Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commitc325a76

Browse files
committed
aio: Add io_method=io_uring
Performing AIO using io_uring can be considerably faster thanio_method=worker, particularly when lots of small IOs are issued, asa) the context-switch overhead for worker based AIO becomes more significantb) the number of IO workers can become limitingio_uring, however, is linux specific and requires an additional compile-timedependency (liburing).This implementation is fairly simple and there are substantial optimizationopportunities.The description of the existing AIO_IO_COMPLETION wait event is updated tomake the difference between it and the new AIO_IO_URING_EXECUTION clearer.Reviewed-by: Noah Misch <noah@leadboat.com>Reviewed-by: Jakub Wartak <jakub.wartak@enterprisedb.com>Discussion:https://postgr.es/m/uvrtrknj4kdytuboidbhwclo4gxhswwcpgadptsjvjqcluzmah%40brqs62irg4dtDiscussion:https://postgr.es/m/20210223100344.llw5an2aklengrmn@alap3.anarazel.deDiscussion:https://postgr.es/m/stj36ea6yyhoxtqkhpieia2z4krnam7qyetc57rfezgk4zgapf@gcnactj4z56m
1 parent8eadd5c commitc325a76

File tree

14 files changed

+589
-2
lines changed

14 files changed

+589
-2
lines changed

‎.cirrus.tasks.yml

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -493,11 +493,14 @@ task:
493493
# - Uses undefined behaviour and alignment sanitizers, sanitizer failures
494494
# are typically printed in the server log
495495
# - Test both 64bit and 32 bit builds
496+
# - uses io_method=io_uring
496497
-name:Linux - Debian Bookworm - Meson
497498

498499
env:
499500
CCACHE_MAXSIZE:"400M"# tests two different builds
500501
SANITIZER_FLAGS:-fsanitize=alignment,undefined
502+
PG_TEST_INITDB_EXTRA_OPTS:>-
503+
-c io_method=io_uring
501504
502505
configure_script:|
503506
su postgres <<-EOF

‎doc/src/sgml/config.sgml

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2710,6 +2710,14 @@ include_dir 'conf.d'
27102710
<literal>worker</literal> (execute asynchronous I/O using worker processes)
27112711
</para>
27122712
</listitem>
2713+
<listitem>
2714+
<para>
2715+
<literal>io_uring</literal> (execute asynchronous I/O using
2716+
io_uring, requires a build with
2717+
<link linkend="configure-option-with-liburing"><option>--with-liburing</option></link> /
2718+
<link linkend="configure-with-liburing-meson"><option>-Dliburing</option></link>)
2719+
</para>
2720+
</listitem>
27132721
<listitem>
27142722
<para>
27152723
<literal>sync</literal> (execute asynchronous-eligible I/O synchronously)

‎src/backend/storage/aio/Makefile

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -14,6 +14,7 @@ OBJS = \
1414
aio_init.o\
1515
aio_io.o\
1616
aio_target.o\
17+
method_io_uring.o\
1718
method_sync.o\
1819
method_worker.o\
1920
read_stream.o

‎src/backend/storage/aio/aio.c

Lines changed: 41 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -65,6 +65,9 @@ static void pgaio_io_wait(PgAioHandle *ioh, uint64 ref_generation);
6565
conststructconfig_enum_entryio_method_options[]= {
6666
{"sync",IOMETHOD_SYNC, false},
6767
{"worker",IOMETHOD_WORKER, false},
68+
#ifdefIOMETHOD_IO_URING_ENABLED
69+
{"io_uring",IOMETHOD_IO_URING, false},
70+
#endif
6871
{NULL,0, false}
6972
};
7073

@@ -82,6 +85,9 @@ PgAioBackend *pgaio_my_backend;
8285
staticconstIoMethodOps*constpgaio_method_ops_table[]= {
8386
[IOMETHOD_SYNC]=&pgaio_sync_ops,
8487
[IOMETHOD_WORKER]=&pgaio_worker_ops,
88+
#ifdefIOMETHOD_IO_URING_ENABLED
89+
[IOMETHOD_IO_URING]=&pgaio_uring_ops,
90+
#endif
8591
};
8692

8793
/* callbacks for the configured io_method, set by assign_io_method */
@@ -1118,6 +1124,41 @@ pgaio_closing_fd(int fd)
11181124
* it's probably not worth it.
11191125
*/
11201126
pgaio_submit_staged();
1127+
1128+
/*
1129+
* If requested by the IO method, wait for all IOs that use the
1130+
* to-be-closed FD.
1131+
*/
1132+
if (pgaio_method_ops->wait_on_fd_before_close)
1133+
{
1134+
/*
1135+
* As waiting for one IO to complete may complete multiple IOs, we
1136+
* can't just use a mutable list iterator. The maximum number of
1137+
* in-flight IOs is fairly small, so just restart the loop after
1138+
* waiting for an IO.
1139+
*/
1140+
while (!dclist_is_empty(&pgaio_my_backend->in_flight_ios))
1141+
{
1142+
dlist_iteriter;
1143+
PgAioHandle*ioh=NULL;
1144+
1145+
dclist_foreach(iter,&pgaio_my_backend->in_flight_ios)
1146+
{
1147+
ioh=dclist_container(PgAioHandle,node,iter.cur);
1148+
1149+
if (pgaio_io_uses_fd(ioh,fd))
1150+
break;
1151+
else
1152+
ioh=NULL;
1153+
}
1154+
1155+
if (!ioh)
1156+
break;
1157+
1158+
/* see comment in pgaio_io_wait_for_free() about raciness */
1159+
pgaio_io_wait(ioh,ioh->generation);
1160+
}
1161+
}
11211162
}
11221163

11231164
/*

‎src/backend/storage/aio/aio_io.c

Lines changed: 22 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -188,3 +188,25 @@ pgaio_io_get_op_name(PgAioHandle *ioh)
188188

189189
returnNULL;/* silence compiler */
190190
}
191+
192+
/*
193+
* Used to determine if an IO needs to be waited upon before the file
194+
* descriptor can be closed.
195+
*/
196+
bool
197+
pgaio_io_uses_fd(PgAioHandle*ioh,intfd)
198+
{
199+
Assert(ioh->state >=PGAIO_HS_DEFINED);
200+
201+
switch (ioh->op)
202+
{
203+
casePGAIO_OP_READV:
204+
returnioh->op_data.read.fd==fd;
205+
casePGAIO_OP_WRITEV:
206+
returnioh->op_data.write.fd==fd;
207+
casePGAIO_OP_INVALID:
208+
return false;
209+
}
210+
211+
return false;/* silence compiler */
212+
}

‎src/backend/storage/aio/meson.build

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -6,6 +6,7 @@ backend_sources += files(
66
'aio_init.c',
77
'aio_io.c',
88
'aio_target.c',
9+
'method_io_uring.c',
910
'method_sync.c',
1011
'method_worker.c',
1112
'read_stream.c',

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp