MDEV-36234: Add innodb_linux_aio

This controls which linux implementation to use for
innodb_use_native_aio=ON.

innodb_linux_aio=auto is equivalent to innodb_linux_aio=io_uring when
it is available, and falling back to innodb_linux_aio=aio when not.

Debian packaging is no longer aio exclusive or uring, so
for those older Debian or Ubuntu releases, its a remove_uring directive.
For more recent releases, add mandatory liburing for consistent packaging.

WITH_LIBAIO is now an independent option from WITH_URING.

is_linux_native_aio_supported(): Remove. This had originally been added in
mysql/mysql-server@0da310b in 2012
to fix an issue where io_submit() on CentOS 5.5 would return EINVAL
for a /tmp/#sql*.ibd file associated with CREATE TEMPORARY TABLE.
But, starting with commit 2e814d4 InnoDB
temporary tables will be written to innodb_temp_data_file_path.
The 2012 commit said that the error could occur on "old kernels".
Any GNU/Linux distribution that we currently support should be based
on a newer Linux kernel; for example, Red Hat Enterprise Linux 7
was released in 2014.

This is joint work with Daniel Black and Vladislav Vaintroub.

Author: dr-m

debian/autobake-deb.sh

@@ -64,11 +64,10 @@ add_lsb_base_depends()
   sed -e 's#lsof #lsb-base (>= 3.0-10),\n         lsof #' -i debian/control
 }
 
-replace_uring_with_aio()
+remove_uring()
 {
-  sed 's/liburing-dev/libaio-dev/g' -i debian/control
-  sed -e '/-DIGNORE_AIO_CHECK=ON/d' \
-      -e '/-DWITH_URING=ON/d' -i debian/rules
+  sed -e '/liburing-dev/d' -i debian/control
+  sed -e '/-DWITH_URING=ON/d' -i debian/rules
 }
 
 disable_libfmt()
@@ -116,7 +115,7 @@ in
   # Debian
   "buster")
     disable_libfmt
-    replace_uring_with_aio
+    remove_uring
     ;&
   "bullseye")
     add_lsb_base_depends
@@ -127,7 +126,7 @@ in
     # so no removal is necessary.
     if [[ ! "$architecture" =~ amd64|arm64|armel|armhf|i386|mips64el|mipsel|ppc64el|s390x ]]
     then
-      replace_uring_with_aio
+      remove_uring
     fi
     ;&
   "trixie"|"sid")
@@ -136,8 +135,8 @@ in
     ;;
   # Ubuntu
   "focal")
-    replace_uring_with_aio
     disable_libfmt
+    remove_uring
     ;&
   "jammy"|"kinetic")
     add_lsb_base_depends

debian/rules

@@ -87,9 +87,6 @@ endif
 	# quality standards in Debian. Also building it requires an extra 4 GB of disk
 	# space which makes native Debian builds fail as the total disk space needed
 	# for MariaDB becomes over 10 GB. Only build CS via autobake-deb.sh.
-	#
-	# Note: Don't use '-DWITH_URING=ON' as some Buildbot builders are missing it
-	# and would fail permanently.
 	PATH=$${MYSQL_BUILD_PATH:-"/usr/lib/ccache:/usr/local/bin:/usr/bin:/bin"} \
 	    dh_auto_configure --builddirectory=$(BUILDDIR) -- \
 	    -DCMAKE_BUILD_TYPE=RelWithDebInfo \
@@ -103,6 +100,8 @@ endif
 	    -DPLUGIN_AWS_KEY_MANAGEMENT=NO \
 	    -DPLUGIN_COLUMNSTORE=NO \
 	    -DIGNORE_AIO_CHECK=ON \
+	    -DWITH_URING=ON \
+	    -DWITH_LIBAIO=ON \
 	    -DDEB=$(DEB_VENDOR)
 
 # This is needed, otherwise 'make test' will run before binaries have been built

extra/mariabackup/xtrabackup.cc

@@ -378,6 +378,10 @@ extern const char *innodb_checksum_algorithm_names[];
 extern TYPELIB innodb_checksum_algorithm_typelib;
 extern const char *innodb_flush_method_names[];
 extern TYPELIB innodb_flush_method_typelib;
+#ifdef __linux__
+extern const char *innodb_linux_aio_names[];
+extern TYPELIB innodb_linux_aio_typelib;
+#endif
 
