Merge lp:~akopytov/percona-server/i26611-bug1059812-5.5 into lp:percona-server/5.5

[Alexey Kopytov]

[Alexey Kopytov (akopytov)]

http://jenkins.percona.com/view/PS%205.5/job/percona-server-5.5-param/558/

[Laurynas Biveinis (laurynas-biveinis)]

Alexey -

It's really uncommon to see my_bool instead of ibool outside of handler/ and to lesser extent srv/. I would contain it to these two dirs by changing srv0start.c to

buf_pool_init (... srv_buf_pool_populate ? TRUE : FALSE, ... );

and making the rest of changes use ibool. I think that this horrible one line is still better than mixing bool types in the lower InnoDB subsystems.

The 2.6.23 version check is something I'd rather not see, but brief checking seems to show that this indeed the only way to find out if MAP_PRIVATE | MAP_POPULATE works :(

review: Needs Fixing

[Jeremy Cole (jeremycole)]

BTW, it seems that the get_time/set_time changes in sql/sp_head.cc and sql/sql_class.h are not related to the main bug being fixed here.

[Davi Arnaut (davi)]

Should also take this bug fix: https://github.com/twitter/mysql/commit/9fa2236c15b80b3d8abdf464087689c58b9c6364

review: Needs Fixing

[Alexey Kopytov (akopytov)]

Merged the bug fix, thanks!

[Davi Arnaut (davi)]

The my_bool/ibool cleanup suggestion is good, will take it.

[Laurynas Biveinis (laurynas-biveinis)]

review: Approve

=== modified file 'Percona-Server/mysql-test/r/percona_server_variables_debug.result'
--- Percona-Server/mysql-test/r/percona_server_variables_debug.result	2012-10-12 06:24:13 +0000
+++ Percona-Server/mysql-test/r/percona_server_variables_debug.result	2012-10-18 18:03:21 +0000
@@ -88,6 +88,7 @@
 INNODB_AUTOINC_LOCK_MODE
 INNODB_BLOCKING_BUFFER_POOL_RESTORE
 INNODB_BUFFER_POOL_INSTANCES
+INNODB_BUFFER_POOL_POPULATE
 INNODB_BUFFER_POOL_RESTORE_AT_STARTUP
 INNODB_BUFFER_POOL_SHM_CHECKSUM
 INNODB_BUFFER_POOL_SHM_KEY

=== modified file 'Percona-Server/mysql-test/r/percona_server_variables_release.result'
--- Percona-Server/mysql-test/r/percona_server_variables_release.result	2012-09-28 01:36:45 +0000
+++ Percona-Server/mysql-test/r/percona_server_variables_release.result	2012-10-18 18:03:21 +0000
@@ -86,6 +86,7 @@
 INNODB_AUTOINC_LOCK_MODE
 INNODB_BLOCKING_BUFFER_POOL_RESTORE
 INNODB_BUFFER_POOL_INSTANCES
+INNODB_BUFFER_POOL_POPULATE
 INNODB_BUFFER_POOL_RESTORE_AT_STARTUP
 INNODB_BUFFER_POOL_SHM_CHECKSUM
 INNODB_BUFFER_POOL_SHM_KEY

=== added file 'Percona-Server/mysql-test/suite/sys_vars/r/innodb_buffer_pool_populate_basic.result'
--- Percona-Server/mysql-test/suite/sys_vars/r/innodb_buffer_pool_populate_basic.result	1970-01-01 00:00:00 +0000
+++ Percona-Server/mysql-test/suite/sys_vars/r/innodb_buffer_pool_populate_basic.result	2012-10-18 18:03:21 +0000
@@ -0,0 +1,12 @@
+CALL mtr.add_suppression(".* Forcing preallocation by faulting in pages.");
+SELECT @@GLOBAL.innodb_buffer_pool_populate;
+@@GLOBAL.innodb_buffer_pool_populate
+1
+1 Expected
+SET @@GLOBAL.innodb_buffer_pool_populate=0;
+ERROR HY000: Variable 'innodb_buffer_pool_populate' is a read only variable
+Expected error 'Read only variable'
+SELECT @@GLOBAL.innodb_buffer_pool_populate;
+@@GLOBAL.innodb_buffer_pool_populate
+1
+1 Expected

=== added file 'Percona-Server/mysql-test/suite/sys_vars/t/innodb_buffer_pool_populate_basic-master.opt'
--- Percona-Server/mysql-test/suite/sys_vars/t/innodb_buffer_pool_populate_basic-master.opt	1970-01-01 00:00:00 +0000
+++ Percona-Server/mysql-test/suite/sys_vars/t/innodb_buffer_pool_populate_basic-master.opt	2012-10-18 18:03:21 +0000
@@ -0,0 +1,1 @@
+--innodb-buffer-pool-populate=true

=== added file 'Percona-Server/mysql-test/suite/sys_vars/t/innodb_buffer_pool_populate_basic.test'
--- Percona-Server/mysql-test/suite/sys_vars/t/innodb_buffer_pool_populate_basic.test	1970-01-01 00:00:00 +0000
+++ Percona-Server/mysql-test/suite/sys_vars/t/innodb_buffer_pool_populate_basic.test	2012-10-18 18:03:21 +0000
@@ -0,0 +1,16 @@
+--source include/have_innodb.inc
+
+CALL mtr.add_suppression(".* Forcing preallocation by faulting in pages.");
+
+# Display current value of innodb_use_sys_malloc
+SELECT @@GLOBAL.innodb_buffer_pool_populate;
+--echo 1 Expected
+
+# Variable should be read-only
+--error ER_INCORRECT_GLOBAL_LOCAL_VAR
+SET @@GLOBAL.innodb_buffer_pool_populate=0;
+--echo Expected error 'Read only variable'
+
+SELECT @@GLOBAL.innodb_buffer_pool_populate;
+--echo 1 Expected
+

=== modified file 'Percona-Server/scripts/mysqld_safe.sh'
--- Percona-Server/scripts/mysqld_safe.sh	2011-11-10 13:46:51 +0000
+++ Percona-Server/scripts/mysqld_safe.sh	2012-10-18 18:03:21 +0000
@@ -17,6 +17,8 @@
 niceness=0
 mysqld_ld_preload=
 mysqld_ld_library_path=
+flush_caches=0
+numa_interleave=0
 
 # Initial logging status: error log is not open, and not using syslog
 logging=init
@@ -60,6 +62,10 @@
   --syslog                   Log messages to syslog with 'logger'
   --skip-syslog              Log messages to error log (default)
   --syslog-tag=TAG           Pass -t "mysqld-TAG" to 'logger'
+  --flush-caches             Flush and purge buffers/caches before
+                             starting the server
+  --numa-interleave          Run mysqld with its memory interleaved
+                             on all NUMA nodes
 
 All other options are passed to the mysqld program.
 
@@ -205,6 +211,8 @@
       --skip-syslog) want_syslog=0 ;;
       --syslog-tag=*) syslog_tag="$val" ;;
       --timezone=*) TZ="$val"; export TZ; ;;
+      --flush-caches) flush_caches=1 ;;
+      --numa-interleave) numa_interleave=1 ;;
 
       --help) usage ;;
 
@@ -711,6 +719,41 @@
 fi
 
 #
+# Flush and purge buffers/caches.
+#
+
+if @TARGET_LINUX@ && test $flush_caches -eq 1
+then
+  # Locate sync, ensure it exists.
+  if ! my_which sync > /dev/null 2>&1
+  then
+    log_error "sync command not found, required for --flush-caches"
+    exit 1
+  # Flush file system buffers.
+  elif ! sync
+  then
+    # Huh, the sync() function is always successful...
+    log_error "sync failed, check if sync is properly installed"
+  fi
+
+  # Locate sysctl, ensure it exists.
+  if ! my_which sysctl > /dev/null 2>&1
+  then
+    log_error "sysctl command not found, required for --flush-caches"
+    exit 1
+  # Purge page cache, dentries and inodes.
+  elif ! sysctl -q -w vm.drop_caches=3
+  then
+    log_error "sysctl failed, check the error message for details"
+    exit 1
+  fi
+elif test $flush_caches -eq 1
+then
+  log_error "--flush-caches is not supported on this platform"
+  exit 1
+fi
+
+#
 # Uncomment the following lines if you want all tables to be automatically
 # checked and repaired during startup. You should add sensible key_buffer
 # and sort_buffer values to my.cnf to improve check performance or require
@@ -730,6 +773,31 @@
 
 cmd="`mysqld_ld_preload_text`$NOHUP_NICENESS"
 
