Merge lp:~percona-dev/percona-server/release-5.1.52-12-bug643463 into lp:~percona-dev/percona-server/release-5.1.52-12

[Alexey Kopytov]

[Vadim Tkachenko (vadim-tk)]

Please rename patch to have more descriptive name, i.e innodb_fast_shutdown.

review: Needs Fixing

[Vadim Tkachenko (vadim-tk)]

Fred,
it has to be documented on wiki.

[Alexey Kopytov (akopytov)]

> Please rename patch to have more descriptive name, i.e innodb_fast_shutdown.
Done.

[Yasufumi Kinoshita (yasufumi-kinoshita)]

innodb_fast_shutdown is existed variable name already...
Is it related to the variavle?
Why do you choose more confusable name than "bug643463"?

[Alexey Kopytov (akopytov)]

Note that this patch will unlikely provide any measurable benefits for production systems (decreasing shutdown time by at most 10 seconds does not look significant).

It's the test suite that will mostly benefit from this patch, where there are many server restarts. So this patch looks more like a bugfix to me rather than a new feature.

Does it really make sense to give it a descriptive name?

[Fred Linhoss (fred-linhoss)]

review: Approve (documentation)

[Vadim Tkachenko (vadim-tk)]

ok to merge

review: Approve

=== added file 'innodb_fast_shutdown.patch'
--- innodb_fast_shutdown.patch	1970-01-01 00:00:00 +0000
+++ innodb_fast_shutdown.patch	2010-11-16 18:46:59 +0000
@@ -0,0 +1,235 @@
+# name       : innodb_fast_shutdown
+# introduced : 13
+# maintainer : Alexey
+#
+# Shutting down XtraDB takes uninterruptible sleep()s up to 10
+# seconds, even when there is no actual work to do during shutdown.
+#
+# This patch removes most such delays during shutdown, as found using
+# PMP. This makes standard test run very close in speed to with
+# --loose-innodb-fast-shutdown=2, and greatly speeds up running the test
+# suite.
+#
+# The patch also implements os_event_wait_time() for POSIX systems.
+--- /dev/null	1970-01-01 00:00:00.000000000 +0000
++++ b/COPYING.innodb_fast_shutdown	2010-11-16 21:37:51.000000000 +0300
+@@ -0,0 +1,10 @@
++Copyright (c) 2010, Kristian Nielsen
++All rights reserved.
++
++Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
++
++    * Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
++    * Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
++    * Neither the name of the author nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.
++
++THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+--- a/storage/innodb_plugin/include/os0sync.h	2010-11-16 21:33:00.000000000 +0300
++++ b/storage/innodb_plugin/include/os0sync.h	2010-11-16 21:34:06.000000000 +0300
+@@ -189,14 +189,14 @@
+ 
+ /**********************************************************//**
+ Waits for an event object until it is in the signaled state or
+-a timeout is exceeded. In Unix the timeout is always infinite.
++a timeout is exceeded.
+ @return	0 if success, OS_SYNC_TIME_EXCEEDED if timeout was exceeded */
+ UNIV_INTERN
+ ulint
+ os_event_wait_time(
+ /*===============*/
+ 	os_event_t	event,	/*!< in: event to wait */
+-	ulint		time);	/*!< in: timeout in microseconds, or
++	ulint		wtime);	/*!< in: timeout in microseconds, or
+ 				OS_SYNC_INFINITE_TIME */
+ #ifdef __WIN__
+ /**********************************************************//**
+--- a/storage/innodb_plugin/include/srv0srv.h	2010-11-16 21:33:00.000000000 +0300
++++ b/storage/innodb_plugin/include/srv0srv.h	2010-11-16 21:34:06.000000000 +0300
+@@ -57,6 +57,9 @@
+ thread starts running */
+ extern os_event_t	srv_lock_timeout_thread_event;
+ 
++/* This event is set at shutdown to wakeup threads from sleep */
++extern os_event_t	srv_shutdown_event;
++
+ /* If the last data file is auto-extended, we add this many pages to it
+ at a time */
+ #define SRV_AUTO_EXTEND_INCREMENT	\
+--- a/storage/innodb_plugin/log/log0log.c	2010-11-16 21:33:00.000000000 +0300
++++ b/storage/innodb_plugin/log/log0log.c	2010-11-16 21:34:06.000000000 +0300
+@@ -3102,6 +3102,7 @@
+ 	algorithm only works if the server is idle at shutdown */
+ 
+ 	srv_shutdown_state = SRV_SHUTDOWN_CLEANUP;
++	os_event_set(srv_shutdown_event);
+ loop:
+ 	os_thread_sleep(100000);
+ 
+--- a/storage/innodb_plugin/os/os0sync.c	2010-11-16 21:33:00.000000000 +0300
++++ b/storage/innodb_plugin/os/os0sync.c	2010-11-16 21:34:06.000000000 +0300
+@@ -31,6 +31,9 @@
+ 
+ #ifdef __WIN__
+ #include <windows.h>
++#else
++#include <sys/time.h>
++#include <time.h>
+ #endif