 static const char *binlog_info_values[] = {"off", "lockless", "on", "auto",
 					   NullS};
@@ -1325,6 +1329,9 @@ enum options_xtrabackup
   OPT_INNODB_READ_IO_THREADS,
   OPT_INNODB_WRITE_IO_THREADS,
   OPT_INNODB_USE_NATIVE_AIO,
+#ifdef __linux__
+  OPT_INNODB_LINUX_AIO,
+#endif
   OPT_INNODB_PAGE_SIZE,
   OPT_INNODB_BUFFER_POOL_FILENAME,
   OPT_INNODB_LOCK_WAIT_TIMEOUT,
@@ -1925,6 +1932,14 @@ struct my_option xb_server_options[] =
    (G_PTR*) &srv_use_native_aio,
    (G_PTR*) &srv_use_native_aio, 0, GET_BOOL, NO_ARG,
    TRUE, 0, 0, 0, 0, 0},
+#ifdef __linux__
+  {"innodb_linux_aio", OPT_INNODB_LINUX_AIO,
+   "Which linux AIO implementation to use, auto (io_uring, failing to aio) or explicit",
+   (G_PTR*) &srv_linux_aio_method,
+   (G_PTR*) &srv_linux_aio_method,
+   &innodb_linux_aio_typelib, GET_ENUM, REQUIRED_ARG,
+   SRV_LINUX_AIO_AUTO, 0, 0, 0, 0, 0},
+#endif
   {"innodb_page_size", OPT_INNODB_PAGE_SIZE,
    "The universal page size of the database.",
    (G_PTR*) &innobase_page_size, (G_PTR*) &innobase_page_size, 0,
@@ -2520,19 +2535,8 @@ static bool innodb_init_param()
 
 	ut_ad(DATA_MYSQL_BINARY_CHARSET_COLL == my_charset_bin.number);
 
-#ifdef _WIN32
+#if defined(_WIN32) || defined(LINUX_NATIVE_AIO) || defined(HAVE_URING)
 	srv_use_native_aio = TRUE;
-
-#elif defined(LINUX_NATIVE_AIO)
-
-	if (srv_use_native_aio) {
-		msg("InnoDB: Using Linux native AIO");
-	}
-#elif defined(HAVE_URING)
-
-	if (srv_use_native_aio) {
-		msg("InnoDB: Using liburing");
-	}
 #else
 	/* Currently native AIO is supported only on windows and linux
 	and that also when the support is compiled in. In all other
@@ -5467,6 +5471,11 @@ static bool xtrabackup_backup_func()
 		msg("Error: cannot initialize AIO subsystem");
 		goto fail;
 	}
+#ifdef __linux__
+	if (srv_use_native_aio) {
+		msg("InnoDB: Using %s", srv_thread_pool->get_implementation());
+	}
+#endif
 
 	log_sys.create();
 

include/my_time.h

@@ -230,7 +230,6 @@ static inline longlong sec_part_unshift(longlong second_part, uint digits)
 /* Date/time rounding and truncation functions */
 static inline long my_time_fraction_remainder(long nr, uint decimals)
 {
-  DBUG_ASSERT(decimals <= TIME_SECOND_PART_DIGITS);
   return nr % (long) log_10_int[TIME_SECOND_PART_DIGITS - decimals];
 }
 static inline void my_datetime_trunc(MYSQL_TIME *ltime, uint decimals)

mysql-test/mariadb-test-run.pl

@@ -4548,7 +4548,7 @@ ($$)
      qr|InnoDB: io_setup\(\) attempt|,
      qr|InnoDB: io_setup\(\) failed with EAGAIN|,
      qr|io_uring_queue_init\(\) failed with|,
-     qr|InnoDB: liburing disabled|,
+     qr|InnoDB: io_uring failed: falling back to libaio|,
      qr/InnoDB: Failed to set O_DIRECT on file/,
      qr|setrlimit could not change the size of core files to 'infinity';|,
      qr|failed to retrieve the MAC address|,

mysql-test/suite/sys_vars/r/innodb_linux_aio_basic.result

@@ -0,0 +1,21 @@
+select @@global.innodb_linux_aio;
+@@global.innodb_linux_aio
+auto
+select @@session.innodb_linux_aio;
+ERROR HY000: Variable 'innodb_linux_aio' is a GLOBAL variable
+show global variables like 'innodb_linux_aio';
+Variable_name	Value
+innodb_linux_aio	auto
+show session variables like 'innodb_linux_aio';
+Variable_name	Value
+innodb_linux_aio	auto
+select * from information_schema.global_variables where variable_name='innodb_linux_aio';
+VARIABLE_NAME	VARIABLE_VALUE
+INNODB_LINUX_AIO	auto
+select * from information_schema.session_variables where variable_name='innodb_linux_aio';
+VARIABLE_NAME	VARIABLE_VALUE
+INNODB_LINUX_AIO	auto
+set global innodb_linux_aio='auto';
+ERROR HY000: Variable 'innodb_linux_aio' is a read only variable
+set session innodb_linux_aio='aio';
+ERROR HY000: Variable 'innodb_linux_aio' is a read only variable

mysql-test/suite/sys_vars/r/sysvars_innodb.result

@@ -5,6 +5,7 @@ variable_name not in (
 'innodb_evict_tables_on_commit_debug', # one may want to override this
 'innodb_use_native_aio',            # default value depends on OS
 'innodb_log_file_buffering',        # only available on Linux and Windows
+'innodb_linux_aio',                 # existence depends on OS
 'innodb_buffer_pool_load_pages_abort')            # debug build only, and is only for testing
 order by variable_name;
 VARIABLE_NAME	INNODB_ADAPTIVE_FLUSHING

