On 4/3/25 15:12, Jakub Wartak wrote:
> On Thu, Apr 3, 2025 at 1:52 PM Tomas Vondra <to...@vondra.me> wrote:
>
>> ...
>>
>> So unless someone can demonstrate a use case where this would matter,
>> I'd not worry about it too much.
>
> OK, fine for me - just 3 cols for pg_buffercache_numa is fine for me,
> it's just that I don't have cycles left today and probably lack skills
> (i've never dealt with arrays so far) thus it would be slow to get it
> right... but I can pick up anything tomorrow morning.
>
OK, I took a stab at reworking/simplifying this the way I proposed.
Here's v24 - needs more polishing, but hopefully enough to show what I
had in mind.
It does these changes:
1) Drops 0002 with the pg_buffercache refactoring, because the new view
is not "extending" the existing one.
2) Reworks pg_buffercache_num to return just three columns, bufferid,
page_num and node_id. page_num is a sequence starting from 0 for each
buffer.
3) It now builds an array of records, with one record per buffer/page.
4) I realized we don't really need to worry about buffers_per_page very
much, except for logging/debugging. There's always "at least one page"
per buffer, even if an incomplete one, so we can do this:
os_page_count = NBuffers * Max(1, pages_per_buffer);
and then
for (i = 0; i < NBuffers; i++)
{
for (j = 0; j < Max(1, pages_per_buffer); j++)
{
..
}
}
and everything just works fine, I think.
Opinions? I personally find this much cleaner / easier to understand.
regards
--
Tomas VondraFrom 70018899b698b186ffebb03c7336022ae83fbfb8 Mon Sep 17 00:00:00 2001
From: Jakub Wartak <jakub.war...@enterprisedb.com>
Date: Wed, 2 Apr 2025 12:29:22 +0200
Subject: [PATCH v24 1/3] Add support for basic NUMA awareness
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Add basic NUMA awareness routines, using a minimal src/port/pg_numa.c
portability wrapper and an optional build dependency, enabled by
--with-libnuma configure option. For now this is Linux-only, other
platforms may be supported later.
A built-in SQL function pg_numa_available() allows checking NUMA
support, i.e. that the server was built/linked with NUMA library.
The libnuma library is not available on 32-bit builds (there's no shared
object for i386), so we disable it in that case. The i386 is very memory
limited anyway, even with PAE, so NUMA is mostly irrelevant.
On Linux we use move_pages(2) syscall for speed instead of
get_mempolicy(2).
Author: Jakub Wartak <jakub.war...@enterprisedb.com>
Co-authored-by: Bertrand Drouvot <bertranddrouvot...@gmail.com>
Reviewed-by: Andres Freund <and...@anarazel.de>
Reviewed-by: Álvaro Herrera <alvhe...@alvh.no-ip.org>
Reviewed-by: Tomas Vondra <to...@vondra.me>
Discussion: https://postgr.es/m/CAKZiRmxh6KWo0aqRqvmcoaX2jUxZYb4kGp3N%3Dq1w%2BDiH-696Xw%40mail.gmail.com
---
.cirrus.tasks.yml | 2 +
configure | 187 ++++++++++++++++++++++++++++
configure.ac | 14 +++
doc/src/sgml/func.sgml | 13 ++
doc/src/sgml/installation.sgml | 21 ++++
meson.build | 23 ++++
meson_options.txt | 3 +
src/Makefile.global.in | 6 +-
src/backend/utils/misc/guc_tables.c | 2 +-
src/include/catalog/pg_proc.dat | 4 +
src/include/pg_config.h.in | 3 +
src/include/port/pg_numa.h | 40 ++++++
src/include/storage/pg_shmem.h | 1 +
src/makefiles/meson.build | 3 +
src/port/Makefile | 1 +
src/port/meson.build | 1 +
src/port/pg_numa.c | 120 ++++++++++++++++++
17 files changed, 442 insertions(+), 2 deletions(-)
create mode 100644 src/include/port/pg_numa.h
create mode 100644 src/port/pg_numa.c
diff --git a/.cirrus.tasks.yml b/.cirrus.tasks.yml
index 86a1fa9bbdb..6f4f5c674a1 100644
--- a/.cirrus.tasks.yml
+++ b/.cirrus.tasks.yml
@@ -471,6 +471,7 @@ task:
--enable-cassert --enable-injection-points --enable-debug \
--enable-tap-tests --enable-nls \
--with-segsize-blocks=6 \
+ --with-libnuma \
--with-liburing \
\
${LINUX_CONFIGURE_FEATURES} \
@@ -523,6 +524,7 @@ task:
-Dllvm=disabled \
--pkg-config-path /usr/lib/i386-linux-gnu/pkgconfig/ \
-DPERL=perl5.36-i386-linux-gnu \
+ -Dlibnuma=disabled \
build-32
EOF
diff --git a/configure b/configure
index 11615d1122d..e27badd83c3 100755
--- a/configure
+++ b/configure
@@ -708,6 +708,9 @@ XML2_LIBS
XML2_CFLAGS
XML2_CONFIG
with_libxml
+LIBNUMA_LIBS
+LIBNUMA_CFLAGS
+with_libnuma
LIBCURL_LIBS
LIBCURL_CFLAGS
with_libcurl
@@ -872,6 +875,7 @@ with_liburing
with_uuid
with_ossp_uuid
with_libcurl
+with_libnuma
with_libxml
with_libxslt
with_system_tzdata
@@ -906,6 +910,8 @@ LIBURING_CFLAGS
LIBURING_LIBS
LIBCURL_CFLAGS
LIBCURL_LIBS
+LIBNUMA_CFLAGS
+LIBNUMA_LIBS
XML2_CONFIG
XML2_CFLAGS
XML2_LIBS
@@ -1588,6 +1594,7 @@ Optional Packages:
--with-uuid=LIB build contrib/uuid-ossp using LIB (bsd,e2fs,ossp)
--with-ossp-uuid obsolete spelling of --with-uuid=ossp
--with-libcurl build with libcurl support
+ --with-libnuma build with libnuma support
--with-libxml build with XML support
--with-libxslt use XSLT support when building contrib/xml2
--with-system-tzdata=DIR
@@ -1629,6 +1636,10 @@ Some influential environment variables:
C compiler flags for LIBCURL, overriding pkg-config
LIBCURL_LIBS
linker flags for LIBCURL, overriding pkg-config
+ LIBNUMA_CFLAGS
+ C compiler flags for LIBNUMA, overriding pkg-config
+ LIBNUMA_LIBS
+ linker flags for LIBNUMA, overriding pkg-config
XML2_CONFIG path to xml2-config utility
XML2_CFLAGS C compiler flags for XML2, overriding pkg-config
XML2_LIBS linker flags for XML2, overriding pkg-config
@@ -9063,6 +9074,182 @@ $as_echo "$as_me: WARNING: *** OAuth support tests require --with-python to run"
fi
+#
+# libnuma
+#
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether to build with libnuma support" >&5
+$as_echo_n "checking whether to build with libnuma support... " >&6; }
+
+
+
+# Check whether --with-libnuma was given.
+if test "${with_libnuma+set}" = set; then :
+ withval=$with_libnuma;
+ case $withval in
+ yes)
+
+$as_echo "#define USE_LIBNUMA 1" >>confdefs.h
+
+ ;;
+ no)
+ :
+ ;;
+ *)
+ as_fn_error $? "no argument expected for --with-libnuma option" "$LINENO" 5
+ ;;
+ esac
+
+else
+ with_libnuma=no
+
+fi
+
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $with_libnuma" >&5
+$as_echo "$with_libnuma" >&6; }
+
+
+if test "$with_libnuma" = yes ; then
+ { $as_echo "$as_me:${as_lineno-$LINENO}: checking for numa_available in -lnuma" >&5
+$as_echo_n "checking for numa_available in -lnuma... " >&6; }
+if ${ac_cv_lib_numa_numa_available+:} false; then :
+ $as_echo_n "(cached) " >&6
+else
+ ac_check_lib_save_LIBS=$LIBS
+LIBS="-lnuma $LIBS"
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h. */
+
+/* Override any GCC internal prototype to avoid an error.
+ Use char because int might match the return type of a GCC
+ builtin and then its argument prototype would still apply. */
+#ifdef __cplusplus
+extern "C"
+#endif
+char numa_available ();
+int
+main ()
+{
+return numa_available ();
+ ;
+ return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+ ac_cv_lib_numa_numa_available=yes
+else
+ ac_cv_lib_numa_numa_available=no
+fi
+rm -f core conftest.err conftest.$ac_objext \
+ conftest$ac_exeext conftest.$ac_ext
+LIBS=$ac_check_lib_save_LIBS
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_cv_lib_numa_numa_available" >&5
+$as_echo "$ac_cv_lib_numa_numa_available" >&6; }
+if test "x$ac_cv_lib_numa_numa_available" = xyes; then :
+ cat >>confdefs.h <<_ACEOF
+#define HAVE_LIBNUMA 1
+_ACEOF
+
+ LIBS="-lnuma $LIBS"
+
+else
+ as_fn_error $? "library 'libnuma' is required for NUMA support" "$LINENO" 5
+fi
+
+
+pkg_failed=no
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for numa" >&5
+$as_echo_n "checking for numa... " >&6; }
+
+if test -n "$LIBNUMA_CFLAGS"; then
+ pkg_cv_LIBNUMA_CFLAGS="$LIBNUMA_CFLAGS"
+ elif test -n "$PKG_CONFIG"; then
+ if test -n "$PKG_CONFIG" && \
+ { { $as_echo "$as_me:${as_lineno-$LINENO}: \$PKG_CONFIG --exists --print-errors \"numa\""; } >&5
+ ($PKG_CONFIG --exists --print-errors "numa") 2>&5
+ ac_status=$?