+ 
+ #include "ut0mem.h"
+@@ -407,14 +410,14 @@
+ 
+ /**********************************************************//**
+ Waits for an event object until it is in the signaled state or
+-a timeout is exceeded. In Unix the timeout is always infinite.
++a timeout is exceeded.
+ @return	0 if success, OS_SYNC_TIME_EXCEEDED if timeout was exceeded */
+ UNIV_INTERN
+ ulint
+ os_event_wait_time(
+ /*===============*/
+ 	os_event_t	event,	/*!< in: event to wait */
+-	ulint		time)	/*!< in: timeout in microseconds, or
++	ulint		wtime)	/*!< in: timeout in microseconds, or
+ 				OS_SYNC_INFINITE_TIME */
+ {
+ #ifdef __WIN__
+@@ -422,8 +425,8 @@
+ 
+ 	ut_a(event);
+ 
+-	if (time != OS_SYNC_INFINITE_TIME) {
+-		err = WaitForSingleObject(event->handle, (DWORD) time / 1000);
++	if (wtime != OS_SYNC_INFINITE_TIME) {
++		err = WaitForSingleObject(event->handle, (DWORD) wtime / 1000);
+ 	} else {
+ 		err = WaitForSingleObject(event->handle, INFINITE);
+ 	}
+@@ -439,13 +442,47 @@
+ 		return(1000000); /* dummy value to eliminate compiler warn. */
+ 	}
+ #else
+-	UT_NOT_USED(time);
++	int		err;
++	int		ret = 0;
++	ulint		tmp;
++	ib_int64_t	old_count;
++	struct timeval	tv_start;
++	struct timespec	timeout;
++
++	if (wtime == OS_SYNC_INFINITE_TIME) {
++		os_event_wait(event);
++		return 0;
++	}
++
++	/* Compute the absolute point in time at which to time out. */
++	gettimeofday(&tv_start, NULL);
++	tmp = tv_start.tv_usec + wtime;
++	timeout.tv_sec = tv_start.tv_sec + (tmp / 1000000);
++	timeout.tv_nsec = (tmp % 1000000) * 1000;
++
++	os_fast_mutex_lock(&(event->os_mutex));
++	old_count = event->signal_count;
+ 
+-	/* In Posix this is just an ordinary, infinite wait */
++	for (;;) {
++		if (event->is_set == TRUE || event->signal_count != old_count)
++			break;
++
++		err = pthread_cond_timedwait(&(event->cond_var),
++					     &(event->os_mutex), &timeout);
++		if (err == ETIMEDOUT) {
++			ret = OS_SYNC_TIME_EXCEEDED;
++			break;
++		}
++	}
+ 
+-	os_event_wait(event);
++	os_fast_mutex_unlock(&(event->os_mutex));
++
++	if (srv_shutdown_state == SRV_SHUTDOWN_EXIT_THREADS) {
++
++		os_thread_exit(NULL);
++	}
+ 
+-	return(0);
++	return ret;
+ #endif
+ }
+ 
+--- a/storage/innodb_plugin/srv/srv0srv.c	2010-11-16 21:33:00.000000000 +0300
++++ b/storage/innodb_plugin/srv/srv0srv.c	2010-11-16 21:34:06.000000000 +0300
+@@ -704,6 +704,8 @@
+ 
+ UNIV_INTERN os_event_t	srv_lock_timeout_thread_event;
+ 
++UNIV_INTERN os_event_t	srv_shutdown_event;
++
+ UNIV_INTERN srv_sys_t*	srv_sys	= NULL;
+ 
+ /* padding to prevent other memory update hotspots from residing on
+@@ -1009,6 +1011,7 @@
+ 	}
+ 
+ 	srv_lock_timeout_thread_event = os_event_create(NULL);
++	srv_shutdown_event = os_event_create(NULL);
+ 
+ 	for (i = 0; i < SRV_MASTER + 1; i++) {
+ 		srv_n_threads_active[i] = 0;
+@@ -2236,7 +2239,7 @@
+ 	/* Wake up every 5 seconds to see if we need to print
+ 	monitor information. */
+ 
+-	os_thread_sleep(5000000);
++	os_event_wait_time(srv_shutdown_event, 5000000);
+ 
+ 	current_time = time(NULL);
+ 
+@@ -2378,7 +2381,7 @@
+ 	/* When someone is waiting for a lock, we wake up every second
+ 	and check if a timeout has passed for a lock wait */
+ 
+-	os_thread_sleep(1000000);
++	os_event_wait_time(srv_shutdown_event, 1000000);
+ 
+ 	srv_lock_timeout_active = TRUE;
+ 
+@@ -2543,7 +2546,7 @@
+ 
+ 	fflush(stderr);
+ 
+-	os_thread_sleep(1000000);
++	os_event_wait_time(srv_shutdown_event, 1000000);
+ 
+ 	if (srv_shutdown_state < SRV_SHUTDOWN_CLEANUP) {
+ 
+@@ -2587,7 +2590,7 @@
+ 	last_dump_time = time(NULL);
+ 
+ loop:
+-	os_thread_sleep(5000000);
++	os_event_wait_time(srv_shutdown_event, 5000000);
+ 
+ 	if (srv_shutdown_state >= SRV_SHUTDOWN_CLEANUP) {
+ 		goto exit_func;
+@@ -2751,7 +2754,7 @@
+ 
+ 		if (!skip_sleep) {
+ 
+-			os_thread_sleep(1000000);
++			os_event_wait_time(srv_shutdown_event, 1000000);
+ 			srv_main_sleeps++;
+ 
+ 			/*
+@@ -3337,9 +3340,10 @@
+ 		mutex_exit(&kernel_mutex);
+ 
+ 		sleep_ms = 10;
++		os_event_reset(srv_shutdown_event);
+ 	}
+ 
+-	os_thread_sleep( sleep_ms * 1000 );
++	os_event_wait_time(srv_shutdown_event, sleep_ms * 1000);
+ 
+ 	history_len = trx_sys->rseg_history_len;
+ 	if (history_len > 1000)
 
=== modified file 'series'
--- series	2010-10-29 09:20:36 +0000
+++ series	2010-11-16 18:46:59 +0000
@@ -49,3 +49,4 @@
 remove_fcntl_excessive_calls.patch
 sql_no_fcache.patch
 show_slave_status_nolock.patch
+innodb_fast_shutdow.patch