/std_data/ndb_backup50
 	$(INSTALL_DATA) $(srcdir)/std_data/ndb_backup51/BACKUP* $(DESTDIR)$(testdir)/std_data/ndb_backup51
 	$(INSTALL_DATA) $(srcdir)/lib/*.pl $(DESTDIR)$(testdir)/lib
-	for f in `(cd $(srcdir); find suite -type f | grep -v SCCS)`; \
+	for f in `(cd $(srcdir); find suite -type f | egrep -v 'SCCS|row_lock')`; \
 	do \
 	  d=$(DESTDIR)$(testdir)/`dirname $$f`; \
 	  mkdir -p $$d ; \

mysql-test/include/federated.inc

@@ -1,3 +1,4 @@
+--source include/have_log_bin.inc
 --source include/not_embedded.inc
 --source ./include/have_federated_db.inc
 

mysql-test/mysql-test-run.pl

@@ -3815,8 +3815,7 @@ sub mysqld_arguments ($$$$) {
 	      "%s--log-slow-queries=%s-slow.log", $prefix, $log_base_path);
 
   # Check if "extra_opt" contains --skip-log-bin
-  my $skip_binlog= grep(/^--skip-log-bin/, @$extra_opt);
+  my $skip_binlog= grep(/^--skip-log-bin/, @$extra_opt, @opt_extra_mysqld_opt);
-
   if ( $mysqld->{'type'} eq 'master' )
   {
     if (! ($opt_skip_master_binlog || $skip_binlog) )

mysql-test/ndb/ndb_config_1_node.ini

@@ -10,7 +10,8 @@ DataDir= CHOOSE_FILESYSTEM
 MaxNoOfOrderedIndexes= CHOOSE_MaxNoOfOrderedIndexes
 MaxNoOfAttributes= CHOOSE_MaxNoOfAttributes
 TimeBetweenGlobalCheckpoints= 500
-NoOfFragmentLogFiles= 3
+NoOfFragmentLogFiles= 8
+FragmentLogFileSize= 6M
 DiskPageBufferMemory= CHOOSE_DiskPageBufferMemory 
 
 #

mysql-test/ndb/ndb_config_2_node.ini

@@ -10,8 +10,10 @@ DataDir= CHOOSE_FILESYSTEM
 MaxNoOfOrderedIndexes= CHOOSE_MaxNoOfOrderedIndexes
 MaxNoOfAttributes= CHOOSE_MaxNoOfAttributes
 TimeBetweenGlobalCheckpoints= 500
-NoOfFragmentLogFiles= 3
+NoOfFragmentLogFiles= 4
+FragmentLogFileSize=12M
 DiskPageBufferMemory= CHOOSE_DiskPageBufferMemory
+ODirect= 1
 # the following parametes just function as a small regression
 # test that the parameter exists
 InitialNoOfOpenFiles= 27

mysql-test/r/date_formats.result

@@ -1,14 +1,12 @@
 drop table if exists t1;
-SHOW GLOBAL VARIABLES LIKE "%_format%";
+SHOW GLOBAL VARIABLES WHERE Variable_name LIKE "%_format%" AND Variable_name != "binlog_format";
 Variable_name	Value
-binlog_format	<format>
 date_format	%d.%m.%Y
 datetime_format	%Y-%m-%d %H:%i:%s
 default_week_format	0
 time_format	%H.%i.%s
-SHOW SESSION VARIABLES LIKE "%_format%";
+SHOW SESSION VARIABLES WHERE Variable_name LIKE "%_format%" AND Variable_name != "binlog_format";
 Variable_name	Value
-binlog_format	<format>
 date_format	%d.%m.%Y
 datetime_format	%Y-%m-%d %H:%i:%s
 default_week_format	0
@@ -30,9 +28,8 @@ set datetime_format= '%H:%i:%s %Y-%m-%d';
 set datetime_format= '%H:%i:%s.%f %m-%d-%Y';
 set datetime_format= '%h:%i:%s %p %Y-%m-%d';
 set datetime_format= '%h:%i:%s.%f %p %Y-%m-%d';
-SHOW SESSION VARIABLES LIKE "%format";
+SHOW SESSION VARIABLES WHERE Variable_name LIKE "%format" AND Variable_name != "binlog_format";
 Variable_name	Value
-binlog_format	<format>
 date_format	%m-%d-%Y
 datetime_format	%h:%i:%s.%f %p %Y-%m-%d
 default_week_format	0

mysql-test/r/have_log_bin.require

@@ -1,2 +1,2 @@
 Variable_name	Value
-have_log_bin	ON
+log_bin	ON

mysql-test/r/ndb_basic.result

@@ -770,35 +770,6 @@ c	abc	ab
 d	ab	ab
 e	abc	abc
 DROP TABLE t1;
-End of 5.0 tests
-CREATE TABLE t1 (a VARCHAR(255) NOT NULL,
-CONSTRAINT pk_a PRIMARY KEY (a))engine=ndb;
-CREATE TABLE t2(a VARCHAR(255) NOT NULL,
-b VARCHAR(255) NOT NULL,
-c VARCHAR(255) NOT NULL,
-CONSTRAINT pk_b_c_id PRIMARY KEY (b,c),
-CONSTRAINT fk_a FOREIGN KEY(a) REFERENCES t1(a))engine=ndb;
-drop table t1, t2;
-create table t1 (a int not null primary key, b int) engine=ndb;
-insert into t1 values(1,1),(2,2),(3,3);
-create table t2 like t1;
-insert into t2 select * from t1;
-select * from t1 order by a;
-a	b
-1	1
-2	2
-3	3
-select * from t2 order by a;
-a	b
-1	1
-2	2
-3	3
-drop table t1, t2;
-create table t1 (a int not null primary key, b int not null default 0, c varchar(254)) engine=ndb;
-create table if not exists t1 (a int not null primary key, b int not null default 0, c varchar(254)) engine=ndb;
-create table t2 like t1;
-rename table t1 to t10, t2 to t20;
-drop table t10,t20;
 create table t1 (a int not null primary key, b int not null) engine=ndb;
 create table t2 (a int not null primary key, b int not null) engine=ndb;
 insert into t1 values (1,10), (2,20), (3,30);
@@ -867,7 +838,36 @@ select * from t1 order by a;
 a	b
 1	10
 2	10
-3	1
+3	30
 4	1
 drop table t1,t2;
+End of 5.0 tests
+CREATE TABLE t1 (a VARCHAR(255) NOT NULL,
+CONSTRAINT pk_a PRIMARY KEY (a))engine=ndb;
+CREATE TABLE t2(a VARCHAR(255) NOT NULL,
+b VARCHAR(255) NOT NULL,
+c VARCHAR(255) NOT NULL,
+CONSTRAINT pk_b_c_id PRIMARY KEY (b,c),
+CONSTRAINT fk_a FOREIGN KEY(a) REFERENCES t1(a))engine=ndb;
+drop table t1, t2;
+create table t1 (a int not null primary key, b int) engine=ndb;
+insert into t1 values(1,1),(2,2),(3,3);
+create table t2 like t1;
+insert into t2 select * from t1;
+select * from t1 order by a;
+a	b
+1	1
+2	2
+3	3
+select * from t2 order by a;
+a	b
+1	1
+2	2
+3	3
+drop table t1, t2;
+create table t1 (a int not null primary key, b int not null default 0, c varchar(254)) engine=ndb;
+create table if not exists t1 (a int not null primary key, b int not null default 0, c varchar(254)) engine=ndb;
+create table t2 like t1;
+rename table t1 to t10, t2 to t20;
+drop table t10,t20;
 End of 5.1 tests

mysql-test/t/date_formats.test

@@ -6,10 +6,8 @@
 drop table if exists t1;
 --enable_warnings
 
---replace_result ROW <format> STATEMENT <format> MIXED <format>
+SHOW GLOBAL VARIABLES WHERE Variable_name LIKE "%_format%" AND Variable_name != "binlog_format";
-SHOW GLOBAL VARIABLES LIKE "%_format%";
+SHOW SESSION VARIABLES WHERE Variable_name LIKE "%_format%" AND Variable_name != "binlog_format";
---replace_result ROW <format> STATEMENT <format> MIXED <format>
-SHOW SESSION VARIABLES LIKE "%_format%";
 
 #
 # Test setting a lot of different formats to see which formats are accepted and
@@ -36,8 +34,7 @@ set datetime_format= '%H:%i:%s.%f %m-%d-%Y';
 set datetime_format= '%h:%i:%s %p %Y-%m-%d';
 set datetime_format= '%h:%i:%s.%f %p %Y-%m-%d';
 
---replace_result ROW <format> STATEMENT <format> MIXED <format>
+SHOW SESSION VARIABLES WHERE Variable_name LIKE "%format" AND Variable_name != "binlog_format";
-SHOW SESSION VARIABLES LIKE "%format";
 
 --error 1231
 SET time_format='%h:%i:%s';

mysql-test/t/disabled.def

@@ -43,7 +43,6 @@ rpl_ndb_ddl              : BUG#28798 2007-05-31  lars  Valgrind failure in NDB
 
 #rpl_ndb_dd_advance	 : Bug#25913 rpl_ndb_dd_advance fails randomly
 
-rpl_ndb_stm_innodb       : Bug#26783
 ndb_partition_error2	 : HF is not sure if the test can work as internded on all the platforms
 
 im_options_set       : Bug#20294: Instance manager tests fail randomly

mysql-test/t/flush_block_commit_notembedded.test

@@ -3,6 +3,8 @@
 # We verify that we did not introduce a deadlock.
 # This is intended to mimick how mysqldump and innobackup work.
 
+-- source include/have_log_bin.inc
+
 # And it requires InnoDB
 -- source include/not_embedded.inc
 -- source include/have_innodb.inc

mysql-test/t/mysqlbinlog.test

@@ -2,6 +2,8 @@
 # TODO: Need to look at making a row based version once the new row based client is completed. [jbm]
 -- source include/have_binlog_format_mixed_or_statement.inc
 
+-- source include/have_log_bin.inc
+
 # Embedded server doesn't support binlogging
 -- source include/not_embedded.inc
 

mysql-test/t/mysqldump.test

@@ -1,6 +1,9 @@
 # Embedded server doesn't support external clients
 --source include/not_embedded.inc
 
+# Binlog is required
+--source include/have_log_bin.inc
+
 --disable_warnings
 DROP TABLE IF EXISTS t1, `"t"1`, t1aa, t2, t2aa, t3;
 drop database if exists mysqldump_test_db;

mysql-test/t/mysqltest.test

@@ -1,3 +1,5 @@
+-- source include/have_log_bin.inc
+
 # This test should work in embedded server after mysqltest is fixed
 -- source include/not_embedded.inc
 

mysql-test/t/ndb_basic.test

@@ -752,46 +752,6 @@ INSERT INTO t1 VALUES
 SELECT * FROM t1 ORDER BY a;
 DROP TABLE t1;
 
-# End of 5.0 tests
---echo End of 5.0 tests
-
-
-#
-# Bug #18483 Cannot create table with FK constraint
-# ndb does not support foreign key constraint, it is silently ignored
-# in line with other storage engines
-#
-CREATE TABLE t1 (a VARCHAR(255) NOT NULL,
-                 CONSTRAINT pk_a PRIMARY KEY (a))engine=ndb;
-CREATE TABLE t2(a VARCHAR(255) NOT NULL,
-                b VARCHAR(255) NOT NULL,
-                c VARCHAR(255) NOT NULL,
-                CONSTRAINT pk_b_c_id PRIMARY KEY (b,c),
-                CONSTRAINT fk_a FOREIGN KEY(a) REFERENCES t1(a))engine=ndb;
-drop table t1, t2;
-
-# bug#24301
-create table t1 (a int not null primary key, b int) engine=ndb;
-insert into t1 values(1,1),(2,2),(3,3);
-create table t2 like t1;
-insert into t2 select * from t1;
-select * from t1 order by a;
-select * from t2 order by a;
-drop table t1, t2;
-
-# create table if not exists
---disable_warnings
-create table t1 (a int not null primary key, b int not null default 0, c varchar(254)) engine=ndb;
-create table if not exists t1 (a int not null primary key, b int not null default 0, c varchar(254)) engine=ndb;
---enable_warnings
-
-# create like
-create table t2 like t1;
-
-# multi rename
-rename table t1 to t10, t2 to t20;
-drop table t10,t20;
-
 # delete
 create table t1 (a int not null primary key, b int not null) engine=ndb;
 create table t2 (a int not null primary key, b int not null) engine=ndb;
@@ -832,4 +792,44 @@ update ignore t1,t2 set a = 1, c = 1 where a = 3 and c = 3;
 select * from t1 order by a;
 drop table t1,t2;
 
+# End of 5.0 tests
+--echo End of 5.0 tests
+
+
+#
+# Bug #18483 Cannot create table with FK constraint
+# ndb does not support foreign key constraint, it is silently ignored
+# in line with other storage engines
+#
+CREATE TABLE t1 (a VARCHAR(255) NOT NULL,
+                 CONSTRAINT pk_a PRIMARY KEY (a))engine=ndb;
+CREATE TABLE t2(a VARCHAR(255) NOT NULL,
+                b VARCHAR(255) NOT NULL,
+                c VARCHAR(255) NOT NULL,
+                CONSTRAINT pk_b_c_id PRIMARY KEY (b,c),
+                CONSTRAINT fk_a FOREIGN KEY(a) REFERENCES t1(a))engine=ndb;
+drop table t1, t2;
+
+# bug#24301
+create table t1 (a int not null primary key, b int) engine=ndb;
+insert into t1 values(1,1),(2,2),(3,3);
+create table t2 like t1;
+insert into t2 select * from t1;
+select * from t1 order by a;
+select * from t2 order by a;
+drop table t1, t2;
+
+# create table if not exists
+--disable_warnings
+create table t1 (a int not null primary key, b int not null default 0, c varchar(254)) engine=ndb;
+create table if not exists t1 (a int not null primary key, b int not null default 0, c varchar(254)) engine=ndb;
+--enable_warnings
+
+# create like
+create table t2 like t1;
+
+# multi rename
+rename table t1 to t10, t2 to t20;
+drop table t10,t20;
+
 --echo End of 5.1 tests

mysql-test/t/ps.test

@@ -1,4 +1,5 @@
 -- source include/not_embedded.inc
+-- source include/have_log_bin.inc
 #
 # SQL Syntax for Prepared Statements test
 #

mysql-test/t/sp_trans.test

@@ -2,6 +2,7 @@
 # tests that require InnoDB...
 #
 
+-- source include/have_log_bin.inc
 -- source include/have_innodb.inc
 
 --disable_warnings

mysql-test/t/user_var-binlog.test

@@ -3,6 +3,7 @@
 # TODO: Create row based version once $MYSQL_BINLOG has new RB version
 # Embedded server does not support binlogging
 --source include/not_embedded.inc
+--source include/have_log_bin.inc
 
 # Check that user variables are binlogged correctly (BUG#3875)
 create table t1 (a varchar(50));

mysys/mf_keycache.c

@@ -325,6 +325,9 @@ static int keycache_pthread_cond_signal(pthread_cond_t *cond);
 #endif /* defined(KEYCACHE_DEBUG) */
 