+ $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+ test $ac_status = 0; }; then
+ pkg_cv_LIBNUMA_CFLAGS=`$PKG_CONFIG --cflags "numa" 2>/dev/null`
+ test "x$?" != "x0" && pkg_failed=yes
+else
+ pkg_failed=yes
+fi
+ else
+ pkg_failed=untried
+fi
+if test -n "$LIBNUMA_LIBS"; then
+ pkg_cv_LIBNUMA_LIBS="$LIBNUMA_LIBS"
+ elif test -n "$PKG_CONFIG"; then
+ if test -n "$PKG_CONFIG" && \
+ { { $as_echo "$as_me:${as_lineno-$LINENO}: \$PKG_CONFIG --exists --print-errors \"numa\""; } >&5
+ ($PKG_CONFIG --exists --print-errors "numa") 2>&5
+ ac_status=$?
+ $as_echo "$as_me:${as_lineno-$LINENO}: \$? = $ac_status" >&5
+ test $ac_status = 0; }; then
+ pkg_cv_LIBNUMA_LIBS=`$PKG_CONFIG --libs "numa" 2>/dev/null`
+ test "x$?" != "x0" && pkg_failed=yes
+else
+ pkg_failed=yes
+fi
+ else
+ pkg_failed=untried
+fi
+
+
+
+if test $pkg_failed = yes; then
+ { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+
+if $PKG_CONFIG --atleast-pkgconfig-version 0.20; then
+ _pkg_short_errors_supported=yes
+else
+ _pkg_short_errors_supported=no
+fi
+ if test $_pkg_short_errors_supported = yes; then
+ LIBNUMA_PKG_ERRORS=`$PKG_CONFIG --short-errors --print-errors --cflags --libs "numa" 2>&1`
+ else
+ LIBNUMA_PKG_ERRORS=`$PKG_CONFIG --print-errors --cflags --libs "numa" 2>&1`
+ fi
+ # Put the nasty error message in config.log where it belongs
+ echo "$LIBNUMA_PKG_ERRORS" >&5
+
+ as_fn_error $? "Package requirements (numa) were not met:
+
+$LIBNUMA_PKG_ERRORS
+
+Consider adjusting the PKG_CONFIG_PATH environment variable if you
+installed software in a non-standard prefix.
+
+Alternatively, you may set the environment variables LIBNUMA_CFLAGS
+and LIBNUMA_LIBS to avoid the need to call pkg-config.
+See the pkg-config man page for more details." "$LINENO" 5
+elif test $pkg_failed = untried; then
+ { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
+$as_echo "no" >&6; }
+ { { $as_echo "$as_me:${as_lineno-$LINENO}: error: in \`$ac_pwd':" >&5
+$as_echo "$as_me: error: in \`$ac_pwd':" >&2;}
+as_fn_error $? "The pkg-config script could not be found or is too old. Make sure it
+is in your PATH or set the PKG_CONFIG environment variable to the full
+path to pkg-config.
+
+Alternatively, you may set the environment variables LIBNUMA_CFLAGS
+and LIBNUMA_LIBS to avoid the need to call pkg-config.
+See the pkg-config man page for more details.
+
+To get pkg-config, see <http://pkg-config.freedesktop.org/>.
+See \`config.log' for more details" "$LINENO" 5; }
+else
+ LIBNUMA_CFLAGS=$pkg_cv_LIBNUMA_CFLAGS
+ LIBNUMA_LIBS=$pkg_cv_LIBNUMA_LIBS
+ { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
+$as_echo "yes" >&6; }
+
+fi
+fi
+
#
# XML
#
diff --git a/configure.ac b/configure.ac
index debdf165044..d365a486d3d 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1053,6 +1053,20 @@ if test "$with_libcurl" = yes ; then
fi
+#
+# libnuma
+#
+AC_MSG_CHECKING([whether to build with libnuma support])
+PGAC_ARG_BOOL(with, libnuma, no, [build with libnuma support],
+ [AC_DEFINE([USE_LIBNUMA], 1, [Define to build with NUMA support. (--with-libnuma)])])
+AC_MSG_RESULT([$with_libnuma])
+AC_SUBST(with_libnuma)
+
+if test "$with_libnuma" = yes ; then
+ AC_CHECK_LIB(numa, numa_available, [], [AC_MSG_ERROR([library 'libnuma' is required for NUMA support])])
+ PKG_CHECK_MODULES(LIBNUMA, numa)
+fi
+
#
# XML
#
diff --git a/doc/src/sgml/func.sgml b/doc/src/sgml/func.sgml
index 2488e9ba998..4bb60e9e080 100644
--- a/doc/src/sgml/func.sgml
+++ b/doc/src/sgml/func.sgml
@@ -25143,6 +25143,19 @@ SELECT * FROM pg_ls_dir('.') WITH ORDINALITY AS t(ls,n);
</para></entry>
</row>
+ <row>
+ <entry role="func_table_entry"><para role="func_signature">
+ <indexterm>
+ <primary>pg_numa_available</primary>
+ </indexterm>
+ <function>pg_numa_available</function> ()
+ <returnvalue>boolean</returnvalue>
+ </para>
+ <para>
+ Returns true if the server has been compiled with <acronym>NUMA</acronym> support.
+ </para></entry>
+ </row>
+
<row>
<entry role="func_table_entry"><para role="func_signature">
<indexterm>
diff --git a/doc/src/sgml/installation.sgml b/doc/src/sgml/installation.sgml
index cc28f041330..8ebf0b03ec0 100644
--- a/doc/src/sgml/installation.sgml
+++ b/doc/src/sgml/installation.sgml
@@ -1156,6 +1156,16 @@ build-postgresql:
</listitem>
</varlistentry>
+ <varlistentry id="configure-option-with-libnuma">
+ <term><option>--with-libnuma</option></term>
+ <listitem>
+ <para>
+ Build with libnuma support for basic NUMA support.
+ Only supported on platforms for which the <productname>libnuma</productname> library is implemented.
+ </para>
+ </listitem>
+ </varlistentry>
+
<varlistentry id="configure-option-with-liburing">
<term><option>--with-liburing</option></term>
<listitem>
@@ -2645,6 +2655,17 @@ ninja install
</listitem>
</varlistentry>
+ <varlistentry id="configure-with-libnuma-meson">
+ <term><option>-Dlibnuma={ auto | enabled | disabled }</option></term>
+ <listitem>
+ <para>
+ Build with libnuma support for basic NUMA support.
+ Only supported on platforms for which the <productname>libnuma</productname> library is implemented.
+ The default for this option is auto.