mysql-test/suite/sys_vars/t/innodb_linux_aio_basic.test

@@ -0,0 +1,23 @@
+--source include/have_innodb.inc
+--source include/linux.inc
+# enum readonly
+
+#
+# show values;
+#
+select @@global.innodb_linux_aio;
+--error ER_INCORRECT_GLOBAL_LOCAL_VAR
+select @@session.innodb_linux_aio;
+show global variables like 'innodb_linux_aio';
+show session variables like 'innodb_linux_aio';
+select * from information_schema.global_variables where variable_name='innodb_linux_aio';
+select * from information_schema.session_variables where variable_name='innodb_linux_aio';
+
+#
+# show that it's read-only
+#
+--error ER_INCORRECT_GLOBAL_LOCAL_VAR
+set global innodb_linux_aio='auto';
+--error ER_INCORRECT_GLOBAL_LOCAL_VAR
+set session innodb_linux_aio='aio';
+

mysql-test/suite/sys_vars/t/sysvars_innodb.test

@@ -16,5 +16,6 @@ select VARIABLE_NAME, SESSION_VALUE, DEFAULT_VALUE, VARIABLE_SCOPE, VARIABLE_TYP
     'innodb_evict_tables_on_commit_debug', # one may want to override this
     'innodb_use_native_aio',            # default value depends on OS
     'innodb_log_file_buffering',        # only available on Linux and Windows
+    'innodb_linux_aio',                 # existence depends on OS
     'innodb_buffer_pool_load_pages_abort')            # debug build only, and is only for testing
   order by variable_name;

storage/innobase/handler/ha_innodb.cc

@@ -315,6 +315,25 @@ static TYPELIB innodb_stats_method_typelib = {
 	NULL
 };
 
+/** Possible values for system variable "innodb_linux_aio" */
+#ifdef __linux__
+const char* innodb_linux_aio_names[] = {
+	"auto",     /* SRV_LINUX_AIO_AUTO */
+	"io_uring", /* SRV_LINUX_AIO_IO_URING */
+	"aio",      /* SRV_LINUX_AIO_LIBAIO */
+	NullS
+};
+
+/** Used to define an enumerate type of the system variable
+innodb_linux_aio. Used by mariadb-backup too. */
+TYPELIB innodb_linux_aio_typelib = {
+	array_elements(innodb_linux_aio_names) - 1,
+	"innodb_linux_aio_typelib",
+	innodb_linux_aio_names,
+	NULL
+};
+#endif
+
 /** Possible values of the parameter innodb_checksum_algorithm */
 const char* innodb_checksum_algorithm_names[] = {
 	"crc32",
@@ -19659,6 +19678,15 @@ static MYSQL_SYSVAR_BOOL(use_native_aio, srv_use_native_aio,
   "Use native AIO if supported on this platform.",
   NULL, NULL, TRUE);
 
+#ifdef __linux__
+static MYSQL_SYSVAR_ENUM(linux_aio, srv_linux_aio_method,
+  PLUGIN_VAR_RQCMDARG | PLUGIN_VAR_READONLY,
+  "Specifies which Linux AIO implementation should be used."
+  " Possible value are \"auto\" (default) to select io_uring"
+  " and fallback to aio, or explicit \"io_uring\" or \"aio\"",
+  nullptr, nullptr, SRV_LINUX_AIO_AUTO, &innodb_linux_aio_typelib);
+#endif
+
 #ifdef HAVE_LIBNUMA
 static MYSQL_SYSVAR_BOOL(numa_interleave, srv_numa_interleave,
   PLUGIN_VAR_NOCMDARG | PLUGIN_VAR_READONLY,
@@ -20054,6 +20082,9 @@ static struct st_mysql_sys_var* innobase_system_variables[]= {
   MYSQL_SYSVAR(tmpdir),
   MYSQL_SYSVAR(autoinc_lock_mode),
   MYSQL_SYSVAR(use_native_aio),
+#ifdef __linux__
+  MYSQL_SYSVAR(linux_aio),
+#endif
 #ifdef HAVE_LIBNUMA
   MYSQL_SYSVAR(numa_interleave),
 #endif /* HAVE_LIBNUMA */

storage/innobase/include/fil0fil.h

@@ -77,6 +77,19 @@ enum srv_flush_t
 #endif
 };
 
+/** Possible values of innodb_linux_aio */
+#ifdef __linux__
+enum srv_linux_aio_t
+{
+  /** auto, io_uring first and then aio */
+  SRV_LINUX_AIO_AUTO,
+  /** io_uring */
+  SRV_LINUX_AIO_IO_URING,
+  /** aio (libaio interface) */
+  SRV_LINUX_AIO_LIBAIO
+};
+#endif
+
 /** innodb_flush_method */
 extern ulong srv_file_flush_method;
 

storage/innobase/include/srv0srv.h

@@ -178,6 +178,12 @@ OS (provided we compiled Innobase with it in), otherwise we will
 use simulated aio.
 Currently we support native aio on windows and linux */
 extern my_bool	srv_use_native_aio;
+
+#ifdef __linux__
+/* This enum is defined which linux native io method to use */
+extern ulong	srv_linux_aio_method;
+#endif
+
 extern my_bool	srv_numa_interleave;
 
 /* Use atomic writes i.e disable doublewrite buffer */