 #if !defined(DBUG_OFF)
+#if defined(inline)
+#undef inline
+#endif
 #define inline  /* disabled inline for easier debugging */
 static int fail_block(BLOCK_LINK *block);
 static int fail_hlink(HASH_LINK *hlink);

plugin/daemon_example/Makefile.am

@@ -36,3 +36,6 @@ noinst_LIBRARIES =	@plugin_daemon_example_static_target@
 libdaemon_example_a_CXXFLAGS =	$(AM_CFLAGS)
 libdaemon_example_a_CFLAGS =	$(AM_CFLAGS)
 libdaemon_example_a_SOURCES=	daemon_example.cc
+
+# Don't update the files from bitkeeper
+%::SCCS/s.%

plugin/fulltext/Makefile.am

@@ -22,3 +22,6 @@ pkglib_LTLIBRARIES= mypluglib.la
 mypluglib_la_SOURCES= plugin_example.c
 mypluglib_la_LDFLAGS= -module -rpath $(pkglibdir)
 mypluglib_la_CFLAGS= -DMYSQL_DYNAMIC_PLUGIN
+
+# Don't update the files from bitkeeper
+%::SCCS/s.%

sql/Makefile.am

@@ -38,7 +38,8 @@ mysqld_LDADD =		@MYSQLD_EXTRA_LDFLAGS@ \
 			@pstack_libs@ \
 			@mysql_plugin_libs@ \
 			$(LDADD)  $(CXXLDFLAGS) $(WRAPLIBS) @LIBDL@ \
-                        $(yassl_libs) $(openssl_libs)
+			$(yassl_libs) $(openssl_libs) \
+			@MYSQLD_EXTRA_LIBS@
 
 noinst_HEADERS =	item.h item_func.h item_sum.h item_cmpfunc.h \
 			item_strfunc.h item_timefunc.h \

sql/ha_ndbcluster.cc

@@ -278,11 +278,6 @@ inline
 int execute_no_commit(ha_ndbcluster *h, NdbTransaction *trans,
 		      bool force_release)
 {
-#ifdef NOT_USED
-  int m_batch_execute= 0;
-  if (m_batch_execute)
-    return 0;
-#endif
   h->release_completed_operations(trans, force_release);
   return h->m_ignore_no_key ?
     execute_no_commit_ignore_no_key(h,trans) :
@@ -294,11 +289,6 @@ int execute_no_commit(ha_ndbcluster *h, NdbTransaction *trans,
 inline
 int execute_commit(ha_ndbcluster *h, NdbTransaction *trans)
 {
-#ifdef NOT_USED
-  int m_batch_execute= 0;
-  if (m_batch_execute)
-    return 0;
-#endif
   return trans->execute(NdbTransaction::Commit,
                         NdbOperation::AbortOnError,
                         h->m_force_send);
@@ -307,11 +297,6 @@ int execute_commit(ha_ndbcluster *h, NdbTransaction *trans)
 inline
 int execute_commit(THD *thd, NdbTransaction *trans)
 {
-#ifdef NOT_USED
-  int m_batch_execute= 0;
-  if (m_batch_execute)
-    return 0;
-#endif
   return trans->execute(NdbTransaction::Commit,
                         NdbOperation::AbortOnError,
                         thd->variables.ndb_force_send);
@@ -321,11 +306,6 @@ inline
 int execute_no_commit_ie(ha_ndbcluster *h, NdbTransaction *trans,
 			 bool force_release)
 {
-#ifdef NOT_USED
-  int m_batch_execute= 0;
-  if (m_batch_execute)
-    return 0;
-#endif
   h->release_completed_operations(trans, force_release);
   return trans->execute(NdbTransaction::NoCommit,
                         NdbOperation::AO_IgnoreError,
@@ -2925,7 +2905,8 @@ int ha_ndbcluster::update_row(const uchar *old_data, uchar *new_data)
    * If IGNORE the ignore constraint violations on primary and unique keys,
    * but check that it is not part of INSERT ... ON DUPLICATE KEY UPDATE
    */
-  if (m_ignore_dup_key && thd->lex->sql_command == SQLCOM_UPDATE)
+  if (m_ignore_dup_key && (thd->lex->sql_command == SQLCOM_UPDATE ||
+                           thd->lex->sql_command == SQLCOM_UPDATE_MULTI))
   {
     int peek_res= peek_indexed_rows(new_data, pk_update);
     
@@ -4267,8 +4248,6 @@ THR_LOCK_DATA **ha_ndbcluster::store_lock(THD *thd,
 extern MASTER_INFO *active_mi;
 static int ndbcluster_update_apply_status(THD *thd, int do_update)
 {
-  return 0;
-
   Thd_ndb *thd_ndb= get_thd_ndb(thd);
   Ndb *ndb= thd_ndb->ndb;
   NDBDICT *dict= ndb->getDictionary();

sql/my_decimal.h

@@ -40,6 +40,7 @@ C_MODE_END
 /* the number of digits that my_decimal can possibly contain */
 #define DECIMAL_MAX_POSSIBLE_PRECISION (DECIMAL_BUFF_LENGTH * 9)
 
+
 /*
   maximum guaranteed precision of number in decimal digits (number of our
   digits * number of decimal digits in one our big digit - number of decimal

sql/sql_parse.cc

@@ -5022,17 +5022,14 @@ bool check_merge_table_access(THD *thd, char *db,
 	Check stack size; Send error if there isn't enough stack to continue
 ****************************************************************************/
 
-#if STACK_DIRECTION < 0
+#ifndef EMBEDDED_LIBRARY
-#define used_stack(A,B) (long) (A - B)
+
-#else
+#define used_stack(A,B) (long)(A > B ? A - B : B - A)
-#define used_stack(A,B) (long) (B - A)
-#endif
 
 #ifndef DBUG_OFF
 long max_stack_used;
 #endif
 
-#ifndef EMBEDDED_LIBRARY
 /*
   Note: The 'buf' parameter is necessary, even if it is unused here.
   - fix_fields functions has a "dummy" buffer large enough for the

storage/ndb/include/mgmapi/mgmapi_config_parameters.h

@@ -64,6 +64,7 @@
 
 #define CFG_DB_FILESYSTEM_PATH        125
 #define CFG_DB_NO_REDOLOG_FILES       126
+#define CFG_DB_REDOLOG_FILE_SIZE      140
 
 #define CFG_DB_LCP_DISC_PAGES_TUP     127
 #define CFG_DB_LCP_DISC_PAGES_TUP_SR  128
@@ -81,6 +82,8 @@
 #define CFG_DB_BACKUP_WRITE_SIZE          136
 #define CFG_DB_BACKUP_MAX_WRITE_SIZE      139
 
+#define CFG_DB_WATCHDOG_INTERVAL_INITIAL  141
+
 #define CFG_LOG_DESTINATION           147
 
 #define CFG_DB_DISCLESS               148
@@ -113,6 +116,8 @@
 
 #define CFG_DB_MEMREPORT_FREQUENCY    166
 
+#define CFG_DB_O_DIRECT               168
+
 #define CFG_DB_SGA                    198 /* super pool mem */
 #define CFG_DB_DATA_MEM_2             199 /* used in special build in 5.1 */
 

storage/ndb/include/ndb_global.h.in

@@ -144,4 +144,6 @@ extern "C" {
 #define MAX(x,y) (((x)>(y))?(x):(y))
 #endif
 
+#define NDB_O_DIRECT_WRITE_ALIGNMENT 512
+
 #endif

storage/ndb/include/portlib/NdbTick.h

@@ -37,9 +37,6 @@ NDB_TICKS NdbTick_CurrentMillisecond(void);
  */
 int NdbTick_CurrentMicrosecond(NDB_TICKS * secs, Uint32 * micros);
 
-  /*#define TIME_MEASUREMENT*/
-#ifdef TIME_MEASUREMENT
-
 struct MicroSecondTimer {
   NDB_TICKS seconds;
   NDB_TICKS micro_seconds;
@@ -54,7 +51,6 @@ struct MicroSecondTimer {
 NDB_TICKS NdbTick_getMicrosPassed(struct MicroSecondTimer start,
                             struct MicroSecondTimer stop);
 int NdbTick_getMicroTimer(struct MicroSecondTimer* time_now);
-#endif
 
 #ifdef	__cplusplus
 }

storage/ndb/ndbapi-examples/mgmapi_logevent/Makefile

@@ -1,6 +1,6 @@
 TARGET = mgmapi_logevent
-SRCS = $(TARGET).cpp
+SRCS = main.cpp
-OBJS = $(TARGET).o
+OBJS = main.o
 CXX = g++
 CFLAGS = -c -Wall -fno-rtti -fno-exceptions
 CXXFLAGS = 
@@ -17,7 +17,7 @@ SYS_LIB =
 $(TARGET): $(OBJS)
 	$(CXX) $(CXXFLAGS) $(LFLAGS) $(LIB_DIR) $(OBJS) -lndbclient -lmysqlclient_r -lmysys -lmystrings -lz $(SYS_LIB) -o $(TARGET)
 
-$(TARGET).o: $(SRCS)
+$(OBJS): $(SRCS)
 	$(CXX) $(CFLAGS) -I$(TOP_SRCDIR)/include -I$(INCLUDE_DIR) -I$(INCLUDE_DIR)/mgmapi  -I$(INCLUDE_DIR)/ndbapi $(SRCS)
 
 clean:

storage/ndb/ndbapi-examples/mgmapi_logevent2/Makefile

@@ -1,6 +1,6 @@
 TARGET = mgmapi_logevent2
-SRCS = $(TARGET).cpp
+SRCS = main.cpp
-OBJS = $(TARGET).o
+OBJS = main.o
 CXX = g++
 CFLAGS = -c -Wall -fno-rtti -fno-exceptions
 CXXFLAGS = 
@@ -17,7 +17,7 @@ SYS_LIB =
 $(TARGET): $(OBJS)
 	$(CXX) $(CXXFLAGS) $(LFLAGS) $(LIB_DIR) $(OBJS) -lndbclient -lmysqlclient_r -lmysys -lmystrings -lz $(SYS_LIB) -o $(TARGET)
 
-$(TARGET).o: $(SRCS)
+$(OBJS): $(SRCS)
 	$(CXX) $(CFLAGS) -I$(TOP_SRCDIR)/include -I$(INCLUDE_DIR) -I$(INCLUDE_DIR)/mgmapi  -I$(INCLUDE_DIR)/ndbapi $(SRCS)
 
 clean:

storage/ndb/ndbapi-examples/ndbapi_simple_dual/Makefile

@@ -1,6 +1,6 @@
 TARGET = ndbapi_simple_dual
-SRCS = $(TARGET).cpp
+SRCS = main.cpp
-OBJS = $(TARGET).o
+OBJS = main.o
 CXX = g++
 CFLAGS = -c -Wall -fno-rtti -fno-exceptions
 CXXFLAGS = 
@@ -17,7 +17,7 @@ SYS_LIB =
 $(TARGET): $(OBJS)
 	$(CXX) $(CXXFLAGS) $(LFLAGS) $(LIB_DIR) $(OBJS) -lndbclient -lmysqlclient_r -lmysys -lmystrings  -lz $(SYS_LIB) -o $(TARGET)
 
-$(TARGET).o: $(SRCS)
+$(OBJS): $(SRCS)
 	$(CXX) $(CFLAGS) -I$(TOP_SRCDIR)/include -I$(INCLUDE_DIR) -I$(INCLUDE_DIR)/ndbapi $(SRCS)
 
 clean:

storage/ndb/ndbapi-examples/ndbapi_simple_index/Makefile

@@ -1,6 +1,6 @@
 TARGET = ndbapi_simple_index
-SRCS = $(TARGET).cpp
+SRCS = main.cpp
-OBJS = $(TARGET).o
+OBJS = main.o
 CXX = g++
 CFLAGS = -c -Wall -fno-rtti -fno-exceptions
 CXXFLAGS = 
@@ -17,7 +17,7 @@ SYS_LIB =
 $(TARGET): $(OBJS)
 	$(CXX) $(CXXFLAGS) $(LFLAGS) $(LIB_DIR) $(OBJS) -lndbclient -lmysqlclient_r -lmysys -lmystrings -lz $(SYS_LIB) -o $(TARGET)
 
-$(TARGET).o: $(SRCS)
+$(OBJS): $(SRCS)
 	$(CXX) $(CFLAGS) -I$(INCLUDE_DIR)/include -I$(INCLUDE_DIR)/storage/ndb/include -I$(INCLUDE_DIR)/storage/ndb/include/ndbapi $(SRCS)
 
 clean:

storage/ndb/src/common/portlib/NdbTick.c

@@ -15,7 +15,7 @@
 
 
 #include <ndb_global.h>
-#include "NdbTick.h"
+#include <NdbTick.h>
 
 #define NANOSEC_PER_SEC  1000000000
 #define MICROSEC_PER_SEC 1000000
@@ -71,7 +71,6 @@ NdbTick_CurrentMicrosecond(NDB_TICKS * secs, Uint32 * micros){
 }
 
 #endif
-#ifdef TIME_MEASUREMENT
 int
 NdbTick_getMicroTimer(struct MicroSecondTimer* input_timer)
 {
@@ -102,4 +101,3 @@ NdbTick_getMicrosPassed(struct MicroSecondTimer start,
   }
   return ret_value;
 }
-#endif

storage/ndb/src/common/transporter/TransporterRegistry.cpp

@@ -818,6 +818,7 @@ TransporterRegistry::performReceive()
         {
           Uint32 * ptr;
           Uint32 sz = t->getReceiveData(&ptr);
+          transporter_recv_from(callbackObj, nodeId);
           Uint32 szUsed = unpack(ptr, sz, nodeId, ioStates[nodeId]);
           t->updateReceiveDataPtr(szUsed);
 	}

storage/ndb/src/kernel/blocks/ERROR_codes.txt

@@ -1,5 +1,5 @@
 Next QMGR 1
-Next NDBCNTR 1001
+Next NDBCNTR 1002
 Next NDBFS 2000
 Next DBACC 3002
 Next DBTUP 4029
@@ -523,3 +523,4 @@ Dbtup:
 NDBCNTR:
 
 1000: Crash insertion on SystemError::CopyFragRef
+1001: Delay sending NODE_FAILREP (to own node), until error is cleared

storage/ndb/src/kernel/blocks/backup/Backup.cpp

@@ -2771,6 +2771,8 @@ Backup::openFiles(Signal* signal, BackupRecordPtr ptr)
   c_backupFilePool.getPtr(filePtr, ptr.p->dataFilePtr);
   filePtr.p->m_flags |= BackupFile::BF_OPENING;
 
+  if (c_defaults.m_o_direct)
+    req->fileFlags |= FsOpenReq::OM_DIRECT;
   req->userPointer = filePtr.i;
   FsOpenReq::setVersion(req->fileNumber, 2);
   FsOpenReq::setSuffix(req->fileNumber, FsOpenReq::S_DATA);
@@ -3745,12 +3747,31 @@ Backup::OperationRecord::newFragment(Uint32 tableId, Uint32 fragNo)
 }
 
 bool
-Backup::OperationRecord::fragComplete(Uint32 tableId, Uint32 fragNo)
+Backup::OperationRecord::fragComplete(Uint32 tableId, Uint32 fragNo, bool fill_record)
 {
   Uint32 * tmp;
   const Uint32 footSz = sizeof(BackupFormat::DataFile::FragmentFooter) >> 2;
+  Uint32 sz = footSz + 1;
 
-  if(dataBuffer.getWritePtr(&tmp, footSz + 1)) {
+  if (fill_record)
+  {
+    Uint32 * new_tmp;
+    if (!dataBuffer.getWritePtr(&tmp, sz))
+      return false;
+    new_tmp = tmp + sz;
+
+    if ((UintPtr)new_tmp & (sizeof(Page32)-1))
+    {
+      /* padding is needed to get full write */
+      new_tmp += 2 /* to fit empty header minimum 2 words*/;
+      new_tmp = (Uint32 *)(((UintPtr)new_tmp + sizeof(Page32)-1) &
+                            ~(UintPtr)(sizeof(Page32)-1));
+      /* new write sz */
+      sz = new_tmp - tmp;
+    }
+  }
+
+  if(dataBuffer.getWritePtr(&tmp, sz)) {
     jam();
     * tmp = 0; // Finish record stream
     tmp++;
@@ -3762,7 +3783,17 @@ Backup::OperationRecord::fragComplete(Uint32 tableId, Uint32 fragNo)
     foot->FragmentNo    = htonl(fragNo);
     foot->NoOfRecords   = htonl(noOfRecords);
     foot->Checksum      = htonl(0);
-    dataBuffer.updateWritePtr(footSz + 1);
+
+    if (sz != footSz + 1)
+    {
+      tmp += footSz;
+      memset(tmp, 0, (sz - footSz - 1) * 4);
+      *tmp = htonl(BackupFormat::EMPTY_ENTRY);
+      tmp++;
+      *tmp = htonl(sz - footSz - 1);
+    }
+
+    dataBuffer.updateWritePtr(sz);
     return true;
   }//if
   return false;
@@ -3864,8 +3895,13 @@ Backup::fragmentCompleted(Signal* signal, BackupFilePtr filePtr)
     return;
   }//if
     
+  BackupRecordPtr ptr LINT_SET_PTR;
+  c_backupPool.getPtr(ptr, filePtr.p->backupPtr);
+
   OperationRecord & op = filePtr.p->operation;
-  if(!op.fragComplete(filePtr.p->tableId, filePtr.p->fragmentNo)) {
+  if(!op.fragComplete(filePtr.p->tableId, filePtr.p->fragmentNo,
+                      c_defaults.m_o_direct))
+  {
     jam();
     signal->theData[0] = BackupContinueB::BUFFER_FULL_FRAG_COMPLETE;
     signal->theData[1] = filePtr.i;
@@ -3875,9 +3911,6 @@ Backup::fragmentCompleted(Signal* signal, BackupFilePtr filePtr)
   
   filePtr.p->m_flags &= ~(Uint32)BackupFile::BF_SCAN_THREAD;
   