+ </para>
+ </listitem>
+ </varlistentry>
+
<varlistentry id="configure-with-libxml-meson">
<term><option>-Dlibxml={ auto | enabled | disabled }</option></term>
<listitem>
diff --git a/meson.build b/meson.build
index 454ed81f5ea..46e92daeb62 100644
--- a/meson.build
+++ b/meson.build
@@ -943,6 +943,27 @@ else
endif
+###############################################################
+# Library: libnuma
+###############################################################
+
+libnumaopt = get_option('libnuma')
+if not libnumaopt.disabled()
+ # via pkg-config
+ libnuma = dependency('numa', required: libnumaopt)
+ if not libnuma.found()
+ libnuma = cc.find_library('numa', required: libnumaopt)
+ endif
+ if not cc.has_header('numa.h', dependencies: libnuma, required: libnumaopt)
+ libnuma = not_found_dep
+ endif
+ if libnuma.found()
+ cdata.set('USE_LIBNUMA', 1)
+ endif
+else
+ libnuma = not_found_dep
+endif
+
###############################################################
# Library: liburing
@@ -3243,6 +3264,7 @@ backend_both_deps += [
icu_i18n,
ldap,
libintl,
+ libnuma,
liburing,
libxml,
lz4,
@@ -3899,6 +3921,7 @@ if meson.version().version_compare('>=0.57')
'icu': icu,
'ldap': ldap,
'libcurl': libcurl,
+ 'libnuma': libnuma,
'liburing': liburing,
'libxml': libxml,
'libxslt': libxslt,
diff --git a/meson_options.txt b/meson_options.txt
index dd7126da3a7..06bf5627d3c 100644
--- a/meson_options.txt
+++ b/meson_options.txt
@@ -106,6 +106,9 @@ option('libcurl', type : 'feature', value: 'auto',
option('libedit_preferred', type: 'boolean', value: false,
description: 'Prefer BSD Libedit over GNU Readline')
+option('libnuma', type: 'feature', value: 'auto',
+ description: 'NUMA support')
+
option('liburing', type : 'feature', value: 'auto',
description: 'io_uring support, for asynchronous I/O')
diff --git a/src/Makefile.global.in b/src/Makefile.global.in
index 737b2dd1869..6722fbdf365 100644
--- a/src/Makefile.global.in
+++ b/src/Makefile.global.in
@@ -196,6 +196,7 @@ with_gssapi = @with_gssapi@
with_krb_srvnam = @with_krb_srvnam@
with_ldap = @with_ldap@
with_libcurl = @with_libcurl@
+with_libnuma = @with_libnuma@
with_liburing = @with_liburing@
with_libxml = @with_libxml@
with_libxslt = @with_libxslt@
@@ -223,6 +224,9 @@ krb_srvtab = @krb_srvtab@
ICU_CFLAGS = @ICU_CFLAGS@
ICU_LIBS = @ICU_LIBS@
+LIBNUMA_CFLAGS = @LIBNUMA_CFLAGS@
+LIBNUMA_LIBS = @LIBNUMA_LIBS@
+
LIBURING_CFLAGS = @LIBURING_CFLAGS@
LIBURING_LIBS = @LIBURING_LIBS@
@@ -250,7 +254,7 @@ CPP = @CPP@
CPPFLAGS = @CPPFLAGS@
PG_SYSROOT = @PG_SYSROOT@
-override CPPFLAGS := $(ICU_CFLAGS) $(LIBURING_CFLAGS) $(CPPFLAGS)
+override CPPFLAGS := $(ICU_CFLAGS) $(LIBNUMA_CFLAGS) $(LIBURING_CFLAGS) $(CPPFLAGS)
ifdef PGXS
override CPPFLAGS := -I$(includedir_server) -I$(includedir_internal) $(CPPFLAGS)
diff --git a/src/backend/utils/misc/guc_tables.c b/src/backend/utils/misc/guc_tables.c
index 4eaeca89f2c..ea8d796e7c4 100644
--- a/src/backend/utils/misc/guc_tables.c
+++ b/src/backend/utils/misc/guc_tables.c
@@ -566,7 +566,7 @@ static int ssl_renegotiation_limit;
*/
int huge_pages = HUGE_PAGES_TRY;
int huge_page_size;
-static int huge_pages_status = HUGE_PAGES_UNKNOWN;
+int huge_pages_status = HUGE_PAGES_UNKNOWN;
/*
* These variables are all dummies that don't do anything, except in some
diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat
index a28a15993a2..63859661951 100644
--- a/src/include/catalog/pg_proc.dat
+++ b/src/include/catalog/pg_proc.dat
@@ -8518,6 +8518,10 @@
proargnames => '{name,off,size,allocated_size}',
prosrc => 'pg_get_shmem_allocations' },
+{ oid => '9685', descr => 'Is NUMA compilation available?',
+ proname => 'pg_numa_available', provolatile => 's', prorettype => 'bool',
+ proargtypes => '', prosrc => 'pg_numa_available' },
+
# memory context of local backend
{ oid => '2282',
descr => 'information about all memory contexts of local backend',
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index c2f1241b234..b3166ec8f42 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -686,6 +686,9 @@
/* Define to 1 to build with libcurl support. (--with-libcurl) */
#undef USE_LIBCURL
+/* Define to 1 to build with NUMA support. (--with-libnuma) */
+#undef USE_LIBNUMA
+
/* Define to build with io_uring support. (--with-liburing) */
#undef USE_LIBURING
diff --git a/src/include/port/pg_numa.h b/src/include/port/pg_numa.h
new file mode 100644
index 00000000000..314cff94dbc
--- /dev/null
+++ b/src/include/port/pg_numa.h
@@ -0,0 +1,40 @@
+/*-------------------------------------------------------------------------
+ *
+ * pg_numa.h
+ * Basic NUMA portability routines
+ *
+ *
+ * Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ * IDENTIFICATION
+ * src/include/port/pg_numa.h
+ *
+ *-------------------------------------------------------------------------
+ */
+#ifndef PG_NUMA_H
+#define PG_NUMA_H
+
+#include "fmgr.h"
+
+extern PGDLLIMPORT int pg_numa_init(void);
+extern PGDLLIMPORT int pg_numa_query_pages(int pid, unsigned long count, void **pages, int *status);
+extern PGDLLIMPORT int pg_numa_get_max_node(void);
+extern PGDLLIMPORT Size pg_numa_get_pagesize(void);
+
+#ifdef USE_LIBNUMA
+
+/*
+ * This is required on Linux, before pg_numa_query_pages() as we
+ * need to page-fault before move_pages(2) syscall returns valid results.
+ */
+#define pg_numa_touch_mem_if_required(ro_volatile_var, ptr) \
+ ro_volatile_var = *(uint64 *)ptr
+
+#else
+
+#define pg_numa_touch_mem_if_required(ro_volatile_var, ptr) \
+ do {} while(0)
+
+#endif
+
+#endif /* PG_NUMA_H */
diff --git a/src/include/storage/pg_shmem.h b/src/include/storage/pg_shmem.h
index b99ebc9e86f..5f7d4b83a60 100644
--- a/src/include/storage/pg_shmem.h
+++ b/src/include/storage/pg_shmem.h
@@ -45,6 +45,7 @@ typedef struct PGShmemHeader /* standard header for all Postgres shmem */
extern PGDLLIMPORT int shared_memory_type;
extern PGDLLIMPORT int huge_pages;
extern PGDLLIMPORT int huge_page_size;
+extern PGDLLIMPORT int huge_pages_status;
/* Possible values for huge_pages and huge_pages_status */
typedef enum
diff --git a/src/makefiles/meson.build b/src/makefiles/meson.build
index 46d8da070e8..55da678ec27 100644
--- a/src/makefiles/meson.build
+++ b/src/makefiles/meson.build
@@ -200,6 +200,8 @@ pgxs_empty = [
'ICU_LIBS',
+ 'LIBNUMA_CFLAGS', 'LIBNUMA_LIBS',
+
'LIBURING_CFLAGS', 'LIBURING_LIBS',
]
@@ -232,6 +234,7 @@ pgxs_deps = {
'icu': icu,
'ldap': ldap,
'libcurl': libcurl,
+ 'libnuma': libnuma,
'liburing': liburing,
'libxml': libxml,
'libxslt': libxslt,
diff --git a/src/port/Makefile b/src/port/Makefile
index f11896440d5..4274949dfa4 100644
--- a/src/port/Makefile
+++ b/src/port/Makefile
@@ -45,6 +45,7 @@ OBJS = \
path.o \
pg_bitutils.o \
pg_localeconv_r.o \
+ pg_numa.o \
pg_popcount_aarch64.o \
pg_popcount_avx512.o \
pg_strong_random.o \
diff --git a/src/port/meson.build b/src/port/meson.build
index 51041e75609..228888b2f66 100644
--- a/src/port/meson.build
+++ b/src/port/meson.build
@@ -8,6 +8,7 @@ pgport_sources = [
'path.c',
'pg_bitutils.c',
'pg_localeconv_r.c',
+ 'pg_numa.c',
'pg_popcount_aarch64.c',
'pg_popcount_avx512.c',
'pg_strong_random.c',
diff --git a/src/port/pg_numa.c b/src/port/pg_numa.c
new file mode 100644
index 00000000000..5e2523cf798
--- /dev/null
+++ b/src/port/pg_numa.c
@@ -0,0 +1,120 @@
+/*-------------------------------------------------------------------------
+ *
+ * pg_numa.c
+ * Basic NUMA portability routines
+ *
+ *
+ * Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ *
+ * IDENTIFICATION
+ * src/port/pg_numa.c
+ *
+ *-------------------------------------------------------------------------
+ */
+
+#include "postgres.h"
+#include <unistd.h>
+
+#ifdef WIN32
+#include <windows.h>
+#endif
+
+#include "fmgr.h"
+#include "miscadmin.h"
+#include "port/pg_numa.h"
+#include "storage/pg_shmem.h"
+
+/*
+ * At this point we provide support only for Linux thanks to libnuma, but in
+ * future support for other platforms e.g. Win32 or FreeBSD might be possible
+ * too. For Win32 NUMA APIs see
+ * https://learn.microsoft.com/en-us/windows/win32/procthread/numa-support
+ */
+#ifdef USE_LIBNUMA
+
+#include <numa.h>
+#include <numaif.h>
+
+Datum pg_numa_available(PG_FUNCTION_ARGS);
+
+/* libnuma requires initialization as per numa(3) on Linux */
+int
+pg_numa_init(void)
+{
+ int r = numa_available();
+
+ return r;
+}
+
+/*
+ * We use move_pages(2) syscall here - instead of get_mempolicy(2) - as the
+ * first one allows us to batch and query about many memory pages in one single
+ * giant system call that is way faster.