+#
+# Set mysqld's memory interleave policy.
+#
+
+if @TARGET_LINUX@ && test $numa_interleave -eq 1
+then
+  # Locate numactl, ensure it exists.
+  if ! my_which numactl > /dev/null 2>&1
+  then
+    log_error "numactl command not found, required for --numa-interleave"
+    exit 1
+  # Attempt to run a command, ensure it works.
+  elif ! numactl --interleave=all true
+  then
+    log_error "numactl failed, check if numactl is properly installed"
+  fi
+
+  # Launch mysqld with numactl.
+  cmd="$cmd numactl --interleave=all"
+elif test $numa_interleave -eq 1
+then
+  log_error "--numa-interleave is not supported on this platform"
+  exit 1
+fi
+
 for i in  "$ledir/$MYSQLD" "$defaults" "--basedir=$MY_BASEDIR_VERSION" \
   "--datadir=$DATADIR" "--plugin-dir=$plugin_dir" "$USER_OPTION"
 do

=== modified file 'Percona-Server/storage/innobase/buf/buf0buf.c'
--- Percona-Server/storage/innobase/buf/buf0buf.c	2012-10-12 06:24:13 +0000
+++ Percona-Server/storage/innobase/buf/buf0buf.c	2012-10-18 18:03:21 +0000
@@ -1028,7 +1028,8 @@
 /*===========*/
 	buf_pool_t*	buf_pool,	/*!< in: buffer pool instance */
 	buf_chunk_t*	chunk,		/*!< out: chunk of buffers */
-	ulint		mem_size)	/*!< in: requested size in bytes */
+	ulint		mem_size,	/*!< in: requested size in bytes */
+	ibool		populate)	/*!< in: virtual page preallocation */
 {
 	buf_block_t*	block;
 	byte*		frame;
@@ -1044,7 +1045,7 @@
 				  + (UNIV_PAGE_SIZE - 1), UNIV_PAGE_SIZE);
 
 	chunk->mem_size = mem_size;
-	chunk->mem = os_mem_alloc_large(&chunk->mem_size);
+	chunk->mem = os_mem_alloc_large(&chunk->mem_size, populate);
 
 	if (UNIV_UNLIKELY(chunk->mem == NULL)) {
 
@@ -1254,6 +1255,7 @@
 /*===================*/
 	buf_pool_t*	buf_pool,	/*!< in: buffer pool instance */
 	ulint		buf_pool_size,	/*!< in: size in bytes */
+	ibool		populate,	/*!< in: virtual page preallocation */
 	ulint		instance_no)	/*!< in: id of the instance */
 {
 	ulint		i;
@@ -1286,7 +1288,7 @@
 
 		UT_LIST_INIT(buf_pool->free);
 
-		if (!buf_chunk_init(buf_pool, chunk, buf_pool_size)) {
+		if (!buf_chunk_init(buf_pool, chunk, buf_pool_size, populate)) {
 			mem_free(chunk);
 			mem_free(buf_pool);
 
@@ -1381,6 +1383,7 @@
 buf_pool_init(
 /*==========*/
 	ulint	total_size,	/*!< in: size of the total pool in bytes */
+	ibool	populate,	/*!< in: virtual page preallocation */
 	ulint	n_instances)	/*!< in: number of instances */
 {
 	ulint		i;
@@ -1398,7 +1401,7 @@
 	for (i = 0; i < n_instances; i++) {
 		buf_pool_t*	ptr	= &buf_pool_ptr[i];
 
-		if (buf_pool_init_instance(ptr, size, i) != DB_SUCCESS) {
+		if (buf_pool_init_instance(ptr, size, populate, i) != DB_SUCCESS) {
 
 			/* Free all the instances created so far. */
 			buf_pool_free(i);

=== modified file 'Percona-Server/storage/innobase/handler/ha_innodb.cc'
--- Percona-Server/storage/innobase/handler/ha_innodb.cc	2012-10-01 00:49:13 +0000
+++ Percona-Server/storage/innobase/handler/ha_innodb.cc	2012-10-18 18:03:21 +0000
@@ -12494,6 +12494,12 @@
   "The size of the memory buffer InnoDB uses to cache data and indexes of its tables.",
   NULL, NULL, 128*1024*1024L, 32*1024*1024L, LONGLONG_MAX, 1024*1024L);
 
+static MYSQL_SYSVAR_BOOL(buffer_pool_populate, srv_buf_pool_populate,
+  PLUGIN_VAR_NOCMDARG | PLUGIN_VAR_READONLY,
+  "Preallocate (pre-fault) the page frames required for the mapping "
+  "established by the buffer pool memory region. Disabled by default.",
+  NULL, NULL, FALSE);
+
 static MYSQL_SYSVAR_LONG(buffer_pool_instances, innobase_buffer_pool_instances,
   PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
   "Number of buffer pool instances, set to higher value on high-end machines to increase scalability",
@@ -12874,6 +12880,7 @@
   MYSQL_SYSVAR(additional_mem_pool_size),
   MYSQL_SYSVAR(autoextend_increment),
   MYSQL_SYSVAR(buffer_pool_size),
+  MYSQL_SYSVAR(buffer_pool_populate),
   MYSQL_SYSVAR(buffer_pool_instances),
   MYSQL_SYSVAR(buffer_pool_shm_key),
   MYSQL_SYSVAR(buffer_pool_shm_checksum),