-  BackupRecordPtr ptr LINT_SET_PTR;
-  c_backupPool.getPtr(ptr, filePtr.p->backupPtr);
-
   if (ptr.p->is_lcp())
   {
     ptr.p->slaveState.setState(STOPPING);
@@ -4914,6 +4947,8 @@ Backup::lcp_open_file(Signal* signal, BackupRecordPtr ptr)
     FsOpenReq::OM_CREATE | 
     FsOpenReq::OM_APPEND |
     FsOpenReq::OM_AUTOSYNC;
+  if (c_defaults.m_o_direct)
+    req->fileFlags |= FsOpenReq::OM_DIRECT;
   FsOpenReq::v2_setCount(req->fileNumber, 0xFFFFFFFF);
   req->auto_sync_size = c_defaults.m_disk_synch_size;
   

storage/ndb/src/kernel/blocks/backup/Backup.hpp

@@ -240,7 +240,7 @@ public:
      * Once per fragment
      */
     bool newFragment(Uint32 tableId, Uint32 fragNo);
-    bool fragComplete(Uint32 tableId, Uint32 fragNo);
+    bool fragComplete(Uint32 tableId, Uint32 fragNo, bool fill_record);
     
     /**
      * Once per scan frag (next) req/conf
@@ -534,6 +534,7 @@ public:
     Uint32 m_disk_write_speed;
     Uint32 m_disk_synch_size;
     Uint32 m_diskless;
+    Uint32 m_o_direct;
   };
   
   /**

storage/ndb/src/kernel/blocks/backup/BackupFormat.hpp

@@ -32,7 +32,8 @@ struct BackupFormat {
     TABLE_LIST        = 4,
     TABLE_DESCRIPTION = 5,
     GCP_ENTRY         = 6,
-    FRAGMENT_INFO     = 7
+    FRAGMENT_INFO     = 7,
+    EMPTY_ENTRY       = 8
   };
 
   struct FileHeader {
@@ -93,6 +94,13 @@ struct BackupFormat {
       Uint32 NoOfRecords;
       Uint32 Checksum;
     };
+
+    /* optional padding for O_DIRECT */
+    struct EmptyEntry {
+      Uint32 SectionType;
+      Uint32 SectionLength;
+      /* not used data */
+    };
   };
 
   /**

storage/ndb/src/kernel/blocks/backup/BackupInit.cpp

@@ -148,10 +148,13 @@ Backup::execREAD_CONFIG_REQ(Signal* signal)
   c_defaults.m_disk_write_speed = 10 * (1024 * 1024);
   c_defaults.m_disk_write_speed_sr = 100 * (1024 * 1024);
   c_defaults.m_disk_synch_size = 4 * (1024 * 1024);
+  c_defaults.m_o_direct = true;
 
   Uint32 noBackups = 0, noTables = 0, noAttribs = 0, noFrags = 0;
   ndbrequire(!ndb_mgm_get_int_parameter(p, CFG_DB_DISCLESS, 
 					&c_defaults.m_diskless));
+  ndb_mgm_get_int_parameter(p, CFG_DB_O_DIRECT,
+                            &c_defaults.m_o_direct);
   ndb_mgm_get_int_parameter(p, CFG_DB_CHECKPOINT_SPEED_SR,
 			    &c_defaults.m_disk_write_speed_sr);
   ndb_mgm_get_int_parameter(p, CFG_DB_CHECKPOINT_SPEED,
@@ -204,7 +207,7 @@ Backup::execREAD_CONFIG_REQ(Signal* signal)
     / sizeof(Page32);
   // We need to allocate an additional of 2 pages. 1 page because of a bug in
   // ArrayPool and another one for DICTTAINFO.
-  c_pagePool.setSize(noPages + NO_OF_PAGES_META_FILE + 2); 
+  c_pagePool.setSize(noPages + NO_OF_PAGES_META_FILE + 2, true); 
   
   { // Init all tables
     SLList<Table> tables(c_tablePool);

storage/ndb/src/kernel/blocks/backup/FsBuffer.hpp

@@ -270,8 +270,8 @@ FsBuffer::getReadPtr(Uint32 ** ptr, Uint32 * sz, bool * _eof){
     
     * ptr = &Tp[Tr];
 
-    DEBUG(ndbout_c("getReadPtr() Tr: %d Tw: %d Ts: %d Tm: %d sz1: %d -> %d",
+    DEBUG(ndbout_c("getReadPtr() Tr: %d Tmw: %d Ts: %d Tm: %d sz1: %d -> %d",
-		   Tr, Tw, Ts, Tm, sz1, * sz));
+		   Tr, Tmw, Ts, Tm, sz1, * sz));
 
     return true;
   }
@@ -279,8 +279,8 @@ FsBuffer::getReadPtr(Uint32 ** ptr, Uint32 * sz, bool * _eof){
   if(!m_eof){
     * _eof = false;
     
-    DEBUG(ndbout_c("getReadPtr() Tr: %d Tw: %d Ts: %d Tm: %d sz1: %d -> false",
+    DEBUG(ndbout_c("getReadPtr() Tr: %d Tmw: %d Ts: %d Tm: %d sz1: %d -> false",
-		   Tr, Tw, Ts, Tm, sz1));
+		   Tr, Tmw, Ts, Tm, sz1));
     
     return false;
   }
@@ -289,8 +289,8 @@ FsBuffer::getReadPtr(Uint32 ** ptr, Uint32 * sz, bool * _eof){
   * _eof = true;
   * ptr = &Tp[Tr];
 
-  DEBUG(ndbout_c("getReadPtr() Tr: %d Tw: %d Ts: %d Tm: %d sz1: %d -> %d eof",
+  DEBUG(ndbout_c("getReadPtr() Tr: %d Tmw: %d Ts: %d Tm: %d sz1: %d -> %d eof",
-		 Tr, Tw, Ts, Tm, sz1, * sz));
+		 Tr, Tmw, Ts, Tm, sz1, * sz));
   
   return false;
 }
@@ -316,13 +316,13 @@ FsBuffer::getWritePtr(Uint32 ** ptr, Uint32 sz){
   if(sz1 > sz){ // Note at least 1 word of slack
     * ptr = &Tp[Tw];
 
-    DEBUG(ndbout_c("getWritePtr(%d) Tr: %d Tw: %d Ts: %d sz1: %d -> true",
+    DEBUG(ndbout_c("getWritePtr(%d) Tw: %d sz1: %d -> true",
-		   sz, Tr, Tw, Ts, sz1));
+		   sz, Tw, sz1));
     return true;
   }
 
-  DEBUG(ndbout_c("getWritePtr(%d) Tr: %d Tw: %d Ts: %d sz1: %d -> false",
+  DEBUG(ndbout_c("getWritePtr(%d) Tw: %d sz1: %d -> false",
-		 sz, Tr, Tw, Ts, sz1));
+		 sz, Tw, sz1));
 
   return false;
 }
@@ -339,11 +339,15 @@ FsBuffer::updateWritePtr(Uint32 sz){
   m_free -= sz;
   if(Tnew < Ts){
     m_writeIndex = Tnew;
+    DEBUG(ndbout_c("updateWritePtr(%d) m_writeIndex: %d",
+                   sz, m_writeIndex));
     return;
   }
 
   memcpy(Tp, &Tp[Ts], (Tnew - Ts) << 2);
   m_writeIndex = Tnew - Ts;
+  DEBUG(ndbout_c("updateWritePtr(%d) m_writeIndex: %d",
+                 sz, m_writeIndex));
 }
 
 inline

storage/ndb/src/kernel/blocks/dbdict/Dbdict.cpp

@@ -698,6 +698,9 @@ void Dbdict::execFSCLOSECONF(Signal* signal)
   case FsConnectRecord::OPEN_READ_SCHEMA2:
     openSchemaFile(signal, 1, fsPtr.i, false, false);
     break;
+  case FsConnectRecord::OPEN_READ_TAB_FILE2:
+    openTableFile(signal, 1, fsPtr.i, c_readTableRecord.tableId, false);
+    break;
   default:
     jamLine((fsPtr.p->fsState & 0xFFF));
     ndbrequire(false);
@@ -1073,8 +1076,11 @@ void Dbdict::readTableConf(Signal* signal,
 void Dbdict::readTableRef(Signal* signal,
                           FsConnectRecordPtr fsPtr)
 {
+  /**
+   * First close corrupt file
+   */
   fsPtr.p->fsState = FsConnectRecord::OPEN_READ_TAB_FILE2;
-  openTableFile(signal, 1, fsPtr.i, c_readTableRecord.tableId, false);
+  closeFile(signal, fsPtr.p->filePtr, fsPtr.i);
   return;
 }//Dbdict::readTableRef()
 

storage/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp

@@ -4741,12 +4741,18 @@ void Dbdih::failedNodeLcpHandling(Signal* signal, NodeRecordPtr failedNodePtr)
   jam();
   const Uint32 nodeId = failedNodePtr.i;
 
-  if (c_lcpState.m_participatingLQH.get(failedNodePtr.i)){
+  if (isMaster() && c_lcpState.m_participatingLQH.get(failedNodePtr.i))
+  {
     /*----------------------------------------------------*/
     /*  THE NODE WAS INVOLVED IN A LOCAL CHECKPOINT. WE   */
     /* MUST UPDATE THE ACTIVE STATUS TO INDICATE THAT     */
     /* THE NODE HAVE MISSED A LOCAL CHECKPOINT.           */
     /*----------------------------------------------------*/
+
+    /**
+     * Bug#28717, Only master should do this, as this status is copied
+     *   to other nodes
+     */
     switch (failedNodePtr.p->activeStatus) {
     case Sysfile::NS_Active:
       jam();

storage/ndb/src/kernel/blocks/dblqh/Dblqh.hpp

@@ -71,7 +71,6 @@ class Dbtup;
 /*       CONSTANTS OF THE LOG PAGES                                          */
 /* ------------------------------------------------------------------------- */
 #define ZPAGE_HEADER_SIZE 32
-#define ZNO_MBYTES_IN_FILE 16
 #define ZPAGE_SIZE 8192
 #define ZPAGES_IN_MBYTE 32
 #define ZTWOLOG_NO_PAGES_IN_MBYTE 5
@@ -115,9 +114,6 @@ class Dbtup;
 /* ------------------------------------------------------------------------- */
 /*       VARIOUS CONSTANTS USED AS FLAGS TO THE FILE MANAGER.                */
 /* ------------------------------------------------------------------------- */
-#define ZOPEN_READ 0
-#define ZOPEN_WRITE 1
-#define ZOPEN_READ_WRITE 2
 #define ZVAR_NO_LOG_PAGE_WORD 1
 #define ZLIST_OF_PAIRS 0
 #define ZLIST_OF_PAIRS_SYNCH 16
@@ -142,7 +138,7 @@ class Dbtup;
 /*       IN THE MBYTE.                                                       */
 /* ------------------------------------------------------------------------- */
 #define ZFD_HEADER_SIZE 3
-#define ZFD_PART_SIZE 48
+#define ZFD_MBYTE_SIZE 3
 #define ZLOG_HEAD_SIZE 8
 #define ZNEXT_LOG_SIZE 2
 #define ZABORT_LOG_SIZE 3
@@ -169,7 +165,6 @@ class Dbtup;
 #define ZPOS_LOG_TYPE 0
 #define ZPOS_NO_FD 1
 #define ZPOS_FILE_NO 2
-#define ZMAX_LOG_FILES_IN_PAGE_ZERO 40
 /* ------------------------------------------------------------------------- */
 /*       THE POSITIONS WITHIN A PREPARE LOG RECORD AND A NEW PREPARE         */
 /*       LOG RECORD.                                                         */
@@ -1437,17 +1432,17 @@ public:
      *       header of each log file.  That information is used during
      *       system restart to find the tail of the log.  
      */
-    UintR logLastPrepRef[16];
+    UintR *logLastPrepRef;
     /**
      *       The max global checkpoint completed before the mbyte in the
      *       log file was started.  One variable per mbyte.  
      */
-    UintR logMaxGciCompleted[16];
+    UintR *logMaxGciCompleted;
     /**
      *       The max global checkpoint started before the mbyte in the log
      *       file was started.  One variable per mbyte.
      */
-    UintR logMaxGciStarted[16];
+    UintR *logMaxGciStarted;
     /**
      *       This variable contains the file name as needed by the file 
      *       system when opening the file.
@@ -2163,6 +2158,7 @@ private:
   void execSTART_RECREF(Signal* signal);
 
   void execGCP_SAVEREQ(Signal* signal);
+  void execFSOPENREF(Signal* signal);
   void execFSOPENCONF(Signal* signal);
   void execFSCLOSECONF(Signal* signal);
   void execFSWRITECONF(Signal* signal);
@@ -2671,6 +2667,8 @@ private:
   LogPartRecord *logPartRecord;
   LogPartRecordPtr logPartPtr;
   UintR clogPartFileSize;
+  Uint32 clogFileSize; // In MBYTE
+  Uint32 cmaxLogFilesInPageZero; //
 
 // Configurable
   LogFileRecord *logFileRecord;
@@ -2678,13 +2676,15 @@ private:
   UintR cfirstfreeLogFile;
   UintR clogFileFileSize;
 
-#define ZLFO_FILE_SIZE 256            /* MAX 256 OUTSTANDING FILE OPERATIONS */
+#define ZLFO_MIN_FILE_SIZE 256
+// RedoBuffer/32K minimum ZLFO_MIN_FILE_SIZE
   LogFileOperationRecord *logFileOperationRecord;
   LogFileOperationRecordPtr lfoPtr;
   UintR cfirstfreeLfo;
   UintR clfoFileSize;
 
   LogPageRecord *logPageRecord;
+  void *logPageRecordUnaligned;
   LogPageRecordPtr logPagePtr;
   UintR cfirstfreeLogPage;
   UintR clogPageFileSize;
@@ -2695,7 +2695,7 @@ private:
   UintR cfirstfreePageRef;
   UintR cpageRefFileSize;
 
-#define ZSCANREC_FILE_SIZE 100
+// Configurable
   ArrayPool<ScanRecord> c_scanRecordPool;
   ScanRecordPtr scanptr;
   UintR cscanNoFreeRec;
@@ -2888,6 +2888,7 @@ private:
   UintR ctransidHash[1024];
   
   Uint32 c_diskless;
+  Uint32 c_o_direct;
   Uint32 c_error_insert_table_id;
   
 public:

storage/ndb/src/kernel/blocks/dblqh/DblqhInit.cpp

@@ -30,11 +30,11 @@ void Dblqh::initData()
   cgcprecFileSize = ZGCPREC_FILE_SIZE;
   chostFileSize = MAX_NDB_NODES;
   clcpFileSize = ZNO_CONCURRENT_LCP;
-  clfoFileSize = ZLFO_FILE_SIZE;
+  clfoFileSize = 0;
   clogFileFileSize = 0;
   clogPartFileSize = ZLOG_PART_FILE_SIZE;
   cpageRefFileSize = ZPAGE_REF_FILE_SIZE;
-  cscanrecFileSize = ZSCANREC_FILE_SIZE;
+  cscanrecFileSize = 0;
   ctabrecFileSize = 0;
   ctcConnectrecFileSize = 0;
   ctcNodeFailrecFileSize = MAX_NDB_NODES;
@@ -49,6 +49,7 @@ void Dblqh::initData()
   logFileRecord = 0;
   logFileOperationRecord = 0;
   logPageRecord = 0;
+  logPageRecordUnaligned= 0;
   pageRefRecord = 0;
   tablerec = 0;
   tcConnectionrec = 0;
@@ -60,6 +61,8 @@ void Dblqh::initData()
   cLqhTimeOutCheckCount = 0;
   cbookedAccOps = 0;
   m_backup_ptr = RNIL;
+  clogFileSize = 16;
+  cmaxLogFilesInPageZero = 40;
 }//Dblqh::initData()
 
 void Dblqh::initRecords() 
@@ -105,9 +108,12 @@ void Dblqh::initRecords()
 		sizeof(LogFileOperationRecord), 
 		clfoFileSize);
 