+ */
+int
+pg_numa_query_pages(int pid, unsigned long count, void **pages, int *status)
+{
+ return numa_move_pages(pid, count, pages, NULL, status, 0);
+}
+
+int
+pg_numa_get_max_node(void)
+{
+ return numa_max_node();
+}
+
+#else
+
+Datum pg_numa_available(PG_FUNCTION_ARGS);
+
+/* Empty wrappers */
+int
+pg_numa_init(void)
+{
+ /* We state that NUMA is not available */
+ return -1;
+}
+
+int
+pg_numa_query_pages(int pid, unsigned long count, void **pages, int *status)
+{
+ return 0;
+}
+
+int
+pg_numa_get_max_node(void)
+{
+ return 0;
+}
+
+#endif
+
+Datum
+pg_numa_available(PG_FUNCTION_ARGS)
+{
+ PG_RETURN_BOOL(pg_numa_init() != -1);
+}
+
+/* This should be used only after the server is started */
+Size
+pg_numa_get_pagesize(void)
+{
+ Size os_page_size;
+#ifdef WIN32
+ SYSTEM_INFO sysinfo;
+
+ GetSystemInfo(&sysinfo);
+ os_page_size = sysinfo.dwPageSize;
+#else
+ os_page_size = sysconf(_SC_PAGESIZE);
+#endif
+
+ Assert(IsUnderPostmaster);
+ Assert(huge_pages_status != HUGE_PAGES_UNKNOWN);
+
+ if (huge_pages_status == HUGE_PAGES_ON)
+ GetHugePageSize(&os_page_size, NULL);
+
+ return os_page_size;
+}
--
2.49.0
From 70e9617f969c8b4cbb27a6d10afaf0cb6ad90206 Mon Sep 17 00:00:00 2001
From: Tomas Vondra <to...@vondra.me>
Date: Thu, 3 Apr 2025 20:21:25 +0200
Subject: [PATCH v24 2/3] Add pg_buffercache_numa view with NUMA node info
Introduces a new view pg_buffercache_numa, showing a NUMA memory node
for each individual buffer.
To determine the NUMA node for a buffer, we first need to touch the
memory pages using pg_numa_touch_mem_if_required, otherwise we might get
status -2 (ENOENT = The page is not present), indicating the page is
either unmapped or unallocated.
The size of a database block and OS memory page may differ. For example
the default block size (BLCKSZ) is 8KB, while the memory page is 4KB,
but it's also possible to make the block size smaller (e.g. 1KB).
Author: Jakub Wartak <jakub.war...@enterprisedb.com>
Reviewed-by: Andres Freund <and...@anarazel.de>
Reviewed-by: Bertrand Drouvot <bertranddrouvot...@gmail.com>
Reviewed-by: Tomas Vondra <to...@vondra.me>
Discussion: https://postgr.es/m/CAKZiRmxh6KWo0aqRqvmcoaX2jUxZYb4kGp3N%3Dq1w%2BDiH-696Xw%40mail.gmail.com
---
contrib/pg_buffercache/Makefile | 5 +-
.../expected/pg_buffercache_numa.out | 28 ++
.../expected/pg_buffercache_numa_1.out | 3 +
contrib/pg_buffercache/meson.build | 2 +
.../pg_buffercache--1.5--1.6.sql | 22 ++
contrib/pg_buffercache/pg_buffercache.control | 2 +-
contrib/pg_buffercache/pg_buffercache_pages.c | 276 ++++++++++++++++++
.../sql/pg_buffercache_numa.sql | 20 ++
doc/src/sgml/pgbuffercache.sgml | 74 ++++-
9 files changed, 428 insertions(+), 4 deletions(-)
create mode 100644 contrib/pg_buffercache/expected/pg_buffercache_numa.out
create mode 100644 contrib/pg_buffercache/expected/pg_buffercache_numa_1.out
create mode 100644 contrib/pg_buffercache/pg_buffercache--1.5--1.6.sql
create mode 100644 contrib/pg_buffercache/sql/pg_buffercache_numa.sql
diff --git a/contrib/pg_buffercache/Makefile b/contrib/pg_buffercache/Makefile
index eae65ead9e5..5f748543e2e 100644
--- a/contrib/pg_buffercache/Makefile
+++ b/contrib/pg_buffercache/Makefile
@@ -8,10 +8,11 @@ OBJS = \
EXTENSION = pg_buffercache
DATA = pg_buffercache--1.2.sql pg_buffercache--1.2--1.3.sql \
pg_buffercache--1.1--1.2.sql pg_buffercache--1.0--1.1.sql \
- pg_buffercache--1.3--1.4.sql pg_buffercache--1.4--1.5.sql
+ pg_buffercache--1.3--1.4.sql pg_buffercache--1.4--1.5.sql \
+ pg_buffercache--1.5--1.6.sql
PGFILEDESC = "pg_buffercache - monitoring of shared buffer cache in real-time"
-REGRESS = pg_buffercache
+REGRESS = pg_buffercache pg_buffercache_numa
ifdef USE_PGXS
PG_CONFIG = pg_config
diff --git a/contrib/pg_buffercache/expected/pg_buffercache_numa.out b/contrib/pg_buffercache/expected/pg_buffercache_numa.out
new file mode 100644
index 00000000000..d4de5ea52fc
--- /dev/null
+++ b/contrib/pg_buffercache/expected/pg_buffercache_numa.out
@@ -0,0 +1,28 @@
+SELECT NOT(pg_numa_available()) AS skip_test \gset
+\if :skip_test
+\quit
+\endif
+select count(*) = (select setting::bigint
+ from pg_settings
+ where name = 'shared_buffers')
+from pg_buffercache_numa;
+ ?column?
+----------
+ t
+(1 row)
+
+-- Check that the functions / views can't be accessed by default. To avoid
+-- having to create a dedicated user, use the pg_database_owner pseudo-role.
+SET ROLE pg_database_owner;
+SELECT count(*) > 0 FROM pg_buffercache_numa;
+ERROR: permission denied for view pg_buffercache_numa
+RESET role;
+-- Check that pg_monitor is allowed to query view / function
+SET ROLE pg_monitor;
+SELECT count(*) > 0 FROM pg_buffercache_numa;
+ ?column?
+----------
+ t
+(1 row)
+
+RESET role;
diff --git a/contrib/pg_buffercache/expected/pg_buffercache_numa_1.out b/contrib/pg_buffercache/expected/pg_buffercache_numa_1.out
new file mode 100644
index 00000000000..6dd6824b4e4
--- /dev/null
+++ b/contrib/pg_buffercache/expected/pg_buffercache_numa_1.out
@@ -0,0 +1,3 @@
+SELECT NOT(pg_numa_available()) AS skip_test \gset
+\if :skip_test
+\quit
diff --git a/contrib/pg_buffercache/meson.build b/contrib/pg_buffercache/meson.build
index 12d1fe48717..7cd039a1df9 100644
--- a/contrib/pg_buffercache/meson.build
+++ b/contrib/pg_buffercache/meson.build
@@ -23,6 +23,7 @@ install_data(
'pg_buffercache--1.2.sql',
'pg_buffercache--1.3--1.4.sql',
'pg_buffercache--1.4--1.5.sql',
+ 'pg_buffercache--1.5--1.6.sql',
'pg_buffercache.control',
kwargs: contrib_data_args,
)
@@ -34,6 +35,7 @@ tests += {
'regress': {
'sql': [
'pg_buffercache',
+ 'pg_buffercache_numa',
],
},
}
diff --git a/contrib/pg_buffercache/pg_buffercache--1.5--1.6.sql b/contrib/pg_buffercache/pg_buffercache--1.5--1.6.sql
new file mode 100644
index 00000000000..1230e244a5f
--- /dev/null
+++ b/contrib/pg_buffercache/pg_buffercache--1.5--1.6.sql
@@ -0,0 +1,22 @@
+/* contrib/pg_buffercache/pg_buffercache--1.5--1.6.sql */
+
+-- complain if script is sourced in psql, rather than via CREATE EXTENSION
+\echo Use "ALTER EXTENSION pg_buffercache UPDATE TO '1.6'" to load this file. \quit
+
+-- Register the new functions.
+CREATE OR REPLACE FUNCTION pg_buffercache_numa_pages()
+RETURNS SETOF RECORD
+AS 'MODULE_PATHNAME', 'pg_buffercache_numa_pages'
+LANGUAGE C PARALLEL SAFE;
+
+-- Create a view for convenient access.
+CREATE OR REPLACE VIEW pg_buffercache_numa AS
+ SELECT P.* FROM pg_buffercache_numa_pages() AS P
+ (bufferid integer, page_num int4, node_id int4);
+
+-- Don't want these to be available to public.
