log0log.c InnoDB now allows parameter innodb_unix_file_flush_method (221b1ea5) · Commits · Software / OSDI20 Artifacts / mariadb

innobase/log/log0log.c

+10 −4

Original line number	Diff line number	Diff line
		@@ -838,14 +838,18 @@ log_io_complete(
		/* It was a checkpoint write */
		group = (log_group_t*)((ulint)group - 1);

		if (srv_unix_file_flush_method == SRV_UNIX_LITTLESYNC) {
		fil_flush(group->space_id);
		}

		log_io_complete_checkpoint(group);

		return;
		}

		if (srv_unix_file_flush_method == SRV_UNIX_LITTLESYNC) {
		fil_flush(group->space_id);
		}

		mutex_enter(&(log_sys->mutex));

		@@ -1474,7 +1478,9 @@ log_checkpoint(
		recv_apply_hashed_log_recs(TRUE);
		}

		if (srv_unix_file_flush_method == SRV_UNIX_LITTLESYNC) {
		fil_flush_file_spaces(FIL_TABLESPACE);
		}

		mutex_enter(&(log_sys->mutex));

innobase/os/os0file.c

+28 −12

Original line number	Diff line number	Diff line
		@@ -9,6 +9,7 @@ Created 10/21/1995 Heikki Tuuri
		#include "os0file.h"
		#include "os0sync.h"
		#include "ut0mem.h"
		#include "srv0srv.h"


		#ifdef POSIX_ASYNC_IO
		@@ -345,12 +346,11 @@ os_file_create(

		UT_NOT_USED(purpose);

		/* On Linux opening a file in the O_SYNC mode seems to be much
		more efficient for small writes than calling an explicit fsync or
		fdatasync after each write, but on Solaris O_SYNC and O_DSYNC is
		extremely slow in large block writes to a big file. Therefore we
		do not use these options, but use explicit fdatasync. */

		#ifdef O_DSYNC
		if (srv_unix_file_flush_method == SRV_UNIX_O_DSYNC) {
		create_flag = create_flag \| O_DSYNC;
		}
		#endif
		if (create_mode == OS_FILE_CREATE) {
		file = open(name, create_flag, S_IRUSR \| S_IWUSR \| S_IRGRP
		\| S_IWGRP \| S_IROTH \| S_IWOTH);
		@@ -547,6 +547,12 @@ os_file_flush(
		#else
		int ret;

		#ifdef O_DSYNC
		if (srv_unix_file_flush_method == SRV_UNIX_O_DSYNC) {
		return(TRUE);
		}
		#endif

		#ifdef HAVE_FDATASYNC
		ret = fdatasync(file);
		#else
		@@ -621,10 +627,15 @@ os_file_pwrite(
		#ifdef HAVE_PWRITE
		ret = pwrite(file, buf, n, offs);

		/* Always do fsync to reduce the probability that when the OS crashes,
		a database page is only partially physically written to disk. */
		if (srv_unix_file_flush_method != SRV_UNIX_LITTLESYNC
		&& srv_unix_file_flush_method != SRV_UNIX_NOSYNC) {

		/* Always do fsync to reduce the probability that when
		the OS crashes, a database page is only partially
		physically written to disk. */

		ut_a(TRUE == os_file_flush(file));
		}

		return(ret);
		#else
		@@ -645,10 +656,15 @@ os_file_pwrite(

		ret = write(file, buf, n);

		/* Always do fsync to reduce the probability that when the OS crashes,
		a database page is only partially physically written to disk. */
		if (srv_unix_file_flush_method != SRV_UNIX_LITTLESYNC
		&& srv_unix_file_flush_method != SRV_UNIX_NOSYNC) {

		/* Always do fsync to reduce the probability that when
		the OS crashes, a database page is only partially
		physically written to disk. */

		ut_a(TRUE == os_file_flush(file));
		}

		os_mutex_exit(os_file_seek_mutexes[i]);

innobase/srv/srv0start.c

+0 −2

Original line number	Diff line number	Diff line
		@@ -548,8 +548,6 @@ innobase_start_or_create_for_mysql(void)
		return(DB_ERROR);
		}

		printf("InnoDB file flush method %lu\n", srv_unix_file_flush_method);

		os_aio_use_native_aio = srv_use_native_aio;

		err = srv_boot();