=== modified file 'Percona-Server/storage/innobase/include/buf0buf.h'
--- Percona-Server/storage/innobase/include/buf0buf.h	2012-08-07 06:10:00 +0000
+++ Percona-Server/storage/innobase/include/buf0buf.h	2012-10-18 18:03:21 +0000
@@ -233,6 +233,7 @@
 buf_pool_init(
 /*=========*/
 	ulint	size,		/*!< in: Size of the total pool in bytes */
+	ibool	populate,	/*!< in: Force virtual page preallocation */
 	ulint	n_instances);	/*!< in: Number of instances */
 /********************************************************************//**
 Frees the buffer pool at shutdown.  This must not be invoked before

=== modified file 'Percona-Server/storage/innobase/include/os0proc.h'
--- Percona-Server/storage/innobase/include/os0proc.h	2010-06-22 15:58:28 +0000
+++ Percona-Server/storage/innobase/include/os0proc.h	2012-10-18 18:03:21 +0000
@@ -58,7 +58,8 @@
 void*
 os_mem_alloc_large(
 /*===============*/
-	ulint*	n);			/*!< in/out: number of bytes */
+	ulint*	n,			/*!< in/out: number of bytes */
+	ibool	populate);		/*!< in: virtual page preallocation */
 /****************************************************************//**
 Frees large pages memory. */
 UNIV_INTERN

=== modified file 'Percona-Server/storage/innobase/include/srv0srv.h'
--- Percona-Server/storage/innobase/include/srv0srv.h	2012-09-28 07:53:51 +0000
+++ Percona-Server/storage/innobase/include/srv0srv.h	2012-10-18 18:03:21 +0000
@@ -179,6 +179,7 @@
 extern ibool	srv_use_sys_malloc;
 #endif /* UNIV_HOTBACKUP */
 extern ulint	srv_buf_pool_size;	/*!< requested size in bytes */
+extern my_bool	srv_buf_pool_populate;	/*!< virtual page preallocation */
 extern ulint    srv_buf_pool_instances; /*!< requested number of buffer pool instances */
 extern ulint	srv_buf_pool_old_size;	/*!< previously requested size */
 extern ulint	srv_buf_pool_curr_size;	/*!< current size in bytes */

=== modified file 'Percona-Server/storage/innobase/os/os0proc.c'
--- Percona-Server/storage/innobase/os/os0proc.c	2011-12-28 10:31:18 +0000
+++ Percona-Server/storage/innobase/os/os0proc.c	2012-10-18 18:03:21 +0000
@@ -32,6 +32,12 @@
 #include "ut0mem.h"
 #include "ut0byte.h"
 
+/* Linux release version */
+#if defined(UNIV_LINUX) && defined(_GNU_SOURCE)
+#include <string.h>		/* strverscmp() */
+#include <sys/utsname.h>	/* uname() */
+#endif
+
 /* FreeBSD for example has only MAP_ANON, Linux has MAP_ANONYMOUS and
 MAP_ANON but MAP_ANON is marked as deprecated */
 #if defined(MAP_ANONYMOUS)
@@ -40,6 +46,13 @@
 #define OS_MAP_ANON	MAP_ANON
 #endif
 
+/* Linux's MAP_POPULATE */
+#if defined(MAP_POPULATE)
+#define OS_MAP_POPULATE	MAP_POPULATE
+#else
+#define OS_MAP_POPULATE	0
+#endif
+
 UNIV_INTERN ibool os_use_large_pages;
 /* Large page size. This may be a boot-time option on some platforms */
 UNIV_INTERN ulint os_large_page_size;
@@ -63,13 +76,32 @@
 }
 