+REVOKE ALL ON FUNCTION pg_buffercache_numa_pages() FROM PUBLIC;
+REVOKE ALL ON pg_buffercache_numa FROM PUBLIC;
+
+GRANT EXECUTE ON FUNCTION pg_buffercache_numa_pages() TO pg_monitor;
+GRANT SELECT ON pg_buffercache_numa TO pg_monitor;
diff --git a/contrib/pg_buffercache/pg_buffercache.control b/contrib/pg_buffercache/pg_buffercache.control
index 5ee875f77dd..b030ba3a6fa 100644
--- a/contrib/pg_buffercache/pg_buffercache.control
+++ b/contrib/pg_buffercache/pg_buffercache.control
@@ -1,5 +1,5 @@
# pg_buffercache extension
comment = 'examine the shared buffer cache'
-default_version = '1.5'
+default_version = '1.6'
module_pathname = '$libdir/pg_buffercache'
relocatable = true
diff --git a/contrib/pg_buffercache/pg_buffercache_pages.c b/contrib/pg_buffercache/pg_buffercache_pages.c
index 62602af1775..d653f4af394 100644
--- a/contrib/pg_buffercache/pg_buffercache_pages.c
+++ b/contrib/pg_buffercache/pg_buffercache_pages.c
@@ -11,6 +11,7 @@
#include "access/htup_details.h"
#include "catalog/pg_type.h"
#include "funcapi.h"
+#include "port/pg_numa.h"
#include "storage/buf_internals.h"
#include "storage/bufmgr.h"
@@ -20,6 +21,8 @@
#define NUM_BUFFERCACHE_SUMMARY_ELEM 5
#define NUM_BUFFERCACHE_USAGE_COUNTS_ELEM 4
+#define NUM_BUFFERCACHE_NUMA_ELEM 3
+
PG_MODULE_MAGIC_EXT(
.name = "pg_buffercache",
.version = PG_VERSION
@@ -59,15 +62,41 @@ typedef struct
} BufferCachePagesContext;
+typedef struct
+{
+ uint32 bufferid;
+ int32 numa_page;
+ int32 numa_node;
+} BufferCacheNumaRec;
+
+/*
+ * Function context for data persisting over repeated calls.
+ */
+typedef struct
+{
+ TupleDesc tupdesc;
+ int buffers_per_page;
+ int pages_per_buffer;
+ int os_page_size;
+ BufferCacheNumaRec *record;
+} BufferCacheNumaContext;
+
+
/*
* Function returning data from the shared buffer cache - buffer number,
* relation node/tablespace/database/blocknum and dirty indicator.
*/
PG_FUNCTION_INFO_V1(pg_buffercache_pages);
+PG_FUNCTION_INFO_V1(pg_buffercache_numa_pages);
PG_FUNCTION_INFO_V1(pg_buffercache_summary);
PG_FUNCTION_INFO_V1(pg_buffercache_usage_counts);
PG_FUNCTION_INFO_V1(pg_buffercache_evict);
+
+/* Only need to touch memory once per backend process lifetime */
+static bool firstNumaTouch = true;
+
+
Datum
pg_buffercache_pages(PG_FUNCTION_ARGS)
{
@@ -246,6 +275,253 @@ pg_buffercache_pages(PG_FUNCTION_ARGS)
SRF_RETURN_DONE(funcctx);
}
+/*
+ * Inquire about NUMA memory mappings, especially the NUMA node.
+ *
+ * In order to get reliable results we also need to touch memory pages, so
+ * that the inquiry about NUMA memory node doesn't return -2 (which indicates
+ * unmapped/unallocated pages).
+ */
+Datum
+pg_buffercache_numa_pages(PG_FUNCTION_ARGS)
+{
+ FuncCallContext *funcctx;
+ MemoryContext oldcontext;
+ BufferCacheNumaContext *fctx; /* User function context. */
+ TupleDesc tupledesc;
+ TupleDesc expected_tupledesc;
+ HeapTuple tuple;
+ Datum result;
+
+ if (SRF_IS_FIRSTCALL())
+ {
+ int i,
+ j,
+ idx;
+ Size os_page_size = 0;
+ void **os_page_ptrs = NULL;
+ int *os_page_status;
+ uint64 os_page_count;
+ int pages_per_buffer;
+ int buffers_per_page;
+ volatile uint64 touch pg_attribute_unused();
+
+ if (pg_numa_init() == -1)
+ elog(ERROR, "libnuma initialization failed or NUMA is not supported on this platform");
+
+ /*
+ * Different database block sizes (4kB, 8kB, ..., 32kB) can be used,
+ * while the OS may have different memory page sizes.
+ *
+ * To correctly map between them, we need to: 1. Determine the OS
+ * memory page size 2. Calculate how many OS pages are used by all
+ * buffer blocks 3. Calculate how many OS pages are contained within
+ * each database block.
+ *
+ * This information is needed before calling move_pages() for NUMA
+ * node id inquiry.
+ */
+ os_page_size = pg_numa_get_pagesize();
+ buffers_per_page = os_page_size / BLCKSZ;
+ pages_per_buffer = BLCKSZ / os_page_size;
+
+ /*
+ * The pages and block size is expected to be 2^k, so one divides the
+ * other (we don't know in which direction).
+ */
+ Assert((os_page_size % BLCKSZ == 0) || (BLCKSZ % os_page_size == 0));
+
+ /*
+ * Either both counts are 1 (when the pages have the same size), or
+ * exacly one of them is zero. Both can't be zero at the same time.
+ */
+ Assert((buffers_per_page > 0) || (pages_per_buffer > 0));
+ Assert(((buffers_per_page == 1) && (pages_per_buffer == 1)) ||
+ ((buffers_per_page == 0) || (pages_per_buffer == 0)));
+
+ /*
+ * How many addresses we are going to query (store) depends on the
+ * relation between BLCKSZ : PAGESIZE. We need at least one status per
+ * buffer - if the memory page is larger than buffer, we still query
+ * it for each buffer. With multiple memory pages per buffer, we need
+ * that many entries.
+ */
+ os_page_count = NBuffers * Max(1, pages_per_buffer);
+
+ elog(DEBUG1, "NUMA: NBuffers=%d os_page_query_count=" UINT64_FORMAT " os_page_size=%zu buffers_per_page=%d pages_per_buffer=%d",
+ NBuffers, os_page_count, os_page_size, buffers_per_page, pages_per_buffer);
+
+
+ /* initialize the multi-call context */
+
+ funcctx = SRF_FIRSTCALL_INIT();
+
+ /* Switch context when allocating stuff to be used in later calls */
+ oldcontext = MemoryContextSwitchTo(funcctx->multi_call_memory_ctx);
+
+ /* Create a user function context for cross-call persistence */
+ fctx = (BufferCacheNumaContext *) palloc(sizeof(BufferCacheNumaContext));
+
+ /*
+ * To smoothly support upgrades from version 1.0 of this extension
+ * transparently handle the (non-)existence of the pinning_backends
+ * column. We unfortunately have to get the result type for that... -
+ * we can't use the result type determined by the function definition
+ * without potentially crashing when somebody uses the old (or even
+ * wrong) function definition though.
+ */
+ if (get_call_result_type(fcinfo, NULL, &expected_tupledesc) != TYPEFUNC_COMPOSITE)
+ elog(ERROR, "return type must be a row type");
+
+ if (expected_tupledesc->natts != NUM_BUFFERCACHE_NUMA_ELEM)
+ elog(ERROR, "incorrect number of output arguments");
+
+ /* Construct a tuple descriptor for the result rows. */
+ tupledesc = CreateTemplateTupleDesc(expected_tupledesc->natts);
+ TupleDescInitEntry(tupledesc, (AttrNumber) 1, "bufferid",
+ INT4OID, -1, 0);
+ TupleDescInitEntry(tupledesc, (AttrNumber) 2, "page_num",
+ INT4OID, -1, 0);
+ TupleDescInitEntry(tupledesc, (AttrNumber) 3, "node_id",
+ INT4OID, -1, 0);
+
+ fctx->tupdesc = BlessTupleDesc(tupledesc);
+
+ /* Allocate NBuffers worth of BufferCachePagesRec records. */
+ fctx->record = (BufferCacheNumaRec *)
+ MemoryContextAllocHuge(CurrentMemoryContext,
+ sizeof(BufferCacheNumaRec) * os_page_count);
+
+ /* Set max calls and remember the user function context. */
+ funcctx->max_calls = NBuffers;
+ funcctx->user_fctx = fctx;
+
+ /* Return to original context when allocating transient memory */
+ MemoryContextSwitchTo(oldcontext);
+
+
+ /* determine the NUMA node for OS pages */
+ os_page_ptrs = palloc0(sizeof(void *) * os_page_count);
+ os_page_status = palloc(sizeof(uint64) * os_page_count);
+
+ /*
+ * If we ever get 0xff back from kernel inquiry, then we probably have
+ * bug in our buffers to OS page mapping code here.
+ */
+ memset(os_page_status, 0xff, sizeof(int) * os_page_count);
+
+ if (firstNumaTouch)
+ elog(DEBUG1, "NUMA: page-faulting the buffercache for proper NUMA readouts");
+
+ /*
+ * Scan through all the buffers, saving the relevant fields in the
+ * fctx->record structure.
+ *
+ * We don't hold the partition locks, so we don't get a consistent
+ * snapshot across all buffers, but we do grab the buffer header
+ * locks, so the information of each buffer is self-consistent.