-  logPageRecord = (LogPageRecord*)allocRecord("LogPageRecord",
+  logPageRecord =
+    (LogPageRecord*)allocRecordAligned("LogPageRecord",
                                        sizeof(LogPageRecord),
                                        clogPageFileSize,
+                                       &logPageRecordUnaligned,
+                                       NDB_O_DIRECT_WRITE_ALIGNMENT,
                                        false);
 
   pageRefRecord = (PageRefRecord*)allocRecord("PageRefRecord",
@@ -260,6 +266,7 @@ Dblqh::Dblqh(Block_context& ctx):
   addRecSignal(GSN_START_FRAGREQ, &Dblqh::execSTART_FRAGREQ);
   addRecSignal(GSN_START_RECREF, &Dblqh::execSTART_RECREF);
   addRecSignal(GSN_GCP_SAVEREQ, &Dblqh::execGCP_SAVEREQ);
+  addRecSignal(GSN_FSOPENREF, &Dblqh::execFSOPENREF, true);
   addRecSignal(GSN_FSOPENCONF, &Dblqh::execFSOPENCONF);
   addRecSignal(GSN_FSCLOSECONF, &Dblqh::execFSCLOSECONF);
   addRecSignal(GSN_FSWRITECONF, &Dblqh::execFSWRITECONF);
@@ -377,7 +384,7 @@ Dblqh::~Dblqh()
 		sizeof(LogFileOperationRecord), 
 		clfoFileSize);
   
-  deallocRecord((void**)&logPageRecord,
+  deallocRecord((void**)&logPageRecordUnaligned,
 		"LogPageRecord",
 		sizeof(LogPageRecord),
 		clogPageFileSize);

storage/ndb/src/kernel/blocks/dblqh/DblqhMain.cpp

@@ -1023,6 +1023,11 @@ void Dblqh::execREAD_CONFIG_REQ(Signal* signal)
     clogPageFileSize+= (16 - mega_byte_part);
   }
 
+  /* maximum number of log file operations */
+  clfoFileSize = clogPageFileSize;
+  if (clfoFileSize < ZLFO_MIN_FILE_SIZE)
+    clfoFileSize = ZLFO_MIN_FILE_SIZE;
+
   ndbrequire(!ndb_mgm_get_int_parameter(p, CFG_LQH_TABLE, &ctabrecFileSize));
   ndbrequire(!ndb_mgm_get_int_parameter(p, CFG_LQH_TC_CONNECT, 
 					&ctcConnectrecFileSize));
@@ -1031,11 +1036,41 @@ void Dblqh::execREAD_CONFIG_REQ(Signal* signal)
   cmaxAccOps = cscanrecFileSize * MAX_PARALLEL_OP_PER_SCAN;
 
   ndbrequire(!ndb_mgm_get_int_parameter(p, CFG_DB_DISCLESS, &c_diskless));
+  c_o_direct = true;
+  ndb_mgm_get_int_parameter(p, CFG_DB_O_DIRECT, &c_o_direct);
   
   Uint32 tmp= 0;
   ndbrequire(!ndb_mgm_get_int_parameter(p, CFG_LQH_FRAG, &tmp));
   c_fragment_pool.setSize(tmp);
 
+  if (!ndb_mgm_get_int_parameter(p, CFG_DB_REDOLOG_FILE_SIZE,
+                                 &clogFileSize))
+  {
+    // convert to mbyte
+    clogFileSize = (clogFileSize + 1024*1024 - 1) / (1024 * 1024);
+    ndbrequire(clogFileSize >= 4 && clogFileSize <= 1024);
+  }
+
+  cmaxLogFilesInPageZero = (ZPAGE_SIZE - ZPAGE_HEADER_SIZE - 128) /
+    (ZFD_MBYTE_SIZE * clogFileSize);
+
+  /**
+   * "Old" cmaxLogFilesInPageZero was 40
+   * Each FD need 3 words per mb, require that they can fit into 1 page
+   *   (atleast 1 FD)
+   * Is also checked in ConfigInfo.cpp (max FragmentLogFileSize = 1Gb)
+   *   1Gb = 1024Mb => 3(ZFD_MBYTE_SIZE) * 1024 < 8192 (ZPAGE_SIZE)
+   */
+  if (cmaxLogFilesInPageZero > 40)
+  {
+    jam();
+    cmaxLogFilesInPageZero = 40;
+  }
+  else
+  {
+    ndbrequire(cmaxLogFilesInPageZero);
+  }
+  
   initRecords();
   initialiseRecordsLab(signal, 0, ref, senderData);
 
@@ -11788,9 +11823,9 @@ void Dblqh::sendStartLcp(Signal* signal)
 Uint32 Dblqh::remainingLogSize(const LogFileRecordPtr &sltCurrLogFilePtr,
 			       const LogPartRecordPtr &sltLogPartPtr)
 {
-  Uint32 hf = sltCurrLogFilePtr.p->fileNo*ZNO_MBYTES_IN_FILE+sltCurrLogFilePtr.p->currentMbyte;
+  Uint32 hf = sltCurrLogFilePtr.p->fileNo*clogFileSize+sltCurrLogFilePtr.p->currentMbyte;
-  Uint32 tf = sltLogPartPtr.p->logTailFileNo*ZNO_MBYTES_IN_FILE+sltLogPartPtr.p->logTailMbyte;
+  Uint32 tf = sltLogPartPtr.p->logTailFileNo*clogFileSize+sltLogPartPtr.p->logTailMbyte;
-  Uint32 sz = sltLogPartPtr.p->noLogFiles*ZNO_MBYTES_IN_FILE;
+  Uint32 sz = sltLogPartPtr.p->noLogFiles*clogFileSize;
   if (tf > hf) hf += sz;
   return sz-(hf-tf);
 }
@@ -11848,7 +11883,7 @@ void Dblqh::setLogTail(Signal* signal, Uint32 keepGci)
 /* ------------------------------------------------------------------------- */
   SLT_LOOP:
     for (tsltIndex = tsltStartMbyte;
-	 tsltIndex <= ZNO_MBYTES_IN_FILE - 1; 
+	 tsltIndex <= clogFileSize - 1;
 	 tsltIndex++) {
       if (sltLogFilePtr.p->logMaxGciStarted[tsltIndex] >= keepGci) {
 /* ------------------------------------------------------------------------- */
@@ -11864,7 +11899,7 @@ void Dblqh::setLogTail(Signal* signal, Uint32 keepGci)
 /* ------------------------------------------------------------------------- */
 /*STEPPING BACK INCLUDES ALSO STEPPING BACK TO THE PREVIOUS LOG FILE.        */
 /* ------------------------------------------------------------------------- */
-          tsltMbyte = ZNO_MBYTES_IN_FILE - 1;
+          tsltMbyte = clogFileSize - 1;
           sltLogFilePtr.i = sltLogFilePtr.p->prevLogFile;
           ptrCheckGuard(sltLogFilePtr, clogFileFileSize, logFileRecord);
         }//if
@@ -11902,7 +11937,7 @@ void Dblqh::setLogTail(Signal* signal, Uint32 keepGci)
       UintR ToldTailFileNo = sltLogPartPtr.p->logTailFileNo;
       UintR ToldTailMByte = sltLogPartPtr.p->logTailMbyte;
 
-      arrGuard(tsltMbyte, 16);
+      arrGuard(tsltMbyte, clogFileSize);
       sltLogPartPtr.p->logTailFileNo = 
          sltLogFilePtr.p->logLastPrepRef[tsltMbyte] >> 16;
 /* ------------------------------------------------------------------------- */
@@ -12402,6 +12437,26 @@ void Dblqh::execFSOPENCONF(Signal* signal)
   }//switch
 }//Dblqh::execFSOPENCONF()
 
+void
+Dblqh::execFSOPENREF(Signal* signal)
+{
+  jamEntry();
+  FsRef* ref = (FsRef*)signal->getDataPtr();
+  Uint32 err = ref->errorCode;
+  if (err == FsRef::fsErrInvalidFileSize)
+  {
+    char buf[256];
+    BaseString::snprintf(buf, sizeof(buf),
+                         "Invalid file size for redo logfile, "
+                         " size only changable with --initial");
+    progError(__LINE__,
+              NDBD_EXIT_INVALID_CONFIG,
+              buf);
+    return;
+  }
+
+  SimulatedBlock::execFSOPENREF(signal);
+}
 
 /* ************>> */
 /*  FSREADCONF  > */
@@ -13047,7 +13102,7 @@ void Dblqh::openFileInitLab(Signal* signal)
 {
   logFilePtr.p->logFileStatus = LogFileRecord::OPEN_INIT;
   seizeLogpage(signal);
-  writeSinglePage(signal, (ZNO_MBYTES_IN_FILE * ZPAGES_IN_MBYTE) - 1,
+  writeSinglePage(signal, (clogFileSize * ZPAGES_IN_MBYTE) - 1,
                   ZPAGE_SIZE - 1, __LINE__);
   lfoPtr.p->lfoState = LogFileOperationRecord::INIT_WRITE_AT_END;
   return;
@@ -13110,7 +13165,7 @@ void Dblqh::writeInitMbyteLab(Signal* signal)
 {
   releaseLfo(signal);
   logFilePtr.p->currentMbyte = logFilePtr.p->currentMbyte + 1;
-  if (logFilePtr.p->currentMbyte == ZNO_MBYTES_IN_FILE) {
+  if (logFilePtr.p->currentMbyte == clogFileSize) {
     jam();
     releaseLogpage(signal);
     logFilePtr.p->logFileStatus = LogFileRecord::CLOSING_INIT;
@@ -13230,7 +13285,7 @@ void Dblqh::initLogfile(Signal* signal, Uint32 fileNo)
   logFilePtr.p->lastPageWritten = 0;
   logFilePtr.p->logPageZero = RNIL;
   logFilePtr.p->currentMbyte = 0;
-  for (tilIndex = 0; tilIndex <= 15; tilIndex++) {
+  for (tilIndex = 0; tilIndex < clogFileSize; tilIndex++) {
     logFilePtr.p->logMaxGciCompleted[tilIndex] = (UintR)-1;
     logFilePtr.p->logMaxGciStarted[tilIndex] = (UintR)-1;
     logFilePtr.p->logLastPrepRef[tilIndex] = 0;
@@ -13281,8 +13336,14 @@ void Dblqh::openFileRw(Signal* signal, LogFileRecordPtr olfLogFilePtr)
   signal->theData[3] = olfLogFilePtr.p->fileName[1];
   signal->theData[4] = olfLogFilePtr.p->fileName[2];
   signal->theData[5] = olfLogFilePtr.p->fileName[3];
-  signal->theData[6] = ZOPEN_READ_WRITE | FsOpenReq::OM_AUTOSYNC;
+  signal->theData[6] = FsOpenReq::OM_READWRITE | FsOpenReq::OM_AUTOSYNC | FsOpenReq::OM_CHECK_SIZE;
+  if (c_o_direct)
+    signal->theData[6] |= FsOpenReq::OM_DIRECT;
   req->auto_sync_size = MAX_REDO_PAGES_WITHOUT_SYNCH * sizeof(LogPageRecord);
+  Uint64 sz = clogFileSize;
+  sz *= 1024; sz *= 1024;
+  req->file_size_hi = sz >> 32;
+  req->file_size_lo = sz & 0xFFFFFFFF;
   sendSignal(NDBFS_REF, GSN_FSOPENREQ, signal, FsOpenReq::SignalLength, JBA);
 }//Dblqh::openFileRw()
 
@@ -13301,7 +13362,9 @@ void Dblqh::openLogfileInit(Signal* signal)
   signal->theData[3] = logFilePtr.p->fileName[1];
   signal->theData[4] = logFilePtr.p->fileName[2];
   signal->theData[5] = logFilePtr.p->fileName[3];
-  signal->theData[6] = 0x302 | FsOpenReq::OM_AUTOSYNC;
+  signal->theData[6] = FsOpenReq::OM_READWRITE | FsOpenReq::OM_TRUNCATE | FsOpenReq::OM_CREATE | FsOpenReq::OM_AUTOSYNC;
+  if (c_o_direct)
+    signal->theData[6] |= FsOpenReq::OM_DIRECT;
   req->auto_sync_size = MAX_REDO_PAGES_WITHOUT_SYNCH * sizeof(LogPageRecord);
   sendSignal(NDBFS_REF, GSN_FSOPENREQ, signal, FsOpenReq::SignalLength, JBA);
 }//Dblqh::openLogfileInit()
@@ -13337,8 +13400,14 @@ void Dblqh::openNextLogfile(Signal* signal)
     signal->theData[3] = onlLogFilePtr.p->fileName[1];
     signal->theData[4] = onlLogFilePtr.p->fileName[2];
     signal->theData[5] = onlLogFilePtr.p->fileName[3];
-    signal->theData[6] = 2 | FsOpenReq::OM_AUTOSYNC;
+    signal->theData[6] = FsOpenReq::OM_READWRITE | FsOpenReq::OM_AUTOSYNC | FsOpenReq::OM_CHECK_SIZE;
+    if (c_o_direct)
+      signal->theData[6] |= FsOpenReq::OM_DIRECT;
     req->auto_sync_size = MAX_REDO_PAGES_WITHOUT_SYNCH * sizeof(LogPageRecord);
+    Uint64 sz = clogFileSize;
+    sz *= 1024; sz *= 1024;
+    req->file_size_hi = sz >> 32;
+    req->file_size_lo = sz & 0xFFFFFFFF;
     sendSignal(NDBFS_REF, GSN_FSOPENREQ, signal, FsOpenReq::SignalLength, JBA);
   }//if
 }//Dblqh::openNextLogfile()
@@ -13469,7 +13538,7 @@ void Dblqh::writeFileDescriptor(Signal* signal)
 /* -------------------------------------------------- */
 /*       START BY WRITING TO LOG FILE RECORD          */
 /* -------------------------------------------------- */
-  arrGuard(logFilePtr.p->currentMbyte, 16);
+  arrGuard(logFilePtr.p->currentMbyte, clogFileSize);
   logFilePtr.p->logMaxGciCompleted[logFilePtr.p->currentMbyte] = 
     logPartPtr.p->logPartNewestCompletedGCI;
   logFilePtr.p->logMaxGciStarted[logFilePtr.p->currentMbyte] = cnewestGci;
@@ -13495,10 +13564,7 @@ void Dblqh::writeFileDescriptor(Signal* signal)
 /* ------------------------------------------------------------------------- */
 void Dblqh::writeFileHeaderOpen(Signal* signal, Uint32 wmoType) 
 {
-  LogFileRecordPtr wmoLogFilePtr;
   UintR twmoNoLogDescriptors;
-  UintR twmoLoop;
-  UintR twmoIndex;
 