 /****************************************************************//**
+Retrieve and compare operating system release.
+@return	TRUE if the OS release is equal to, or later than release. */
+UNIV_INTERN
+ibool
+os_compare_release(
+/*===============*/
+	const char*	release		/*!< in: OS release */
+	__attribute__((unused)))
+{
+#if defined(UNIV_LINUX) && defined(_GNU_SOURCE)
+	struct utsname name;
+	return uname(&name) == 0 && strverscmp(name.release, release) >= 0;
+#else
+	return 0;
+#endif
+}
+
+/****************************************************************//**
 Allocates large pages memory.
 @return	allocated memory */
 UNIV_INTERN
 void*
 os_mem_alloc_large(
 /*===============*/
-	ulint*	n)			/*!< in/out: number of bytes */
+	ulint*	n,			/*!< in/out: number of bytes */
+	ibool	populate)		/*!< in: virtual page preallocation */
 {
 	void*	ptr;
 	ulint	size;
@@ -155,12 +187,13 @@
 	ut_ad(ut_is_2pow(size));
 	size = *n = ut_2pow_round(*n + (size - 1), size);
 	ptr = mmap(NULL, size, PROT_READ | PROT_WRITE,
-		   MAP_PRIVATE | OS_MAP_ANON, -1, 0);
+		   MAP_PRIVATE | OS_MAP_ANON |
+		   (populate ? OS_MAP_POPULATE : 0), -1, 0);
 	if (UNIV_UNLIKELY(ptr == (void*) -1)) {
 		fprintf(stderr, "InnoDB: mmap(%lu bytes) failed;"
 			" errno %lu\n",
 			(ulong) size, (ulong) errno);
-		ptr = NULL;
+		return(NULL);
 	} else {
 		os_fast_mutex_lock(&ut_list_mutex);
 		ut_total_allocated_memory += size;
@@ -168,6 +201,25 @@
 		UNIV_MEM_ALLOC(ptr, size);
 	}
 #endif
+
+#if OS_MAP_ANON && OS_MAP_POPULATE
+	/* MAP_POPULATE is only supported for private mappings
+	since Linux 2.6.23. */
+	populate = populate && !os_compare_release("2.6.23");
+
+	if (populate) {
+		fprintf(stderr, "InnoDB: Warning: mmap(MAP_POPULATE) "
+			"is not supported for private mappings. "
+			"Forcing preallocation by faulting in pages.\n");
+	}
+#endif
+
+	/* Initialize the entire buffer to force the allocation
+	of physical memory page frames. */
+	if (populate) {
+		memset(ptr, '\0', size);
+	}
+
 	return(ptr);
 }
 

=== modified file 'Percona-Server/storage/innobase/row/row0merge.c'
--- Percona-Server/storage/innobase/row/row0merge.c	2012-07-13 11:34:09 +0000
+++ Percona-Server/storage/innobase/row/row0merge.c	2012-10-18 18:03:21 +0000
@@ -2720,7 +2720,7 @@
 
 	merge_files = mem_alloc(n_indexes * sizeof *merge_files);
 	block_size = 3 * merge_sort_block_size;
-	block_mem = os_mem_alloc_large(&block_size);
+	block_mem = os_mem_alloc_large(&block_size, FALSE);
 
 	for (i = 0; i < UT_ARR_SIZE(block); i++) {
 		block[i] = (row_merge_block_t ) ((byte *) block_mem +

=== modified file 'Percona-Server/storage/innobase/srv/srv0srv.c'
--- Percona-Server/storage/innobase/srv/srv0srv.c	2012-09-28 07:53:51 +0000
+++ Percona-Server/storage/innobase/srv/srv0srv.c	2012-10-18 18:03:21 +0000
@@ -232,6 +232,8 @@
 UNIV_INTERN my_bool	srv_use_sys_malloc	= TRUE;
 /* requested size in kilobytes */
 UNIV_INTERN ulint	srv_buf_pool_size	= ULINT_MAX;
+/* force virtual page preallocation (prefault) */
+UNIV_INTERN my_bool	srv_buf_pool_populate	= FALSE;
 /* requested number of buffer pool instances */
 UNIV_INTERN ulint       srv_buf_pool_instances  = 1;
 /* previously requested size */

=== modified file 'Percona-Server/storage/innobase/srv/srv0start.c'
--- Percona-Server/storage/innobase/srv/srv0start.c	2012-09-28 07:53:51 +0000
+++ Percona-Server/storage/innobase/srv/srv0start.c	2012-10-18 18:03:21 +0000
@@ -1540,7 +1540,8 @@
 			((double) srv_buf_pool_size) / (1024 * 1024));
 	}
 
-	err = buf_pool_init(srv_buf_pool_size, srv_buf_pool_instances);
+	err = buf_pool_init(srv_buf_pool_size, (ibool) srv_buf_pool_populate,
+			    srv_buf_pool_instances);
 
 	ut_print_timestamp(stderr);
 	fprintf(stderr,