+ *
+ * This loop touches and stores addresses into os_page_ptrs[] as input
+ * to one big big move_pages(2) inquiry system call. Basically we ask
+ * for all memory pages for NBuffers.
+ */
+ idx = 0;
+ for (i = 0; i < NBuffers; i++)
+ {
+ BufferDesc *bufHdr;
+ uint32 buf_state;
+ uint32 bufferid;
+
+ CHECK_FOR_INTERRUPTS();
+
+ bufHdr = GetBufferDescriptor(i);
+
+ /* Lock each buffer header before inspecting. */
+ buf_state = LockBufHdr(bufHdr);
+ bufferid = BufferDescriptorGetBuffer(bufHdr);
+
+ UnlockBufHdr(bufHdr, buf_state);
+
+ /*
+ * If we have multiple OS pages per buffer, fill those in too. We
+ * always want at least one OS page, even if there are multiple
+ * buffers per page.
+ *
+ * Altough we could query just once per each OS page, we do it
+ * repeatably for each Buffer and hit the same address as
+ * move_pages(2) requires page aligment. This also simplifies
+ * retrieval code later on. Also NBuffers starts from 1.
+ */
+ for (j = 0; j < Max(1, pages_per_buffer); j++)
+ {
+ char *buffptr = (char *) BufferGetBlock(i + 1);
+
+ fctx->record[idx].bufferid = bufferid;
+ fctx->record[idx].numa_page = j;
+
+ os_page_ptrs[idx]
+ = (char *) TYPEALIGN(os_page_size,
+ buffptr + (os_page_size * j));
+
+ /* Only need to touch memory once per backend process lifetime */
+ if (firstNumaTouch)
+ pg_numa_touch_mem_if_required(touch, os_page_ptrs[idx]);
+
+ ++idx;
+ }
+
+ }
+
+ /* we should get exactly the expected number of entrires */
+ Assert(idx == os_page_count);
+
+ /* query NUMA status for all the pointers */
+ if (pg_numa_query_pages(0, os_page_count, os_page_ptrs, os_page_status) == -1)
+ elog(ERROR, "failed NUMA pages inquiry: %m");
+
+ /*
+ * Update the entries with NUMA node ID. The status array is indexed the
+ * same way as the entry index.
+ */
+ for (i = 0; i < os_page_count; i++)
+ {
+ fctx->record[i].numa_node = os_page_status[i];
+ }
+
+ pfree(os_page_status);
+ pfree(os_page_ptrs);
+ }
+
+ funcctx = SRF_PERCALL_SETUP();
+
+ /* Get the saved state */
+ fctx = funcctx->user_fctx;
+
+ if (funcctx->call_cntr < funcctx->max_calls)
+ {
+ uint32 i = funcctx->call_cntr;
+ Datum values[NUM_BUFFERCACHE_NUMA_ELEM];
+ bool nulls[NUM_BUFFERCACHE_NUMA_ELEM];
+
+ values[0] = Int32GetDatum(fctx->record[i].bufferid);
+ nulls[0] = false;
+
+ values[1] = Int32GetDatum(fctx->record[i].numa_page);
+ nulls[1] = false;
+
+ values[2] = Int32GetDatum(fctx->record[i].numa_node);
+ nulls[2] = false;
+
+ /* Build and return the tuple. */
+ tuple = heap_form_tuple(fctx->tupdesc, values, nulls);
+ result = HeapTupleGetDatum(tuple);
+
+ SRF_RETURN_NEXT(funcctx, result);
+ }
+ else
+ SRF_RETURN_DONE(funcctx);
+}
+
Datum
pg_buffercache_summary(PG_FUNCTION_ARGS)
{
diff --git a/contrib/pg_buffercache/sql/pg_buffercache_numa.sql b/contrib/pg_buffercache/sql/pg_buffercache_numa.sql
new file mode 100644
index 00000000000..2225b879f58
--- /dev/null
+++ b/contrib/pg_buffercache/sql/pg_buffercache_numa.sql
@@ -0,0 +1,20 @@
+SELECT NOT(pg_numa_available()) AS skip_test \gset
+\if :skip_test
+\quit
+\endif
+
+select count(*) = (select setting::bigint
+ from pg_settings
+ where name = 'shared_buffers')
+from pg_buffercache_numa;
+
+-- Check that the functions / views can't be accessed by default. To avoid
+-- having to create a dedicated user, use the pg_database_owner pseudo-role.
+SET ROLE pg_database_owner;
+SELECT count(*) > 0 FROM pg_buffercache_numa;
+RESET role;
+
+-- Check that pg_monitor is allowed to query view / function
+SET ROLE pg_monitor;
+SELECT count(*) > 0 FROM pg_buffercache_numa;
+RESET role;
diff --git a/doc/src/sgml/pgbuffercache.sgml b/doc/src/sgml/pgbuffercache.sgml
index 802a5112d77..59dbbd2b25e 100644
--- a/doc/src/sgml/pgbuffercache.sgml
+++ b/doc/src/sgml/pgbuffercache.sgml
@@ -30,7 +30,9 @@
<para>
This module provides the <function>pg_buffercache_pages()</function>
function (wrapped in the <structname>pg_buffercache</structname> view),
- the <function>pg_buffercache_summary()</function> function, the
+ <function>pg_buffercache_numa_pages()</function> function (wrapped in the
+ <structname>pg_buffercache_numa</structname> view), the
+ <function>pg_buffercache_summary()</function> function, the
<function>pg_buffercache_usage_counts()</function> function and
the <function>pg_buffercache_evict()</function> function.
</para>
@@ -42,6 +44,14 @@
convenient use.
</para>
+ <para>
+ The <function>pg_buffercache_numa_pages()</function> provides the same information
+ as <function>pg_buffercache_pages()</function> but is slower because it also
+ provides the <acronym>NUMA</acronym> node ID per shared buffer entry.
+ The <structname>pg_buffercache_numa</structname> view wraps the function for
+ convenient use.
+ </para>
+
<para>
The <function>pg_buffercache_summary()</function> function returns a single
row summarizing the state of the shared buffer cache.
@@ -200,6 +210,68 @@
</para>
</sect2>
+ <sect2 id="pgbuffercache-pg-buffercache-numa">
+ <title>The <structname>pg_buffercache_numa</structname> View</title>
+
+ <para>
+ The definitions of the columns exposed by the view are shown in <xref linkend="pgbuffercache-numa-columns"/>.
+ </para>
+
+ <table id="pgbuffercache-numa-columns">
+ <title><structname>pg_buffercache_numa</structname> Columns</title>
+ <tgroup cols="1">
+ <thead>
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ Column Type
+ </para>
+ <para>
+ Description
+ </para></entry>
+ </row>
+ </thead>
+
+ <tbody>
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>bufferid</structfield> <type>integer</type>
+ </para>
+ <para>
+ ID, in the range 1..<varname>shared_buffers</varname>
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>page_num</structfield> <type>int</type>
+ </para>
+ <para>
+ number of OS memory page for this buffer
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>node_id</structfield> <type>int</type>
+ </para>
+ <para>
+ ID of NUMA node
+ </para></entry>
+ </row>
+
+ </tbody>
+ </tgroup>
+ </table>
+
+ <para>
+ As <acronym>NUMA</acronym> node ID inquiry for each page requires memory pages
+ to be paged-in, the first execution of this function can take a noticeable
+ amount of time. In all the cases (first execution or not), retrieving this
+ information is costly and querying the view at a high frequency is not recommended.
+ </para>
+
+ </sect2>
+
<sect2 id="pgbuffercache-summary">
<title>The <function>pg_buffercache_summary()</function> Function</title>
--
2.49.0
From b2c8061d09cf13d6713d0b7f41b88b33b0af7710 Mon Sep 17 00:00:00 2001
From: Jakub Wartak <jakub.war...@enterprisedb.com>
Date: Fri, 21 Feb 2025 14:20:18 +0100
Subject: [PATCH v24 3/3] Introduce pg_shmem_allocations_numa view
Introduce new pg_shmem_alloctions_numa view with information about how
shared memory is distributed across NUMA nodes.