 /* -------------------------------------------------- */
 /*       WRITE HEADER INFORMATION IN THE NEW FILE.    */
@@ -13506,52 +13572,44 @@ void Dblqh::writeFileHeaderOpen(Signal* signal, Uint32 wmoType)
   logPagePtr.p->logPageWord[ZPAGE_HEADER_SIZE + ZPOS_LOG_TYPE] = ZFD_TYPE;
   logPagePtr.p->logPageWord[ZPAGE_HEADER_SIZE + ZPOS_FILE_NO] = 
     logFilePtr.p->fileNo;
-  if (logPartPtr.p->noLogFiles > ZMAX_LOG_FILES_IN_PAGE_ZERO) {
+  if (logPartPtr.p->noLogFiles > cmaxLogFilesInPageZero) {
     jam();
-    twmoNoLogDescriptors = ZMAX_LOG_FILES_IN_PAGE_ZERO;
+    twmoNoLogDescriptors = cmaxLogFilesInPageZero;
   } else {
     jam();
     twmoNoLogDescriptors = logPartPtr.p->noLogFiles;
   }//if
   logPagePtr.p->logPageWord[ZPAGE_HEADER_SIZE + ZPOS_NO_FD] = 
     twmoNoLogDescriptors;
-  wmoLogFilePtr.i = logFilePtr.i;
+
-  twmoLoop = 0;
+  {
-WMO_LOOP:
+    Uint32 pos = ZPAGE_HEADER_SIZE + ZFD_HEADER_SIZE;
+    LogFileRecordPtr filePtr = logFilePtr;
+    for (Uint32 fd = 0; fd < twmoNoLogDescriptors; fd++)
+    {
       jam();
-  if (twmoLoop < twmoNoLogDescriptors) {
+      ptrCheckGuard(filePtr, clogFileFileSize, logFileRecord);
+      for (Uint32 mb = 0; mb < clogFileSize; mb ++)
+      {
         jam();
-    ptrCheckGuard(wmoLogFilePtr, clogFileFileSize, logFileRecord);
+        Uint32 pos0 = pos + fd * (ZFD_MBYTE_SIZE * clogFileSize) + mb;
-    for (twmoIndex = 0; twmoIndex <= ZNO_MBYTES_IN_FILE - 1; twmoIndex++) {
+        Uint32 pos1 = pos0 + clogFileSize;
-      jam();
+        Uint32 pos2 = pos1 + clogFileSize;
-      arrGuard(((ZPAGE_HEADER_SIZE + ZFD_HEADER_SIZE) + 
+        arrGuard(pos0, ZPAGE_SIZE);
-                (twmoLoop * ZFD_PART_SIZE)) + twmoIndex, ZPAGE_SIZE);
+        arrGuard(pos1, ZPAGE_SIZE);
-      logPagePtr.p->logPageWord[((ZPAGE_HEADER_SIZE + ZFD_HEADER_SIZE) + 
+        arrGuard(pos2, ZPAGE_SIZE);
-                   (twmoLoop * ZFD_PART_SIZE)) + twmoIndex] = 
+        logPagePtr.p->logPageWord[pos0] = filePtr.p->logMaxGciCompleted[mb];
-            wmoLogFilePtr.p->logMaxGciCompleted[twmoIndex];
+        logPagePtr.p->logPageWord[pos1] = filePtr.p->logMaxGciStarted[mb];
-      arrGuard((((ZPAGE_HEADER_SIZE + ZFD_HEADER_SIZE) +
+        logPagePtr.p->logPageWord[pos2] = filePtr.p->logLastPrepRef[mb];
-                 (twmoLoop * ZFD_PART_SIZE)) + ZNO_MBYTES_IN_FILE) +
+      }
-                  twmoIndex, ZPAGE_SIZE);
+      filePtr.i = filePtr.p->prevLogFile;
-      logPagePtr.p->logPageWord[(((ZPAGE_HEADER_SIZE + ZFD_HEADER_SIZE) + 
+    }
-           (twmoLoop * ZFD_PART_SIZE)) + ZNO_MBYTES_IN_FILE) + twmoIndex] = 
+    pos += (twmoNoLogDescriptors * ZFD_MBYTE_SIZE * clogFileSize);
-         wmoLogFilePtr.p->logMaxGciStarted[twmoIndex];
+    arrGuard(pos, ZPAGE_SIZE);
-      arrGuard((((ZPAGE_HEADER_SIZE + ZFD_HEADER_SIZE) +
+    logPagePtr.p->logPageWord[ZCURR_PAGE_INDEX] = pos;
-        (twmoLoop * ZFD_PART_SIZE)) + (2 * ZNO_MBYTES_IN_FILE)) +
+    logPagePtr.p->logPageWord[pos] = ZNEXT_LOG_RECORD_TYPE;
-         twmoIndex, ZPAGE_SIZE);
+  }
-      logPagePtr.p->logPageWord[(((ZPAGE_HEADER_SIZE + ZFD_HEADER_SIZE) + 
+
-        (twmoLoop * ZFD_PART_SIZE)) + (2 * ZNO_MBYTES_IN_FILE)) + twmoIndex] = 
-          wmoLogFilePtr.p->logLastPrepRef[twmoIndex];
-    }//for
-    wmoLogFilePtr.i = wmoLogFilePtr.p->prevLogFile;
-    twmoLoop = twmoLoop + 1;
-    goto WMO_LOOP;
-  }//if
-  logPagePtr.p->logPageWord[ZCURR_PAGE_INDEX] = 
-    (ZPAGE_HEADER_SIZE + ZFD_HEADER_SIZE) +
-    (ZFD_PART_SIZE * twmoNoLogDescriptors);
-  arrGuard(logPagePtr.p->logPageWord[ZCURR_PAGE_INDEX], ZPAGE_SIZE);
-  logPagePtr.p->logPageWord[logPagePtr.p->logPageWord[ZCURR_PAGE_INDEX]] = 
-       ZNEXT_LOG_RECORD_TYPE;
 /* ------------------------------------------------------- */
 /*       THIS IS A SPECIAL WRITE OF THE FIRST PAGE IN THE  */
 /*       LOG FILE. THIS HAS SPECIAL SIGNIFANCE TO FIND     */
@@ -13696,9 +13754,9 @@ void Dblqh::openSrLastFileLab(Signal* signal)
 void Dblqh::readSrLastFileLab(Signal* signal) 
 {
   logPartPtr.p->logLap = logPagePtr.p->logPageWord[ZPOS_LOG_LAP];
-  if (logPartPtr.p->noLogFiles > ZMAX_LOG_FILES_IN_PAGE_ZERO) {
+  if (logPartPtr.p->noLogFiles > cmaxLogFilesInPageZero) {
     jam();
-    initGciInLogFileRec(signal, ZMAX_LOG_FILES_IN_PAGE_ZERO);
+    initGciInLogFileRec(signal, cmaxLogFilesInPageZero);
   } else {
     jam();
     initGciInLogFileRec(signal, logPartPtr.p->noLogFiles);
@@ -13723,7 +13781,7 @@ void Dblqh::readSrLastMbyteLab(Signal* signal)
       logPartPtr.p->lastMbyte = logFilePtr.p->currentMbyte - 1;
     }//if
   }//if
-  arrGuard(logFilePtr.p->currentMbyte, 16);
+  arrGuard(logFilePtr.p->currentMbyte, clogFileSize);
   logFilePtr.p->logMaxGciCompleted[logFilePtr.p->currentMbyte] = 
     logPagePtr.p->logPageWord[ZPOS_MAX_GCI_COMPLETED];
   logFilePtr.p->logMaxGciStarted[logFilePtr.p->currentMbyte] = 
@@ -13731,7 +13789,7 @@ void Dblqh::readSrLastMbyteLab(Signal* signal)
   logFilePtr.p->logLastPrepRef[logFilePtr.p->currentMbyte] = 
     logPagePtr.p->logPageWord[ZLAST_LOG_PREP_REF];
   releaseLogpage(signal);
-  if (logFilePtr.p->currentMbyte < (ZNO_MBYTES_IN_FILE - 1)) {
+  if (logFilePtr.p->currentMbyte < (clogFileSize - 1)) {
     jam();
     logFilePtr.p->currentMbyte++;
     readSinglePage(signal, ZPAGES_IN_MBYTE * logFilePtr.p->currentMbyte);
@@ -13745,21 +13803,21 @@ void Dblqh::readSrLastMbyteLab(Signal* signal)
      * ---------------------------------------------------------------------- */
     if (logPartPtr.p->lastMbyte == ZNIL) {
       jam();
-      logPartPtr.p->lastMbyte = ZNO_MBYTES_IN_FILE - 1;
+      logPartPtr.p->lastMbyte = clogFileSize - 1;
     }//if
   }//if
   logFilePtr.p->logFileStatus = LogFileRecord::CLOSING_SR;
   closeFile(signal, logFilePtr, __LINE__);
-  if (logPartPtr.p->noLogFiles > ZMAX_LOG_FILES_IN_PAGE_ZERO) {
+  if (logPartPtr.p->noLogFiles > cmaxLogFilesInPageZero) {
     Uint32 fileNo;
-    if (logFilePtr.p->fileNo >= ZMAX_LOG_FILES_IN_PAGE_ZERO) {
+    if (logFilePtr.p->fileNo >= cmaxLogFilesInPageZero) {
       jam();
-      fileNo = logFilePtr.p->fileNo - ZMAX_LOG_FILES_IN_PAGE_ZERO;
+      fileNo = logFilePtr.p->fileNo - cmaxLogFilesInPageZero;
     } else {
       jam();
       fileNo = 
 	(logPartPtr.p->noLogFiles + logFilePtr.p->fileNo) - 
-	ZMAX_LOG_FILES_IN_PAGE_ZERO;
+	cmaxLogFilesInPageZero;
     }//if
     if (fileNo == 0) {
       jam();
@@ -13769,11 +13827,11 @@ void Dblqh::readSrLastMbyteLab(Signal* signal)
        * -------------------------------------------------------------------- */
       fileNo = 1;
       logPartPtr.p->srRemainingFiles = 
-	logPartPtr.p->noLogFiles - (ZMAX_LOG_FILES_IN_PAGE_ZERO - 1);
+	logPartPtr.p->noLogFiles - (cmaxLogFilesInPageZero - 1);
     } else {
       jam();
       logPartPtr.p->srRemainingFiles = 
-	logPartPtr.p->noLogFiles - ZMAX_LOG_FILES_IN_PAGE_ZERO;
+	logPartPtr.p->noLogFiles - cmaxLogFilesInPageZero;
     }//if
     LogFileRecordPtr locLogFilePtr;
     findLogfile(signal, fileNo, logPartPtr, &locLogFilePtr);
@@ -13798,9 +13856,9 @@ void Dblqh::openSrNextFileLab(Signal* signal)
 
 void Dblqh::readSrNextFileLab(Signal* signal) 
 {
-  if (logPartPtr.p->srRemainingFiles > ZMAX_LOG_FILES_IN_PAGE_ZERO) {
+  if (logPartPtr.p->srRemainingFiles > cmaxLogFilesInPageZero) {
     jam();
-    initGciInLogFileRec(signal, ZMAX_LOG_FILES_IN_PAGE_ZERO);
+    initGciInLogFileRec(signal, cmaxLogFilesInPageZero);
   } else {
     jam();
     initGciInLogFileRec(signal, logPartPtr.p->srRemainingFiles);
@@ -13808,16 +13866,16 @@ void Dblqh::readSrNextFileLab(Signal* signal)
   releaseLogpage(signal);
   logFilePtr.p->logFileStatus = LogFileRecord::CLOSING_SR;
   closeFile(signal, logFilePtr, __LINE__);
-  if (logPartPtr.p->srRemainingFiles > ZMAX_LOG_FILES_IN_PAGE_ZERO) {
+  if (logPartPtr.p->srRemainingFiles > cmaxLogFilesInPageZero) {
     Uint32 fileNo;
-    if (logFilePtr.p->fileNo >= ZMAX_LOG_FILES_IN_PAGE_ZERO) {
+    if (logFilePtr.p->fileNo >= cmaxLogFilesInPageZero) {
       jam();
-      fileNo = logFilePtr.p->fileNo - ZMAX_LOG_FILES_IN_PAGE_ZERO;
+      fileNo = logFilePtr.p->fileNo - cmaxLogFilesInPageZero;
     } else {
       jam();
       fileNo = 
 	(logPartPtr.p->noLogFiles + logFilePtr.p->fileNo) - 
-	ZMAX_LOG_FILES_IN_PAGE_ZERO;
+	cmaxLogFilesInPageZero;
     }//if
     if (fileNo == 0) {
       jam();
@@ -13826,11 +13884,11 @@ void Dblqh::readSrNextFileLab(Signal* signal)
        * -------------------------------------------------------------------- */
       fileNo = 1;
       logPartPtr.p->srRemainingFiles = 
-	logPartPtr.p->srRemainingFiles - (ZMAX_LOG_FILES_IN_PAGE_ZERO - 1);
+	logPartPtr.p->srRemainingFiles - (cmaxLogFilesInPageZero - 1);
     } else {
       jam();
       logPartPtr.p->srRemainingFiles = 
-	logPartPtr.p->srRemainingFiles - ZMAX_LOG_FILES_IN_PAGE_ZERO;
+	logPartPtr.p->srRemainingFiles - cmaxLogFilesInPageZero;
     }//if
     LogFileRecordPtr locLogFilePtr;
     findLogfile(signal, fileNo, logPartPtr, &locLogFilePtr);
@@ -14701,7 +14759,7 @@ void Dblqh::srLogLimits(Signal* signal)
    *   EXECUTED.
    * ----------------------------------------------------------------------- */
   while(true) {
-    ndbrequire(tmbyte < 16);
+    ndbrequire(tmbyte < clogFileSize);
     if (logPartPtr.p->logExecState == LogPartRecord::LES_SEARCH_STOP) {
       if (logFilePtr.p->logMaxGciCompleted[tmbyte] < logPartPtr.p->logLastGci) {
         jam();
@@ -14742,7 +14800,7 @@ void Dblqh::srLogLimits(Signal* signal)
     if (logPartPtr.p->logExecState != LogPartRecord::LES_EXEC_LOG) {
       if (tmbyte == 0) {
         jam();
-        tmbyte = ZNO_MBYTES_IN_FILE - 1;
+        tmbyte = clogFileSize - 1;
         logFilePtr.i = logFilePtr.p->prevLogFile;
         ptrCheckGuard(logFilePtr, clogFileFileSize, logFileRecord);
       } else {
@@ -15136,7 +15194,7 @@ void Dblqh::execSr(Signal* signal)
 	  logPagePtr.p->logPageWord[ZPAGE_HEADER_SIZE + ZPOS_NO_FD];
           logPagePtr.p->logPageWord[ZCURR_PAGE_INDEX] = 
 	      (ZPAGE_HEADER_SIZE + ZFD_HEADER_SIZE) + 
-	      (noFdDescriptors * ZFD_PART_SIZE);
+	      (noFdDescriptors * ZFD_MBYTE_SIZE * clogFileSize);
       }
       break;
 /* ========================================================================= */
@@ -15176,11 +15234,11 @@ void Dblqh::execSr(Signal* signal)
 /*---------------------------------------------------------------------------*/
 /* START EXECUTION OF A NEW MBYTE IN THE LOG.                                */
 /*---------------------------------------------------------------------------*/
-      if (logFilePtr.p->currentMbyte < (ZNO_MBYTES_IN_FILE - 1)) {
+      if (logFilePtr.p->currentMbyte < (clogFileSize - 1)) {
         jam();
         logPartPtr.p->logExecState = LogPartRecord::LES_EXEC_LOG_NEW_MBYTE;
       } else {
-        ndbrequire(logFilePtr.p->currentMbyte == (ZNO_MBYTES_IN_FILE - 1));
+        ndbrequire(logFilePtr.p->currentMbyte == (clogFileSize - 1));
         jam();
 /*---------------------------------------------------------------------------*/
 /* WE HAVE TO CHANGE FILE. CLOSE THIS ONE AND THEN OPEN THE NEXT.            */
@@ -15375,7 +15433,7 @@ void Dblqh::invalidateLogAfterLastGCI(Signal* signal) {
     jam();
     releaseLfo(signal);
     releaseLogpage(signal); 
-    if (logPartPtr.p->invalidatePageNo < (ZNO_MBYTES_IN_FILE * ZPAGES_IN_MBYTE - 1)) {
+    if (logPartPtr.p->invalidatePageNo < (clogFileSize * ZPAGES_IN_MBYTE - 1)) {
       // We continue in this file.
       logPartPtr.p->invalidatePageNo++;
     } else {
@@ -16716,6 +16774,22 @@ void Dblqh::initialiseLogFile(Signal* signal)
       ptrAss(logFilePtr, logFileRecord);
       logFilePtr.p->nextLogFile = logFilePtr.i + 1;
       logFilePtr.p->logFileStatus = LogFileRecord::LFS_IDLE;
+
+      logFilePtr.p->logLastPrepRef = new Uint32[clogFileSize];
+      logFilePtr.p->logMaxGciCompleted = new Uint32[clogFileSize];
+      logFilePtr.p->logMaxGciStarted = new Uint32[clogFileSize];
+
+      if (logFilePtr.p->logLastPrepRef == 0 ||
+          logFilePtr.p->logMaxGciCompleted == 0 ||
+          logFilePtr.p->logMaxGciStarted == 0)
+      {
+        char buf[256];
+        BaseString::snprintf(buf, sizeof(buf),
+                             "Failed to alloc mbyte(%u) arrays for logfile %u",
+                             clogFileSize, logFilePtr.i);
+        progError(__LINE__, NDBD_EXIT_MEMALLOC, buf);
+      }
+
     }//for
     logFilePtr.i = clogFileFileSize - 1;
     ptrAss(logFilePtr, logFileRecord);
@@ -17044,41 +17118,31 @@ void Dblqh::initFragrec(Signal* signal,
  * ========================================================================= */
 void Dblqh::initGciInLogFileRec(Signal* signal, Uint32 noFdDescriptors) 
 {
-  LogFileRecordPtr iglLogFilePtr;
+  LogFileRecordPtr filePtr = logFilePtr;
-  UintR tiglLoop;
+  Uint32 pos = ZPAGE_HEADER_SIZE + ZFD_HEADER_SIZE;
-  UintR tiglIndex;
+  for (Uint32 fd = 0; fd < noFdDescriptors; fd++)
-
+  {
-  tiglLoop = 0;
-  iglLogFilePtr.i = logFilePtr.i;
-  iglLogFilePtr.p = logFilePtr.p;
-IGL_LOOP:
-  for (tiglIndex = 0; tiglIndex <= ZNO_MBYTES_IN_FILE - 1; tiglIndex++) {
-    arrGuard(((ZPAGE_HEADER_SIZE + ZFD_HEADER_SIZE) + 
-	      (tiglLoop * ZFD_PART_SIZE)) + tiglIndex, ZPAGE_SIZE);
-    iglLogFilePtr.p->logMaxGciCompleted[tiglIndex] = 
-      logPagePtr.p->logPageWord[((ZPAGE_HEADER_SIZE + ZFD_HEADER_SIZE) + 
-				 (tiglLoop * ZFD_PART_SIZE)) + tiglIndex];
-    arrGuard((((ZPAGE_HEADER_SIZE + ZFD_HEADER_SIZE) + ZNO_MBYTES_IN_FILE) + 
-	      (tiglLoop * ZFD_PART_SIZE)) + tiglIndex, ZPAGE_SIZE);
-    iglLogFilePtr.p->logMaxGciStarted[tiglIndex] = 
-      logPagePtr.p->logPageWord[(((ZPAGE_HEADER_SIZE + ZFD_HEADER_SIZE) + 
-				  ZNO_MBYTES_IN_FILE) + 
-				 (tiglLoop * ZFD_PART_SIZE)) + tiglIndex];
-    arrGuard((((ZPAGE_HEADER_SIZE + ZFD_HEADER_SIZE) + 
-	       (2 * ZNO_MBYTES_IN_FILE)) + (tiglLoop * ZFD_PART_SIZE)) + 
-	     tiglIndex, ZPAGE_SIZE);
-    iglLogFilePtr.p->logLastPrepRef[tiglIndex] = 
-      logPagePtr.p->logPageWord[(((ZPAGE_HEADER_SIZE + ZFD_HEADER_SIZE) + 
-				  (2 * ZNO_MBYTES_IN_FILE)) + 
-				 (tiglLoop * ZFD_PART_SIZE)) + tiglIndex];
-  }//for
-  tiglLoop = tiglLoop + 1;
-  if (tiglLoop < noFdDescriptors) {
     jam();
-    iglLogFilePtr.i = iglLogFilePtr.p->prevLogFile;
+    for (Uint32 mb = 0; mb < clogFileSize; mb++)
-    ptrCheckGuard(iglLogFilePtr, clogFileFileSize, logFileRecord);
+    {
-    goto IGL_LOOP;
+      jam();
-  }//if
+      Uint32 pos0 = pos + fd * (ZFD_MBYTE_SIZE * clogFileSize) + mb;
+      Uint32 pos1 = pos0 + clogFileSize;
+      Uint32 pos2 = pos1 + clogFileSize;
+      arrGuard(pos0, ZPAGE_SIZE);
+      arrGuard(pos1, ZPAGE_SIZE);
+      arrGuard(pos2, ZPAGE_SIZE);
+      filePtr.p->logMaxGciCompleted[mb] = logPagePtr.p->logPageWord[pos0];
+      filePtr.p->logMaxGciStarted[mb] = logPagePtr.p->logPageWord[pos1];
+      filePtr.p->logLastPrepRef[mb] = logPagePtr.p->logPageWord[pos2];
+    }
+    if (fd + 1 < noFdDescriptors)
+    {
+      jam();
+      filePtr.i = filePtr.p->prevLogFile;
+      ptrCheckGuard(filePtr, clogFileFileSize, logFileRecord);
+    }
+  }
 }//Dblqh::initGciInLogFileRec()
 
