InnoDB: Do not flush after each write, not even when creating the

data files.  Previously, writes were flushed until the doublewrite
buffer was created.  That would be too slow on systems where
os_file_flush() [or fsync(2)] is slow.  (Bug #12125)


innobase/include/os0file.h:
  Disable os_do_not_call_flush_at_each_write unless #ifdef UNIV_DO_FLUSH
innobase/os/os0file.c:
  Disable os_do_not_call_flush_at_each_write unless #ifdef UNIV_DO_FLUSH
innobase/trx/trx0sys.c:
  Disable os_do_not_call_flush_at_each_write unless #ifdef UNIV_DO_FLUSH

3 files changed, 18 insertions, 1 deletions

diff --git a/innobase/include/os0file.h b/innobase/include/os0file.h
index 63cd41a6d28..280a949c1c5 100644
--- a/innobase/include/os0file.h
+++ b/innobase/include/os0file.h
@@ -17,7 +17,9 @@ Created 10/21/1995 Heikki Tuuri
 #include <time.h>
 #endif
 
+#ifdef UNIV_DO_FLUSH
 extern ibool	os_do_not_call_flush_at_each_write;
+#endif /* UNIV_DO_FLUSH */
 extern ibool	os_has_said_disk_full;
 extern ibool	os_aio_print_debug;
 
diff --git a/innobase/os/os0file.c b/innobase/os/os0file.c
index 4313b7f7019..49f88c0d62a 100644
--- a/innobase/os/os0file.c
+++ b/innobase/os/os0file.c
@@ -33,9 +33,13 @@ ulint	os_innodb_umask		= S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP;
 ulint	os_innodb_umask		= 0;
 #endif
 
+#ifdef UNIV_DO_FLUSH
 /* If the following is set to TRUE, we do not call os_file_flush in every
 os_file_write. We can set this TRUE when the doublewrite buffer is used. */
 ibool	os_do_not_call_flush_at_each_write	= FALSE;
+#else
+/* We do not call os_file_flush in every os_file_write. */
+#endif /* UNIV_DO_FLUSH */
 
 /* We use these mutexes to protect lseek + file i/o operation, if the
 OS does not provide an atomic pread or pwrite, or similar */
@@ -1974,6 +1978,7 @@ os_file_pwrite(
 	os_file_n_pending_pwrites--;
         os_mutex_exit(os_file_count_mutex);
 
+# ifdef UNIV_DO_FLUSH
 	if (srv_unix_file_flush_method != SRV_UNIX_LITTLESYNC
 	    && srv_unix_file_flush_method != SRV_UNIX_NOSYNC
 	    && !os_do_not_call_flush_at_each_write) {
@@ -1984,6 +1989,7 @@ os_file_pwrite(
 
 	        ut_a(TRUE == os_file_flush(file));
 	}
+# endif /* UNIV_DO_FLUSH */
 
         return(ret);
 #else
@@ -2006,6 +2012,7 @@ os_file_pwrite(
 	
 	ret = write(file, buf, (ssize_t)n);
 
+# ifdef UNIV_DO_FLUSH
 	if (srv_unix_file_flush_method != SRV_UNIX_LITTLESYNC
 	    && srv_unix_file_flush_method != SRV_UNIX_NOSYNC
 	    && !os_do_not_call_flush_at_each_write) {
@@ -2016,6 +2023,7 @@ os_file_pwrite(
 
 	        ut_a(TRUE == os_file_flush(file));
 	}
+# endif /* UNIV_DO_FLUSH */
 
 	os_mutex_exit(os_file_seek_mutexes[i]);
 
@@ -2282,9 +2290,11 @@ retry:
 	/* Always do fsync to reduce the probability that when the OS crashes,
 	a database page is only partially physically written to disk. */
 
+# ifdef UNIV_DO_FLUSH
 	if (!os_do_not_call_flush_at_each_write) {
 		ut_a(TRUE == os_file_flush(file));
 	}
+# endif /* UNIV_DO_FLUSH */
 
 	os_mutex_exit(os_file_seek_mutexes[i]);
 
@@ -3498,10 +3508,12 @@ os_aio_windows_handle(
 	if (ret && len == slot->len) {
 		ret_val = TRUE;
 
+# ifdef UNIV_DO_FLUSH
 		if (slot->type == OS_FILE_WRITE
 				&& !os_do_not_call_flush_at_each_write) {
 		         ut_a(TRUE == os_file_flush(slot->file));
 		}
+# endif /* UNIV_DO_FLUSH */
 	} else {
 		os_file_handle_error(slot->name, "Windows aio");
 		
@@ -3582,10 +3594,12 @@ os_aio_posix_handle(
 	*message1 = slot->message1;
 	*message2 = slot->message2;
 
+# ifdef UNIV_DO_FLUSH
 	if (slot->type == OS_FILE_WRITE
 				&& !os_do_not_call_flush_at_each_write) {
 		ut_a(TRUE == os_file_flush(slot->file));
 	}
+# endif /* UNIV_DO_FLUSH */
 
 	os_mutex_exit(array->mutex);
 
diff --git a/innobase/trx/trx0sys.c b/innobase/trx/trx0sys.c
index 51e193d563b..c7292fb7650 100644
--- a/innobase/trx/trx0sys.c
+++ b/innobase/trx/trx0sys.c
@@ -97,8 +97,9 @@ trx_doublewrite_init(
 
 	/* Since we now start to use the doublewrite buffer, no need to call
 	fsync() after every write to a data file */
-
+#ifdef UNIV_DO_FLUSH
 	os_do_not_call_flush_at_each_write = TRUE;
+#endif /* UNIV_DO_FLUSH */
 
 	mutex_create(&(trx_doublewrite->mutex));
 	mutex_set_level(&(trx_doublewrite->mutex), SYNC_DOUBLEWRITE);