Author: Jakub Wartak <jakub.war...@enterprisedb.com>
Reviewed-by: Andres Freund <and...@anarazel.de>
Reviewed-by: Bertrand Drouvot <bertranddrouvot...@gmail.com>
Reviewed-by: Tomas Vondra <to...@vondra.me>
Discussion: https://postgr.es/m/CAKZiRmxh6KWo0aqRqvmcoaX2jUxZYb4kGp3N%3Dq1w%2BDiH-696Xw%40mail.gmail.com
---
doc/src/sgml/system-views.sgml | 79 ++++++++++++++
src/backend/catalog/system_views.sql | 8 ++
src/backend/storage/ipc/shmem.c | 132 +++++++++++++++++++++++
src/include/catalog/pg_proc.dat | 8 ++
src/test/regress/expected/numa.out | 12 +++
src/test/regress/expected/numa_1.out | 3 +
src/test/regress/expected/privileges.out | 16 ++-
src/test/regress/expected/rules.out | 4 +
src/test/regress/parallel_schedule | 2 +-
src/test/regress/sql/numa.sql | 9 ++
src/test/regress/sql/privileges.sql | 6 +-
11 files changed, 274 insertions(+), 5 deletions(-)
create mode 100644 src/test/regress/expected/numa.out
create mode 100644 src/test/regress/expected/numa_1.out
create mode 100644 src/test/regress/sql/numa.sql
diff --git a/doc/src/sgml/system-views.sgml b/doc/src/sgml/system-views.sgml
index 4f336ee0adf..a83365ae24a 100644
--- a/doc/src/sgml/system-views.sgml
+++ b/doc/src/sgml/system-views.sgml
@@ -181,6 +181,11 @@
<entry>shared memory allocations</entry>
</row>
+ <row>
+ <entry><link linkend="view-pg-shmem-allocations-numa"><structname>pg_shmem_allocations_numa</structname></link></entry>
+ <entry>NUMA node mappings for shared memory allocations</entry>
+ </row>
+
<row>
<entry><link linkend="view-pg-stats"><structname>pg_stats</structname></link></entry>
<entry>planner statistics</entry>
@@ -4051,6 +4056,80 @@ SELECT * FROM pg_locks pl LEFT JOIN pg_prepared_xacts ppx
</para>
</sect1>
+ <sect1 id="view-pg-shmem-allocations-numa">
+ <title><structname>pg_shmem_allocations_numa</structname></title>
+
+ <indexterm zone="view-pg-shmem-allocations-numa">
+ <primary>pg_shmem_allocations_numa</primary>
+ </indexterm>
+
+ <para>
+ The <structname>pg_shmem_allocations_numa</structname> shows how shared
+ memory allocations in the server's main shared memory segment are distributed
+ across NUMA nodes. This includes both memory allocated by
+ <productname>PostgreSQL</productname> itself and memory allocated
+ by extensions using the mechanisms detailed in
+ <xref linkend="xfunc-shared-addin" />.
+ </para>
+
+ <para>
+ Note that this view does not include memory allocated using the dynamic
+ shared memory infrastructure.
+ </para>
+
+ <table>
+ <title><structname>pg_shmem_allocations_numa</structname> Columns</title>
+ <tgroup cols="1">
+ <thead>
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ Column Type
+ </para>
+ <para>
+ Description
+ </para></entry>
+ </row>
+ </thead>
+
+ <tbody>
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>name</structfield> <type>text</type>
+ </para>
+ <para>
+ The name of the shared memory allocation.
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>node_id</structfield> <type>int4</type>
+ </para>
+ <para>
+ ID of <acronym>NUMA</acronym> node
+ </para></entry>
+ </row>
+
+ <row>
+ <entry role="catalog_table_entry"><para role="column_definition">
+ <structfield>size</structfield> <type>int4</type>
+ </para>
+ <para>
+ Size of the allocation on this particular NUMA memory node in bytes
+ </para></entry>
+ </row>
+
+ </tbody>
+ </tgroup>
+ </table>
+
+ <para>
+ By default, the <structname>pg_shmem_allocations_numa</structname> view can be
+ read only by superusers or roles with privileges of the
+ <literal>pg_read_all_stats</literal> role.
+ </para>
+ </sect1>
+
<sect1 id="view-pg-stats">
<title><structname>pg_stats</structname></title>
diff --git a/src/backend/catalog/system_views.sql b/src/backend/catalog/system_views.sql
index 273008db37f..08f780a2e63 100644
--- a/src/backend/catalog/system_views.sql
+++ b/src/backend/catalog/system_views.sql
@@ -658,6 +658,14 @@ GRANT SELECT ON pg_shmem_allocations TO pg_read_all_stats;
REVOKE EXECUTE ON FUNCTION pg_get_shmem_allocations() FROM PUBLIC;
GRANT EXECUTE ON FUNCTION pg_get_shmem_allocations() TO pg_read_all_stats;
+CREATE VIEW pg_shmem_allocations_numa AS
+ SELECT * FROM pg_get_shmem_allocations_numa();
+
+REVOKE ALL ON pg_shmem_allocations_numa FROM PUBLIC;
+GRANT SELECT ON pg_shmem_allocations_numa TO pg_read_all_stats;
+REVOKE EXECUTE ON FUNCTION pg_get_shmem_allocations_numa() FROM PUBLIC;
+GRANT EXECUTE ON FUNCTION pg_get_shmem_allocations_numa() TO pg_read_all_stats;
+
CREATE VIEW pg_backend_memory_contexts AS
SELECT * FROM pg_get_backend_memory_contexts();
diff --git a/src/backend/storage/ipc/shmem.c b/src/backend/storage/ipc/shmem.c
index e453f856794..4313e6db62c 100644
--- a/src/backend/storage/ipc/shmem.c
+++ b/src/backend/storage/ipc/shmem.c
@@ -68,6 +68,7 @@
#include "fmgr.h"
#include "funcapi.h"
#include "miscadmin.h"
+#include "port/pg_numa.h"
#include "storage/lwlock.h"
#include "storage/pg_shmem.h"
#include "storage/shmem.h"
@@ -90,6 +91,8 @@ slock_t *ShmemLock; /* spinlock for shared memory and LWLock
static HTAB *ShmemIndex = NULL; /* primary index hashtable for shmem */
+/* To get reliable results for NUMA inquiry we need to "touch pages" once */
+static bool firstNumaTouch = true;
/*
* InitShmemAccess() --- set up basic pointers to shared memory.
@@ -570,3 +573,132 @@ pg_get_shmem_allocations(PG_FUNCTION_ARGS)
return (Datum) 0;
}
+
+/* SQL SRF showing NUMA memory nodes for allocated shared memory */
+Datum
+pg_get_shmem_allocations_numa(PG_FUNCTION_ARGS)
+{
+#define PG_GET_SHMEM_NUMA_SIZES_COLS 3
+ ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
+ HASH_SEQ_STATUS hstat;
+ ShmemIndexEnt *ent;
+ Datum values[PG_GET_SHMEM_NUMA_SIZES_COLS];
+ bool nulls[PG_GET_SHMEM_NUMA_SIZES_COLS];
+ Size os_page_size;
+ void **page_ptrs;
+ int *pages_status;
+ uint64 shm_total_page_count,
+ shm_ent_page_count,
+ max_nodes;
+ Size *nodes;
+
+ InitMaterializedSRF(fcinfo, 0);
+
+ if (pg_numa_init() == -1)
+ {
+ elog(ERROR, "libnuma initialization failed or NUMA is not supported on this platform");
+ return (Datum) 0;
+ }
+ max_nodes = pg_numa_get_max_node();
+ nodes = palloc(sizeof(Size) * (max_nodes + 1));
+
+ /*
+ * Different database block sizes (4kB, 8kB, ..., 32kB) can be used, while
+ * the OS may have different memory page sizes.
+ *
+ * To correctly map between them, we need to: 1. Determine the OS memory
+ * page size 2. Calculate how many OS pages are used by all buffer blocks
+ * 3. Calculate how many OS pages are contained within each database
+ * block.
+ *
+ * This information is needed before calling move_pages() for NUMA memory
+ * node inquiry.
+ */
+ os_page_size = pg_numa_get_pagesize();
+
+ /*
+ * Allocate memory for page pointers and status based on total shared
+ * memory size. This simplified approach allocates enough space for all
+ * pages in shared memory rather than calculating the exact requirements
+ * for each segment.
+ */
+ shm_total_page_count = ShmemSegHdr->totalsize / os_page_size;
+ page_ptrs = palloc0(sizeof(void *) * shm_total_page_count);
+ pages_status = palloc(sizeof(int) * shm_total_page_count);
+
+ if (firstNumaTouch)
+ elog(DEBUG1, "NUMA: page-faulting shared memory segments for proper NUMA readouts");
+
+ LWLockAcquire(ShmemIndexLock, LW_SHARED);
+
+ hash_seq_init(&hstat, ShmemIndex);
+
+ /* output all allocated entries */
+ memset(nulls, 0, sizeof(nulls));
+ while ((ent = (ShmemIndexEnt *) hash_seq_search(&hstat)) != NULL)
+ {
+ int i;
+
+ /* Get number of OS aliged pages */
+ shm_ent_page_count = TYPEALIGN(os_page_size, ent->allocated_size) / os_page_size;
+
+ /*
+ * If we get ever 0xff back from kernel inquiry, then we probably have
+ * bug in our buffers to OS page mapping code here.
+ */
+ memset(pages_status, 0xff, sizeof(int) * shm_ent_page_count);
+
+ for (i = 0; i < shm_ent_page_count; i++)
+ {
+ /*
+ * In order to get reliable results we also need to touch memory
+ * pages, so that inquiry about NUMA memory node doesn't return -2
+ * (which indicates unmapped/unallocated pages).