 /* ========================================================================== 
@@ -18331,7 +18395,7 @@ void Dblqh::writeNextLog(Signal* signal)
   ndbrequire(logPagePtr.p->logPageWord[ZCURR_PAGE_INDEX] < ZPAGE_SIZE);
   logPagePtr.p->logPageWord[logPagePtr.p->logPageWord[ZCURR_PAGE_INDEX]] = 
     ZNEXT_MBYTE_TYPE;
-  if (logFilePtr.p->currentMbyte == (ZNO_MBYTES_IN_FILE - 1)) {
+  if (logFilePtr.p->currentMbyte == (clogFileSize - 1)) {
     jam();
 /* -------------------------------------------------- */
 /*       CALCULATE THE NEW REMAINING WORDS WHEN       */
@@ -18420,7 +18484,7 @@ void Dblqh::writeNextLog(Signal* signal)
       systemError(signal, __LINE__);
     }//if
   }//if
-  if (logFilePtr.p->currentMbyte == (ZNO_MBYTES_IN_FILE - 1)) {
+  if (logFilePtr.p->currentMbyte == (clogFileSize - 1)) {
     jam();
     twnlNextMbyte = 0;
     if (logFilePtr.p->fileChangeState != LogFileRecord::NOT_ONGOING) {

storage/ndb/src/kernel/blocks/dblqh/Makefile.am

@@ -16,7 +16,7 @@
 EXTRA_PROGRAMS = ndbd_redo_log_reader
 
 ndbd_redo_log_reader_SOURCES = redoLogReader/records.cpp \
-  redoLogReader/redoLogFileReader.cpp
+  redoLogReader/reader.cpp
 
 include $(top_srcdir)/storage/ndb/config/common.mk.am
 include $(top_srcdir)/storage/ndb/config/type_kernel.mk.am

storage/ndb/src/kernel/blocks/ndbcntr/NdbcntrMain.cpp

@@ -277,6 +277,14 @@ void Ndbcntr::execSTTOR(Signal* signal)
     break;
   case ZSTART_PHASE_1:
     jam();
+    {
+      Uint32 db_watchdog_interval = 0;
+      const ndb_mgm_configuration_iterator * p = 
+        m_ctx.m_config.getOwnConfigIterator();
+      ndb_mgm_get_int_parameter(p, CFG_DB_WATCHDOG_INTERVAL, &db_watchdog_interval);
+      ndbrequire(db_watchdog_interval);
+      update_watch_dog_timer(db_watchdog_interval);
+    }
     startPhase1Lab(signal);
     break;
   case ZSTART_PHASE_2:
@@ -1410,6 +1418,13 @@ void Ndbcntr::execNODE_FAILREP(Signal* signal)
 {
   jamEntry();
 
+  if (ERROR_INSERTED(1001))
+  {
+    sendSignalWithDelay(reference(), GSN_NODE_FAILREP, signal, 100, 
+                        signal->getLength());
+    return;
+  }
+  
   const NodeFailRep * nodeFail = (NodeFailRep *)&signal->theData[0];
   NdbNodeBitmask allFailed; 
   allFailed.assign(NdbNodeBitmask::Size, nodeFail->theNodes);
@@ -2734,16 +2749,34 @@ void Ndbcntr::execSTART_ORD(Signal* signal){
   c_missra.execSTART_ORD(signal);
 }
 
+#define CLEAR_DX 13
+#define CLEAR_LCP 3
+
 void
-Ndbcntr::clearFilesystem(Signal* signal){
+Ndbcntr::clearFilesystem(Signal* signal)
+{
+  const Uint32 lcp = c_fsRemoveCount >= CLEAR_DX;
+  
   FsRemoveReq * req  = (FsRemoveReq *)signal->getDataPtrSend();
   req->userReference = reference();
   req->userPointer   = 0;
   req->directory     = 1;
   req->ownDirectory  = 1;
+
+  if (lcp == 0)
+  {
     FsOpenReq::setVersion(req->fileNumber, 3);
     FsOpenReq::setSuffix(req->fileNumber, FsOpenReq::S_CTL); // Can by any...
     FsOpenReq::v1_setDisk(req->fileNumber, c_fsRemoveCount);
+  }
+  else
+  {
+    FsOpenReq::setVersion(req->fileNumber, 5);
+    FsOpenReq::setSuffix(req->fileNumber, FsOpenReq::S_DATA);
+    FsOpenReq::v5_setLcpNo(req->fileNumber, c_fsRemoveCount - CLEAR_DX);
+    FsOpenReq::v5_setTableId(req->fileNumber, 0);
+    FsOpenReq::v5_setFragmentId(req->fileNumber, 0);
+  }
   sendSignal(NDBFS_REF, GSN_FSREMOVEREQ, signal, 
              FsRemoveReq::SignalLength, JBA);
   c_fsRemoveCount++;
@@ -2752,12 +2785,12 @@ Ndbcntr::clearFilesystem(Signal* signal){
 void
 Ndbcntr::execFSREMOVECONF(Signal* signal){
   jamEntry();
-  if(c_fsRemoveCount == 13){
+  if(c_fsRemoveCount == CLEAR_DX + CLEAR_LCP){
     jam();
     sendSttorry(signal);
   } else {
     jam();
-    ndbrequire(c_fsRemoveCount < 13);
+    ndbrequire(c_fsRemoveCount < CLEAR_DX + CLEAR_LCP);
     clearFilesystem(signal);
   }//if
 }

storage/ndb/src/kernel/blocks/ndbfs/AsyncFile.cpp

@@ -163,7 +163,12 @@ AsyncFile::run()
   theStartFlag = true;
   // Create write buffer for bigger writes
   theWriteBufferSize = WRITEBUFFERSIZE;
-  theWriteBuffer = (char *) ndbd_malloc(theWriteBufferSize); 
+  theWriteBufferUnaligned = (char *) ndbd_malloc(theWriteBufferSize +
+                                                 NDB_O_DIRECT_WRITE_ALIGNMENT-1);
+  theWriteBuffer = (char *)
+    (((UintPtr)theWriteBufferUnaligned + NDB_O_DIRECT_WRITE_ALIGNMENT - 1) &
+     ~(UintPtr)(NDB_O_DIRECT_WRITE_ALIGNMENT - 1));
+
   NdbMutex_Unlock(theStartMutexPtr);
   NdbCondition_Signal(theStartConditionPtr);
   
@@ -247,6 +252,78 @@ AsyncFile::run()
 static char g_odirect_readbuf[2*GLOBAL_PAGE_SIZE -1];
 #endif
 
+int
+AsyncFile::check_odirect_write(Uint32 flags, int& new_flags, int mode)
+{
+  assert(new_flags & (O_CREAT | O_TRUNC));
+#ifdef O_DIRECT
+  int ret;
+  char * bufptr = (char*)((UintPtr(g_odirect_readbuf)+(GLOBAL_PAGE_SIZE - 1)) & ~(GLOBAL_PAGE_SIZE - 1));
+  while (((ret = ::write(theFd, bufptr, GLOBAL_PAGE_SIZE)) == -1) && 
+         (errno == EINTR));
+  if (ret == -1)
+  {
+    new_flags &= ~O_DIRECT;
+    ndbout_c("%s Failed to write using O_DIRECT, disabling", 
+             theFileName.c_str());
+  }
+  
+  close(theFd);
+  theFd = ::open(theFileName.c_str(), new_flags, mode);
+  if (theFd == -1)
+    return errno;
+#endif
+
+  return 0;
+}
+
+int
+AsyncFile::check_odirect_read(Uint32 flags, int &new_flags, int mode)
+{
+#ifdef O_DIRECT
+  int ret;
+  char * bufptr = (char*)((UintPtr(g_odirect_readbuf)+(GLOBAL_PAGE_SIZE - 1)) & ~(GLOBAL_PAGE_SIZE - 1));
+  while (((ret = ::read(theFd, bufptr, GLOBAL_PAGE_SIZE)) == -1) && 
+         (errno == EINTR));
+  if (ret == -1)
+  {
+    ndbout_c("%s Failed to read using O_DIRECT, disabling", 
+             theFileName.c_str());
+    goto reopen;
+  }
+  
+  if(lseek(theFd, 0, SEEK_SET) != 0)
+  {
+    return errno;
+  }
+  
+  if ((flags & FsOpenReq::OM_CHECK_SIZE) == 0)
+  {
+    struct stat buf;
+    if ((fstat(theFd, &buf) == -1))
+    {
+      return errno;
+    } 
+    else if ((buf.st_size % GLOBAL_PAGE_SIZE) != 0)
+    {
+      ndbout_c("%s filesize not a multiple of %d, disabling O_DIRECT", 
+               theFileName.c_str(), GLOBAL_PAGE_SIZE);
+      goto reopen;
+    }
+  }
+  
+  return 0;
+  
+reopen:
+  close(theFd);
+  new_flags &= ~O_DIRECT;
+  theFd = ::open(theFileName.c_str(), new_flags, mode);
+  if (theFd == -1)
+    return errno;  
+#endif
+  return 0;
+}
+
 void AsyncFile::openReq(Request* request)
 {  
   m_auto_sync_freq = 0;
@@ -312,7 +389,7 @@ void AsyncFile::openReq(Request* request)
   }
 #else
   Uint32 flags = request->par.open.flags;
-  Uint32 new_flags = 0;
+  int new_flags = 0;
 
   // Convert file open flags from Solaris to Liux
   if (flags & FsOpenReq::OM_CREATE)
@@ -343,10 +420,6 @@ void AsyncFile::openReq(Request* request)
   {
     new_flags |= O_DIRECT;
   }
-#elif defined O_SYNC
-  {
-    flags |= FsOpenReq::OM_SYNC;
-  }
 #endif
   
   if ((flags & FsOpenReq::OM_SYNC) && ! (flags & FsOpenReq::OM_INIT))
@@ -356,14 +429,18 @@ void AsyncFile::openReq(Request* request)
 #endif
   }
 
+  const char * rw = "";
   switch(flags & 0x3){
   case FsOpenReq::OM_READONLY:
+    rw = "r";
     new_flags |= O_RDONLY;
     break;
   case FsOpenReq::OM_WRITEONLY:
+    rw = "w";
     new_flags |= O_WRONLY;
     break;
   case FsOpenReq::OM_READWRITE:
+    rw = "rw";
     new_flags |= O_RDWR;
     break;
   default:
@@ -404,11 +481,6 @@ no_odirect:
 	if (new_flags & O_DIRECT)
 	{
 	  new_flags &= ~O_DIRECT;
-	  flags |= FsOpenReq::OM_SYNC;
-#ifdef O_SYNC
-	  if (! (flags & FsOpenReq::OM_INIT))
-	    new_flags |= O_SYNC;
-#endif
 	  goto no_odirect;
 	}
 #endif
@@ -421,11 +493,6 @@ no_odirect:
     else if (new_flags & O_DIRECT)
     {
       new_flags &= ~O_DIRECT;
-      flags |= FsOpenReq::OM_SYNC;
-#ifdef O_SYNC
-      if (! (flags & FsOpenReq::OM_INIT))
-	new_flags |= O_SYNC;
-#endif
       goto no_odirect;
     }
 #endif
@@ -512,7 +579,6 @@ no_odirect:
 	{
 	  ndbout_c("error on first write(%d), disable O_DIRECT", err);
 	  new_flags &= ~O_DIRECT;
-	  flags |= FsOpenReq::OM_SYNC;
 	  close(theFd);
 	  theFd = ::open(theFileName.c_str(), new_flags, mode);
 	  if (theFd != -1)
@@ -532,26 +598,32 @@ no_odirect:
   else if (flags & FsOpenReq::OM_DIRECT)
   {
 #ifdef O_DIRECT
-    do {
+    if (flags & (FsOpenReq::OM_TRUNCATE | FsOpenReq::OM_CREATE))
-      int ret;
-      char * bufptr = (char*)((UintPtr(g_odirect_readbuf)+(GLOBAL_PAGE_SIZE - 1)) & ~(GLOBAL_PAGE_SIZE - 1));
-      while (((ret = ::read(theFd, bufptr, GLOBAL_PAGE_SIZE)) == -1) && (errno == EINTR));
-      if (ret == -1)
     {
-	ndbout_c("%s Failed to read using O_DIRECT, disabling", theFileName.c_str());
+      request->error = check_odirect_write(flags, new_flags, mode);
-	flags |= FsOpenReq::OM_SYNC;
-	flags |= FsOpenReq::OM_INIT;
-	break;
     }
-      if(lseek(theFd, 0, SEEK_SET) != 0)
+    else
     {
-	request->error = errno;
+      request->error = check_odirect_read(flags, new_flags, mode);
-	return;
-      }
-    } while (0);
-#endif
     }
     
+    if (request->error)
+      return;
+#endif
+  }
+#ifdef VM_TRACE
+  if (flags & FsOpenReq::OM_DIRECT)
+  {
+#ifdef O_DIRECT
+    ndbout_c("%s %s O_DIRECT: %d",
+             theFileName.c_str(), rw,
+             !!(new_flags & O_DIRECT));
+#else
+    ndbout_c("%s %s O_DIRECT: 0",
+             theFileName.c_str(), rw);
+#endif
+  }
+#endif  
   if ((flags & FsOpenReq::OM_SYNC) && (flags & FsOpenReq::OM_INIT))
   {
 #ifdef O_SYNC
@@ -562,6 +634,10 @@ no_odirect:
     new_flags &= ~(O_CREAT | O_TRUNC);
     new_flags |= O_SYNC;
     theFd = ::open(theFileName.c_str(), new_flags, mode);
+    if (theFd == -1)
+    {
+      request->error = errno;
+    }
 #endif
   }
 #endif
@@ -1079,7 +1155,8 @@ AsyncFile::rmrfReq(Request * request, char * path, bool removePath){
 void AsyncFile::endReq()
 {
   // Thread is ended with return
-  if (theWriteBuffer) ndbd_free(theWriteBuffer, theWriteBufferSize);
+  if (theWriteBufferUnaligned)
+    ndbd_free(theWriteBufferUnaligned, theWriteBufferSize);
 }
 
 

storage/ndb/src/kernel/blocks/ndbfs/AsyncFile.hpp

@@ -234,9 +234,13 @@ private:
   bool   theStartFlag;
   int theWriteBufferSize;
   char* theWriteBuffer;
+  void* theWriteBufferUnaligned;
   
   size_t m_write_wo_sync;  // Writes wo/ sync
   size_t m_auto_sync_freq; // Auto sync freq in bytes
+
+  int check_odirect_read(Uint32 flags, int&new_flags, int mode);
+  int check_odirect_write(Uint32 flags, int&new_flags, int mode);
 public:
   SimulatedBlock& m_fs;
   Ptr<GlobalPage> m_page_ptr;

storage/ndb/src/kernel/blocks/ndbfs/Ndbfs.cpp

@@ -652,7 +652,7 @@ AsyncFile*
 Ndbfs::createAsyncFile(){
 
   // Check limit of open files
-  if (m_maxFiles !=0 && theFiles.size()+1 ==  m_maxFiles) {
+  if (m_maxFiles !=0 && theFiles.size() ==  m_maxFiles) {
     // Print info about all open files
     for (unsigned i = 0; i < theFiles.size(); i++){
       AsyncFile* file = theFiles[i];

storage/ndb/src/kernel/blocks/pgman.cpp

@@ -123,8 +123,8 @@ Pgman::execREAD_CONFIG_REQ(Signal* signal)
   if (page_buffer > 0)
   {
     page_buffer /= GLOBAL_PAGE_SIZE; // in pages
-    m_page_entry_pool.setSize(100*page_buffer);
     m_param.m_max_pages = page_buffer;
+    m_page_entry_pool.setSize(m_param.m_lirs_stack_mult * page_buffer);
     m_param.m_max_hot_pages = (page_buffer * 9) / 10;
   }
 
@@ -141,6 +141,7 @@ Pgman::execREAD_CONFIG_REQ(Signal* signal)
 
 Pgman::Param::Param() :
   m_max_pages(64),      // smallish for testing
+  m_lirs_stack_mult(10),
   m_max_hot_pages(56),
   m_max_loop_count(256),
   m_max_io_waits(64),
@@ -301,6 +302,9 @@ Pgman::get_sublist_no(Page_state state)
   {
     return Page_entry::SL_LOCKED;
   }
+  if (state == Page_entry::ONSTACK) {
+    return Page_entry::SL_IDLE;
+  }
   return Page_entry::SL_OTHER;
 }
 
@@ -415,15 +419,55 @@ Pgman::get_page_entry(Ptr<Page_entry>& ptr, Uint32 file_no, Uint32 page_no)
 {
   if (find_page_entry(ptr, file_no, page_no))
   {
+    jam();
     ndbrequire(ptr.p->m_state != 0);
     m_stats.m_page_hits++;
+
+#ifdef VM_TRACE
+  debugOut << "PGMAN: get_page_entry: found" << endl;
+  debugOut << "PGMAN: " << ptr << endl;
+#endif
     return true;
   }
 
+  if (m_page_entry_pool.getNoOfFree() == 0)
+  {
+    jam();
+    Page_sublist& pl_idle = *m_page_sublist[Page_entry::SL_IDLE];
+    Ptr<Page_entry> idle_ptr;
+    if (pl_idle.first(idle_ptr))
+    {
+      jam();
+
+#ifdef VM_TRACE
+    debugOut << "PGMAN: get_page_entry: re-use idle entry" << endl;
+    debugOut << "PGMAN: " << idle_ptr << endl;
+#endif
+
+      Page_state state = idle_ptr.p->m_state;
+      ndbrequire(state == Page_entry::ONSTACK);
+
+      Page_stack& pl_stack = m_page_stack;
+      ndbrequire(pl_stack.hasPrev(idle_ptr));
+      pl_stack.remove(idle_ptr);
+      state &= ~ Page_entry::ONSTACK;
+      set_page_state(idle_ptr, state);
+      ndbrequire(idle_ptr.p->m_state == 0);
+
+      release_page_entry(idle_ptr);
+    }
+  }
+
   if (seize_page_entry(ptr, file_no, page_no))
   {
+    jam();
     ndbrequire(ptr.p->m_state == 0);
     m_stats.m_page_faults++;
+
+#ifdef VM_TRACE
+  debugOut << "PGMAN: get_page_entry: seize" << endl;
+  debugOut << "PGMAN: " << ptr << endl;
+#endif
     return true;
   }
 
@@ -624,6 +668,7 @@ Pgman::lirs_reference(Ptr<Page_entry> ptr)
         jam();
         move_cleanup_ptr(ptr);
         pl_queue.remove(ptr);
+        state &= ~ Page_entry::ONQUEUE;
       }
       if (state & Page_entry::BOUND)
       {
@@ -654,6 +699,12 @@ Pgman::lirs_reference(Ptr<Page_entry> ptr)
     pl_stack.add(ptr);
     state |= Page_entry::ONSTACK;
     state |= Page_entry::HOT;
+    // it could be on queue already
+    if (state & Page_entry::ONQUEUE) {
+      jam();
+      pl_queue.remove(ptr);
+      state &= ~Page_entry::ONQUEUE;
+    }
   }
 
   set_page_state(ptr, state);
@@ -902,9 +953,11 @@ Pgman::process_map(Signal* signal)
 #ifdef VM_TRACE
   debugOut << "PGMAN: >process_map" << endl;
 #endif
-  int max_count = m_param.m_max_io_waits - m_stats.m_current_io_waits;
+  int max_count = 0;
-  if (max_count > 0)
+  if (m_param.m_max_io_waits > m_stats.m_current_io_waits) {
+    max_count = m_param.m_max_io_waits - m_stats.m_current_io_waits;
     max_count = max_count / 2 + 1;
+  }
   Page_sublist& pl_map = *m_page_sublist[Page_entry::SL_MAP];
 
   while (! pl_map.isEmpty() && --max_count >= 0)
@@ -1056,15 +1109,10 @@ Pgman::process_cleanup(Signal* signal)
   }
 
   int max_loop_count = m_param.m_max_loop_count;
-  int max_count = m_param.m_max_io_waits - m_stats.m_current_io_waits;
+  int max_count = 0;
-
+  if (m_param.m_max_io_waits > m_stats.m_current_io_waits) {
-  if (max_count > 0)
+    max_count = m_param.m_max_io_waits - m_stats.m_current_io_waits;
-  {
     max_count = max_count / 2 + 1;
-    /*
-     * Possibly add code here to avoid writing too rapidly.  May be
-     * unnecessary since only cold pages are cleaned.
-     */
   }
 
   Ptr<Page_entry> ptr = m_cleanup_ptr;
@@ -1166,9 +1214,12 @@ bool
 Pgman::process_lcp(Signal* signal)
 {
   Page_hashlist& pl_hash = m_page_hashlist;
-  int max_count = m_param.m_max_io_waits - m_stats.m_current_io_waits;
+
-  if (max_count > 0)
+  int max_count = 0;
+  if (m_param.m_max_io_waits > m_stats.m_current_io_waits) {
+    max_count = m_param.m_max_io_waits - m_stats.m_current_io_waits;
     max_count = max_count / 2 + 1;
+  }
 
 #ifdef VM_TRACE
   debugOut
@@ -1927,6 +1978,8 @@ Pgman::verify_page_entry(Ptr<Page_entry> ptr)
     break;
   case Page_entry::SL_LOCKED:
     break;
+  case Page_entry::SL_IDLE:
+    break;
   case Page_entry::SL_OTHER:
     break;
   default:
@@ -1973,8 +2026,11 @@ Pgman::verify_page_lists()
   ndbrequire(stack_count == pl_stack.count() || dump_page_lists());
   ndbrequire(queue_count == pl_queue.count() || dump_page_lists());
 
+  Uint32 hot_count = 0;
   Uint32 hot_bound_count = 0;
   Uint32 cold_bound_count = 0;
+  Uint32 stack_request_count = 0;
+  Uint32 queue_request_count = 0;
 
   Uint32 i1 = RNIL;
   for (pl_stack.first(ptr); ptr.i != RNIL; pl_stack.next(ptr))
@@ -1985,10 +2041,14 @@ Pgman::verify_page_lists()
     ndbrequire(state & Page_entry::ONSTACK || dump_page_lists());
     if (! pl_stack.hasPrev(ptr))
       ndbrequire(state & Page_entry::HOT || dump_page_lists());
-    if (state & Page_entry::HOT &&
+    if (state & Page_entry::HOT) {
-        state & Page_entry::BOUND)
+      hot_count++;
+      if (state & Page_entry::BOUND)
         hot_bound_count++;
     }
+    if (state & Page_entry::REQUEST)
+      stack_request_count++;
+  }
 
   Uint32 i2 = RNIL;
   for (pl_queue.first(ptr); ptr.i != RNIL; pl_queue.next(ptr))
@@ -1999,6 +2059,8 @@ Pgman::verify_page_lists()
     ndbrequire(state & Page_entry::ONQUEUE || dump_page_lists());
     ndbrequire(state & Page_entry::BOUND || dump_page_lists());
     cold_bound_count++;
+    if (state & Page_entry::REQUEST)
+      queue_request_count++;
   }
 
   Uint32 tot_bound_count =
@@ -2031,7 +2093,11 @@ Pgman::verify_page_lists()
            << " cache:" << m_stats.m_num_pages
            << "(" << locked_bound_count << "L)"
            << " stack:" << pl_stack.count()
+           << " hot:" << hot_count
+           << " hot_bound:" << hot_bound_count
+           << " stack_request:" << stack_request_count
            << " queue:" << pl_queue.count()
+           << " queue_request:" << queue_request_count
            << " queuewait:" << queuewait_count << endl;
 
   debugOut << "PGMAN:";
@@ -2139,6 +2205,8 @@ Pgman::get_sublist_name(Uint32 list_no)
     return "busy";
   case Page_entry::SL_LOCKED:
     return "locked";
+  case Page_entry::SL_IDLE:
+    return "idle";
   case Page_entry::SL_OTHER:
     return "other";
   }

storage/ndb/src/kernel/blocks/pgman.hpp

@@ -325,8 +325,9 @@ private:
       ,SL_CALLBACK_IO = 4
       ,SL_BUSY = 5
       ,SL_LOCKED = 6
-      ,SL_OTHER = 7
+      ,SL_IDLE = 7
-      ,SUBLIST_COUNT = 8
+      ,SL_OTHER = 8
+      ,SUBLIST_COUNT = 9
     };
 
     Uint16 m_file_no;       // disk page address set at seize
@@ -401,6 +402,7 @@ private:
   struct Param {
     Param();
     Uint32 m_max_pages;         // max number of cache pages
+    Uint32 m_lirs_stack_mult;   // in m_max_pages (around 3-10)
     Uint32 m_max_hot_pages;     // max hot cache pages (up to 99%)
     Uint32 m_max_loop_count;    // limit purely local loops
     Uint32 m_max_io_waits;

storage/ndb/src/kernel/blocks/restore.cpp

@@ -557,6 +557,9 @@ Restore::restore_next(Signal* signal, FilePtr file_ptr)
       case BackupFormat::GCP_ENTRY:
 	parse_gcp_entry(signal, file_ptr, data, len);
 	break;
+      case BackupFormat::EMPTY_ENTRY:
+        // skip
+        break;
       case 0x4e444242: // 'NDBB'
 	if (check_file_version(signal, ntohl(* (data+2))) == 0)
 	{

storage/ndb/src/kernel/vm/Configuration.cpp

@@ -443,6 +443,11 @@ Configuration::setupConfiguration(){
 	      "TimeBetweenWatchDogCheck missing");
   }
 
+  if(iter.get(CFG_DB_WATCHDOG_INTERVAL_INITIAL, &_timeBetweenWatchDogCheckInitial)){
+    ERROR_SET(fatal, NDBD_EXIT_INVALID_CONFIG, "Invalid configuration fetched", 
+	      "TimeBetweenWatchDogCheckInitial missing");
+  }
+
   /**
    * Get paths
    */  
@@ -462,9 +467,12 @@ Configuration::setupConfiguration(){
    * Create the watch dog thread
    */
   { 
-    Uint32 t = _timeBetweenWatchDogCheck;
+    if (_timeBetweenWatchDogCheckInitial < _timeBetweenWatchDogCheck)
+      _timeBetweenWatchDogCheckInitial = _timeBetweenWatchDogCheck;
+
+    Uint32 t = _timeBetweenWatchDogCheckInitial;
     t = globalEmulatorData.theWatchDog ->setCheckInterval(t);
-    _timeBetweenWatchDogCheck = t;
+    _timeBetweenWatchDogCheckInitial = t;
   }
   
   ConfigValues* cf = ConfigValuesFactory::extractCurrentSection(iter.m_config);

storage/ndb/src/kernel/vm/Configuration.hpp

@@ -84,6 +84,7 @@ private:
   Uint32 _maxErrorLogs;
   Uint32 _lockPagesInMainMemory;
   Uint32 _timeBetweenWatchDogCheck;
+  Uint32 _timeBetweenWatchDogCheckInitial;
 
   ndb_mgm_configuration * m_ownConfig;
   ndb_mgm_configuration * m_clusterConfig;

storage/ndb/src/kernel/vm/SimulatedBlock.cpp

@@ -19,6 +19,7 @@
 #include <NdbOut.hpp>
 #include <GlobalData.hpp>
 #include <Emulator.hpp>
+#include <WatchDog.hpp>
 #include <ErrorHandlingMacros.hpp>
 #include <TimeQueue.hpp>
 #include <TransporterRegistry.hpp>
@@ -38,6 +39,9 @@
 #include <AttributeDescriptor.hpp>
 #include <NdbSqlUtil.hpp>
 
+#include <EventLogger.hpp>
+extern EventLogger g_eventLogger;
+
 #define ljamEntry() jamEntryLine(30000 + __LINE__)
 #define ljam() jamLine(30000 + __LINE__)
 
@@ -655,14 +659,20 @@ SimulatedBlock::getBatSize(Uint16 blockNo){
   return sb->theBATSize;
 }
 
+void* SimulatedBlock::allocRecord(const char * type, size_t s, size_t n, bool clear, Uint32 paramId)
+{
+  return allocRecordAligned(type, s, n, 0, 0, clear, paramId);
+}
+
 void* 
-SimulatedBlock::allocRecord(const char * type, size_t s, size_t n, bool clear, Uint32 paramId) 
+SimulatedBlock::allocRecordAligned(const char * type, size_t s, size_t n, void **unaligned_buffer, Uint32 align, bool clear, Uint32 paramId)
 {
 
   void * p = NULL;
-  size_t size = n*s;
+  Uint32 over_alloc = unaligned_buffer ? (align - 1) : 0;
-  Uint64 real_size = (Uint64)((Uint64)n)*((Uint64)s);
+  size_t size = n*s + over_alloc;
-  refresh_watch_dog(); 
+  Uint64 real_size = (Uint64)((Uint64)n)*((Uint64)s) + over_alloc;
+  refresh_watch_dog(9);
   if (real_size > 0){
 #ifdef VM_TRACE_MEM
     ndbout_c("%s::allocRecord(%s, %u, %u) = %llu bytes", 
@@ -696,14 +706,24 @@ SimulatedBlock::allocRecord(const char * type, size_t s, size_t n, bool clear, U
       char * ptr = (char*)p;
       const Uint32 chunk = 128 * 1024;
       while(size > chunk){
-	refresh_watch_dog(); 
+	refresh_watch_dog(9);
 	memset(ptr, 0, chunk);
 	ptr += chunk;
 	size -= chunk;
       }
-      refresh_watch_dog(); 
+      refresh_watch_dog(9);
       memset(ptr, 0, size);
     }
+    if (unaligned_buffer)
+    {
+      *unaligned_buffer = p;
+      p = (void *)(((UintPtr)p + over_alloc) & ~(UintPtr)(over_alloc));
+#ifdef VM_TRACE
+      g_eventLogger.info("'%s' (%u) %llu %llu, alignment correction %u bytes",
+                         type, align, (Uint64)p, (Uint64)p+n*s,
+                         (Uint32)((UintPtr)p - (UintPtr)*unaligned_buffer));
+#endif
+    }
   }
   return p;
 }
@@ -720,9 +740,16 @@ SimulatedBlock::deallocRecord(void ** ptr,
 }
 
 void
-SimulatedBlock::refresh_watch_dog()
+SimulatedBlock::refresh_watch_dog(Uint32 place)
 {
-  globalData.incrementWatchDogCounter(1);
+  globalData.incrementWatchDogCounter(place);
+}
+
+void
+SimulatedBlock::update_watch_dog_timer(Uint32 interval)
+{
+  extern EmulatorData globalEmulatorData;
+  globalEmulatorData.theWatchDog->setCheckInterval(interval);
 }
 
 void

storage/ndb/src/kernel/vm/SimulatedBlock.hpp

@@ -334,7 +334,8 @@ protected:
    * Refresh Watch Dog in initialising code
    *
    */
-  void refresh_watch_dog();
+  void refresh_watch_dog(Uint32 place = 1);
+  void update_watch_dog_timer(Uint32 interval);
 