+ */
+ volatile uint64 touch pg_attribute_unused();
+
+ page_ptrs[i] = (char *) ent->location + (i * os_page_size);
+ if (firstNumaTouch)
+ pg_numa_touch_mem_if_required(touch, page_ptrs[i]);
+
+ CHECK_FOR_INTERRUPTS();
+ }
+
+ if (pg_numa_query_pages(0, shm_ent_page_count, page_ptrs, pages_status) == -1)
+ elog(ERROR, "failed NUMA pages inquiry status: %m");
+
+ memset(nodes, 0, sizeof(Size) * (max_nodes + 1));
+ /* Count number of NUMA nodes used for this shared memory entry */
+ for (i = 0; i < shm_ent_page_count; i++)
+ {
+ int s = pages_status[i];
+
+ /* Ensure we are adding only valid index to the array */
+ if (s >= 0 && s <= max_nodes)
+ nodes[s]++;
+ else
+ elog(ERROR, "invalid NUMA node id outside of allowed range [0, %ld]: %d", max_nodes, s);
+ }
+
+ for (i = 0; i <= max_nodes; i++)
+ {
+ values[0] = CStringGetTextDatum(ent->key);
+ values[1] = i;
+ values[2] = Int64GetDatum(nodes[i] * os_page_size);
+
+ tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc,
+ values, nulls);
+ }
+ }
+
+ /*
+ * We are ignoring the following memory regions (as compared to
+ * pg_get_shmem_allocations()): 1. output shared memory allocated but not
+ * counted via the shmem index 2. output as-of-yet unused shared memory.
+ */
+
+ LWLockRelease(ShmemIndexLock);
+ firstNumaTouch = false;
+
+ return (Datum) 0;
+}
diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat
index 63859661951..0a1fc2a25cd 100644
--- a/src/include/catalog/pg_proc.dat
+++ b/src/include/catalog/pg_proc.dat
@@ -8522,6 +8522,14 @@
proname => 'pg_numa_available', provolatile => 's', prorettype => 'bool',
proargtypes => '', prosrc => 'pg_numa_available' },
+# shared memory usage with NUMA info
+{ oid => '9686', descr => 'NUMA mappings for the main shared memory segment',
+ proname => 'pg_get_shmem_allocations_numa', prorows => '50', proretset => 't',
+ provolatile => 'v', prorettype => 'record', proargtypes => '',
+ proallargtypes => '{text,int4,int8}', proargmodes => '{o,o,o}',
+ proargnames => '{name,node_id,size}',
+ prosrc => 'pg_get_shmem_allocations_numa' },
+
# memory context of local backend
{ oid => '2282',
descr => 'information about all memory contexts of local backend',
diff --git a/src/test/regress/expected/numa.out b/src/test/regress/expected/numa.out
new file mode 100644
index 00000000000..668172f7d79
--- /dev/null
+++ b/src/test/regress/expected/numa.out
@@ -0,0 +1,12 @@
+SELECT NOT(pg_numa_available()) AS skip_test \gset
+\if :skip_test
+\quit
+\endif
+-- switch to superuser
+\c -
+SELECT COUNT(*) >= 0 AS ok FROM pg_shmem_allocations_numa;
+ ok
+----
+ t
+(1 row)
+
diff --git a/src/test/regress/expected/numa_1.out b/src/test/regress/expected/numa_1.out
new file mode 100644
index 00000000000..6dd6824b4e4
--- /dev/null
+++ b/src/test/regress/expected/numa_1.out
@@ -0,0 +1,3 @@
+SELECT NOT(pg_numa_available()) AS skip_test \gset
+\if :skip_test
+\quit
diff --git a/src/test/regress/expected/privileges.out b/src/test/regress/expected/privileges.out
index 5588d83e1bf..f9c3ff259bc 100644
--- a/src/test/regress/expected/privileges.out
+++ b/src/test/regress/expected/privileges.out
@@ -3127,8 +3127,8 @@ REVOKE MAINTAIN ON lock_table FROM regress_locktable_user;
-- clean up
DROP TABLE lock_table;
DROP USER regress_locktable_user;
--- test to check privileges of system views pg_shmem_allocations and
--- pg_backend_memory_contexts.
+-- test to check privileges of system views pg_shmem_allocations,
+-- pg_shmem_allocations_numa and pg_backend_memory_contexts.
-- switch to superuser
\c -
CREATE ROLE regress_readallstats;
@@ -3150,6 +3150,12 @@ SELECT has_table_privilege('regress_readallstats','pg_shmem_allocations','SELECT
f
(1 row)
+SELECT has_table_privilege('regress_readallstats','pg_shmem_allocations_numa','SELECT'); -- no
+ has_table_privilege
+---------------------
+ f
+(1 row)
+
GRANT pg_read_all_stats TO regress_readallstats;
SELECT has_table_privilege('regress_readallstats','pg_aios','SELECT'); -- yes
has_table_privilege
@@ -3169,6 +3175,12 @@ SELECT has_table_privilege('regress_readallstats','pg_shmem_allocations','SELECT
t
(1 row)
+SELECT has_table_privilege('regress_readallstats','pg_shmem_allocations_numa','SELECT'); -- yes
+ has_table_privilege
+---------------------
+ t
+(1 row)
+
-- run query to ensure that functions within views can be executed
SET ROLE regress_readallstats;
SELECT COUNT(*) >= 0 AS ok FROM pg_aios;
diff --git a/src/test/regress/expected/rules.out b/src/test/regress/expected/rules.out
index 673c63b8d1b..abfdc97abc5 100644
--- a/src/test/regress/expected/rules.out
+++ b/src/test/regress/expected/rules.out
@@ -1757,6 +1757,10 @@ pg_shmem_allocations| SELECT name,
size,
allocated_size
FROM pg_get_shmem_allocations() pg_get_shmem_allocations(name, off, size, allocated_size);
+pg_shmem_allocations_numa| SELECT name,
+ node_id,
+ size
+ FROM pg_get_shmem_allocations_numa() pg_get_shmem_allocations_numa(name, node_id, size);
pg_stat_activity| SELECT s.datid,
d.datname,
s.pid,
diff --git a/src/test/regress/parallel_schedule b/src/test/regress/parallel_schedule
index 0a35f2f8f6a..0f38caa0d24 100644
--- a/src/test/regress/parallel_schedule
+++ b/src/test/regress/parallel_schedule
@@ -119,7 +119,7 @@ test: plancache limit plpgsql copy2 temp domain rangefuncs prepare conversion tr
# The stats test resets stats, so nothing else needing stats access can be in
# this group.
# ----------
-test: partition_join partition_prune reloptions hash_part indexing partition_aggregate partition_info tuplesort explain compression memoize stats predicate
+test: partition_join partition_prune reloptions hash_part indexing partition_aggregate partition_info tuplesort explain compression memoize stats predicate numa
# event_trigger depends on create_am and cannot run concurrently with
# any test that runs DDL
diff --git a/src/test/regress/sql/numa.sql b/src/test/regress/sql/numa.sql
new file mode 100644
index 00000000000..034098783fb
--- /dev/null
+++ b/src/test/regress/sql/numa.sql
@@ -0,0 +1,9 @@
+SELECT NOT(pg_numa_available()) AS skip_test \gset
+\if :skip_test
+\quit
+\endif
+
+-- switch to superuser
+\c -
+
+SELECT COUNT(*) >= 0 AS ok FROM pg_shmem_allocations_numa;
diff --git a/src/test/regress/sql/privileges.sql b/src/test/regress/sql/privileges.sql
index 286b1d03756..81b40a1a330 100644
--- a/src/test/regress/sql/privileges.sql
+++ b/src/test/regress/sql/privileges.sql
@@ -1911,8 +1911,8 @@ REVOKE MAINTAIN ON lock_table FROM regress_locktable_user;
DROP TABLE lock_table;
DROP USER regress_locktable_user;
--- test to check privileges of system views pg_shmem_allocations and
--- pg_backend_memory_contexts.
+-- test to check privileges of system views pg_shmem_allocations,
+-- pg_shmem_allocations_numa and pg_backend_memory_contexts.
-- switch to superuser
\c -
@@ -1922,12 +1922,14 @@ CREATE ROLE regress_readallstats;
SELECT has_table_privilege('regress_readallstats','pg_aios','SELECT'); -- no
SELECT has_table_privilege('regress_readallstats','pg_backend_memory_contexts','SELECT'); -- no
SELECT has_table_privilege('regress_readallstats','pg_shmem_allocations','SELECT'); -- no
+SELECT has_table_privilege('regress_readallstats','pg_shmem_allocations_numa','SELECT'); -- no
GRANT pg_read_all_stats TO regress_readallstats;
SELECT has_table_privilege('regress_readallstats','pg_aios','SELECT'); -- yes
SELECT has_table_privilege('regress_readallstats','pg_backend_memory_contexts','SELECT'); -- yes
SELECT has_table_privilege('regress_readallstats','pg_shmem_allocations','SELECT'); -- yes
+SELECT has_table_privilege('regress_readallstats','pg_shmem_allocations_numa','SELECT'); -- yes
-- run query to ensure that functions within views can be executed
SET ROLE regress_readallstats;
--
2.49.0