   /**
    * Prog error
@@ -377,6 +378,7 @@ protected:
    *
    */
   void* allocRecord(const char * type, size_t s, size_t n, bool clear = true, Uint32 paramId = 0);
+  void* allocRecordAligned(const char * type, size_t s, size_t n, void **unaligned_buffer, Uint32 align = NDB_O_DIRECT_WRITE_ALIGNMENT, bool clear = true, Uint32 paramId = 0);
   
   /**
    * Deallocate record

storage/ndb/src/kernel/vm/WatchDog.cpp

@@ -16,6 +16,7 @@
 
 #include <ndb_global.h>
 #include <my_pthread.h>
+#include <sys/times.h>
 
 #include "WatchDog.hpp"
 #include "GlobalData.hpp"
@@ -24,6 +25,8 @@
 #include <ErrorHandlingMacros.hpp>
 #include <EventLogger.hpp>
 
+#include <NdbTick.h>
+
 extern EventLogger g_eventLogger;
 
 extern "C" 
@@ -71,69 +74,118 @@ WatchDog::doStop(){
   }
 }
 
-void 
+const char *get_action(Uint32 IPValue)
-WatchDog::run(){
+{
-  unsigned int anIPValue;
+  const char *action;
-  unsigned int alerts = 0;
+  switch (IPValue) {
-  unsigned int oldIPValue = 0;
+  case 1:
-  
+    action = "Job Handling";
-  // WatchDog for the single threaded NDB
+    break;
-  while(!theStop){
+  case 2:
-    Uint32 tmp  = theInterval / 500;
+    action = "Scanning Timers";
-    tmp= (tmp ? tmp : 1);
+    break;
-    
+  case 3:
-    while(!theStop && tmp > 0){
+    action = "External I/O";
-      NdbSleep_MilliSleep(500);
+    break;
-      tmp--;
+  case 4:
+    action = "Print Job Buffers at crash";
+    break;
+  case 5:
+    action = "Checking connections";
+    break;
+  case 6:
+    action = "Performing Send";
+    break;
+  case 7:
+    action = "Polling for Receive";
+    break;
+  case 8:
+    action = "Performing Receive";
+    break;
+  case 9:
+    action = "Allocating memory";
+    break;
+  default:
+    action = "Unknown place";
+    break;
+  }//switch
+  return action;
 }
 
+void 
+WatchDog::run()
+{
+  unsigned int anIPValue, sleep_time;
+  unsigned int oldIPValue = 0;
+  unsigned int theIntervalCheck = theInterval;
+  struct MicroSecondTimer start_time, last_time, now;
+  NdbTick_getMicroTimer(&start_time);
+  last_time = start_time;
+
+  // WatchDog for the single threaded NDB
+  while (!theStop)
+  {
+    sleep_time= 100;
+
+    NdbSleep_MilliSleep(sleep_time);
     if(theStop)
       break;
 
+    NdbTick_getMicroTimer(&now);
+    if (NdbTick_getMicrosPassed(last_time, now)/1000 > sleep_time*2)
+    {
+      struct tms my_tms;
+      times(&my_tms);
+      g_eventLogger.info("Watchdog: User time: %llu  System time: %llu",
+                         (Uint64)my_tms.tms_utime,
+                         (Uint64)my_tms.tms_stime);
+      g_eventLogger.warning("Watchdog: Warning overslept %u ms, expected %u ms.",
+                            NdbTick_getMicrosPassed(last_time, now)/1000,
+                            sleep_time);
+    }
+    last_time = now;
+
     // Verify that the IP thread is not stuck in a loop
     anIPValue = *theIPValue;
-    if(anIPValue != 0) {
+    if (anIPValue != 0)
+    {
       oldIPValue = anIPValue;
       globalData.incrementWatchDogCounter(0);
-      alerts = 0;
+      NdbTick_getMicroTimer(&start_time);
-    } else {
+      theIntervalCheck = theInterval;
-      const char *last_stuck_action;
+    }
-      alerts++;
+    else
-      switch (oldIPValue) {
+    {
-      case 1:
+      int warn = 1;
-        last_stuck_action = "Job Handling";
+      Uint32 elapsed = NdbTick_getMicrosPassed(start_time, now)/1000;
-        break;
+      /*
-      case 2:
+        oldIPValue == 9 indicates malloc going on, this can take some time
-        last_stuck_action = "Scanning Timers";
+        so only warn if we pass the watchdog interval
-        break;
+      */
-      case 3:
+      if (oldIPValue == 9)
-        last_stuck_action = "External I/O";
+        if (elapsed < theIntervalCheck)
-        break;
+          warn = 0;
-      case 4:
+        else
-        last_stuck_action = "Print Job Buffers at crash";
+          theIntervalCheck += theInterval;
-        break;
+
-      case 5:
+      if (warn)
-        last_stuck_action = "Checking connections";
+      {
-        break;
+        const char *last_stuck_action = get_action(oldIPValue);
-      case 6:
-        last_stuck_action = "Performing Send";
-        break;
-      case 7:
-        last_stuck_action = "Polling for Receive";
-        break;
-      case 8:
-        last_stuck_action = "Performing Receive";
-        break;
-      default:
-        last_stuck_action = "Unknown place";
-        break;
-      }//switch
         g_eventLogger.warning("Ndb kernel is stuck in: %s", last_stuck_action);
-      if(alerts == 3){
+        {
+          struct tms my_tms;
+          times(&my_tms);
+          g_eventLogger.info("Watchdog: User time: %llu  System time: %llu",
+                             (Uint64)my_tms.tms_utime,
+                             (Uint64)my_tms.tms_stime);
+        }
+        if (elapsed > 3 * theInterval)
+        {
           shutdownSystem(last_stuck_action);
         }
       }
     }
+  }
   return;
 }
 

storage/ndb/src/mgmsrv/ConfigInfo.cpp

@@ -579,6 +579,18 @@ const ConfigInfo::ParamInfo ConfigInfo::m_ParamInfo[] = {
     "70",
     STR_VALUE(MAX_INT_RNIL) },
 
+  {
+    CFG_DB_WATCHDOG_INTERVAL_INITIAL,
+    "TimeBetweenWatchDogCheckInitial",
+    DB_TOKEN,
+    "Time between execution checks inside a database node in the early start phases when memory is allocated",
+    ConfigInfo::CI_USED,
+    true,
+    ConfigInfo::CI_INT,
+    "6000",
+    "70",
+    STR_VALUE(MAX_INT_RNIL) },
+
   {
     CFG_DB_STOP_ON_ERROR,
     "StopOnError",
@@ -879,6 +891,18 @@ const ConfigInfo::ParamInfo ConfigInfo::m_ParamInfo[] = {
     "3",
     STR_VALUE(MAX_INT_RNIL) },
 
+  {
+    CFG_DB_REDOLOG_FILE_SIZE,
+    "FragmentLogFileSize",
+    DB_TOKEN,
+    "Size of each Redo log file",
+    ConfigInfo::CI_USED,
+    false,
+    ConfigInfo::CI_INT,
+    "16M",
+    "4M",
+    "1G" },
+
   {
     CFG_DB_MAX_OPEN_FILES,
     "MaxNoOfOpenFiles",
@@ -1309,6 +1333,18 @@ const ConfigInfo::ParamInfo ConfigInfo::m_ParamInfo[] = {
     "0",
     STR_VALUE(MAX_INT_RNIL) },
   
+  {
+    CFG_DB_O_DIRECT,
+    "ODirect",
+    DB_TOKEN,
+    "Use O_DIRECT file write/read when possible",
+    ConfigInfo::CI_USED,
+    true,
+    ConfigInfo::CI_BOOL,
+    "false",
+    "false",
+    "true"},
+
   /***************************************************************************
    * API
    ***************************************************************************/

storage/ndb/src/ndbapi/ndberror.c

@@ -179,7 +179,7 @@ ErrorBundle ErrorCodes[] = {
   { 873,  DMEC, TR, "Out of attrinfo records for scan in tuple manager" },
   { 899,  DMEC, TR, "Rowid already allocated" },
   { 1217, DMEC, TR, "Out of operation records in local data manager (increase MaxNoOfLocalOperations)" },
-  { 1220, DMEC, TR, "REDO log files overloaded, consult online manual (decrease TimeBetweenLocalCheckpoints, and|or increase NoOfFragmentLogFiles)" },
+  { 1220, DMEC, TR, "REDO log files overloaded, consult online manual (increase FragmentLogFileSize)" },
   { 1222, DMEC, TR, "Out of transaction markers in LQH" },
   { 4021, DMEC, TR, "Out of Send Buffer space in NDB API" },
   { 4022, DMEC, TR, "Out of Send Buffer space in NDB API" },

storage/ndb/test/ndbapi/testNodeRestart.cpp

@@ -1633,6 +1633,85 @@ runBug28023(NDBT_Context* ctx, NDBT_Step* step)
   return NDBT_OK;
 }
 
+
+int
+runBug28717(NDBT_Context* ctx, NDBT_Step* step)
+{
+  int result = NDBT_OK;
+  int loops = ctx->getNumLoops();
+  int records = ctx->getNumRecords();
+  Ndb* pNdb = GETNDB(step);
+  NdbRestarter res;
+
+  if (res.getNumDbNodes() < 4)
+  {
+    return NDBT_OK;
+  }
+
+  int master = res.getMasterNodeId();
+  int node0 = res.getRandomNodeOtherNodeGroup(master, rand());
+  int node1 = res.getRandomNodeSameNodeGroup(node0, rand());
+  
+  ndbout_c("master: %d node0: %d node1: %d", master, node0, node1);
+  
+  if (res.restartOneDbNode(node0, false, true, true))
+  {
+    return NDBT_FAILED;
+  }
+
+  {
+    int filter[] = { 15, NDB_MGM_EVENT_CATEGORY_CHECKPOINT, 0 };
+    NdbLogEventHandle handle = 
+      ndb_mgm_create_logevent_handle(res.handle, filter);
+    
+
+    int dump[] = { DumpStateOrd::DihStartLcpImmediately };
+    struct ndb_logevent event;
+    
+    for (Uint32 i = 0; i<3; i++)
+    {
+      res.dumpStateOneNode(master, dump, 1);
+      while(ndb_logevent_get_next(handle, &event, 0) >= 0 &&
+            event.type != NDB_LE_LocalCheckpointStarted);
+      while(ndb_logevent_get_next(handle, &event, 0) >= 0 &&
+            event.type != NDB_LE_LocalCheckpointCompleted);
+    } 
+  }
+  
+  if (res.waitNodesNoStart(&node0, 1))
+    return NDBT_FAILED;
+  
+  int val2[] = { DumpStateOrd::CmvmiSetRestartOnErrorInsert, 1 };
+  
+  if (res.dumpStateOneNode(node0, val2, 2))
+    return NDBT_FAILED;
+  
+  if (res.insertErrorInNode(node0, 5010))
+    return NDBT_FAILED;
+  
+  if (res.insertErrorInNode(node1, 1001))
+    return NDBT_FAILED;
+  
+  if (res.startNodes(&node0, 1))
+    return NDBT_FAILED;
+  
+  NdbSleep_SecSleep(3);
+
+  if (res.insertErrorInNode(node1, 0))
+    return NDBT_FAILED;
+
+  if (res.waitNodesNoStart(&node0, 1))
+    return NDBT_FAILED;
+
+  if (res.startNodes(&node0, 1))
+    return NDBT_FAILED;
+
+  if (res.waitClusterStarted())
+    return NDBT_FAILED;
+  
+  return NDBT_OK;
+}
+
 NDBT_TESTSUITE(testNodeRestart);
 TESTCASE("NoLoad", 
 	 "Test that one node at a time can be stopped and then restarted "\
@@ -1993,6 +2072,9 @@ TESTCASE("Bug27466", ""){
 TESTCASE("Bug28023", ""){
   INITIALIZER(runBug28023);
 }
+TESTCASE("Bug28717", ""){
+  INITIALIZER(runBug28717);
+}
 NDBT_TESTSUITE_END(testNodeRestart);
 
 int main(int argc, const char** argv){

storage/ndb/test/run-test/conf-dl145a.cnf

@@ -21,3 +21,6 @@ BackupMemory = 64M
 MaxNoOfConcurrentScans = 100
 MaxNoOfSavedMessages= 1000
 SendBufferMemory = 2M
+NoOfFragmentLogFiles = 4
+FragmentLogFileSize = 64M
+

storage/ndb/test/run-test/daily-basic-tests.txt

@@ -567,6 +567,10 @@ max-time: 1500
 cmd: testDict
 args: -n CreateAndDrop 
 
+max-time: 1000
+cmd: testNodeRestart
+args: -n Bug28717 T1
+
 max-time: 1500
 cmd: testDict
 args: -n CreateAndDropAtRandom -l 200 T1
@@ -706,7 +710,7 @@ args: -n ExecuteAsynch T1
 
 max-time: 1000
 cmd: testNdbApi
-args: -n BugBug28443
+args: -n Bug28443
 
 #max-time: 500
 #cmd: testInterpreter

storage/ndb/tools/restore/Restore.cpp

@@ -873,13 +873,32 @@ bool RestoreDataIterator::readFragmentHeader(int & ret, Uint32 *fragmentId)
   
   debug << "RestoreDataIterator::getNextFragment" << endl;
   
-  if (buffer_read(&Header, sizeof(Header), 1) != 1){
+  while (1)
+  {
+    /* read first part of header */
+    if (buffer_read(&Header, 8, 1) != 1)
+    {
       ret = 0;
       return false;
     } // if
 
+    /* skip if EMPTY_ENTRY */
     Header.SectionType  = ntohl(Header.SectionType);
     Header.SectionLength  = ntohl(Header.SectionLength);
+    if (Header.SectionType == BackupFormat::EMPTY_ENTRY)
+    {
+      void *tmp;
+      buffer_get_ptr(&tmp, Header.SectionLength*4-8, 1);
+      continue;
+    }
+    break;
+  }
+  /* read rest of header */
+  if (buffer_read(((char*)&Header)+8, sizeof(Header)-8, 1) != 1)
+  {
+    ret = 0;
+    return false;
+  }
   Header.TableId  = ntohl(Header.TableId);
   Header.FragmentNo  = ntohl(Header.FragmentNo);
   Header.ChecksumType  = ntohl(Header.ChecksumType);

unittest/Makefile.am

@@ -25,3 +25,6 @@ test:
 
 test-verbose:
 	HARNESS_VERBOSE=1 perl unit.pl run $(unittests)
+
+# Don't update the files from bitkeeper
+%::SCCS/s.%

unittest/examples/Makefile.am

@@ -22,3 +22,5 @@ LDADD              = -lmytap
 
 noinst_PROGRAMS    = simple-t skip-t todo-t skip_all-t no_plan-t core-t
 
+# Don't update the files from bitkeeper
+%::SCCS/s.%

unittest/mysys/Makefile.am

@@ -23,3 +23,5 @@ LDADD 		= $(top_builddir)/unittest/mytap/libmytap.a \
 
 noinst_PROGRAMS  = bitmap-t base64-t my_atomic-t
 
+# Don't update the files from bitkeeper
+%::SCCS/s.%

unittest/mytap/Makefile.am

@@ -21,3 +21,6 @@ noinst_HEADERS     = tap.h
 libmytap_a_SOURCES = tap.c
 
 SUBDIRS            = . t
+
+# Don't update the files from bitkeeper
+%::SCCS/s.%

unittest/mytap/t/Makefile.am

@@ -21,3 +21,5 @@ LDADD           = -lmytap
 
 noinst_PROGRAMS = basic-t
 
+# Don't update the files from bitkeeper
+%::SCCS/s.%