agora inbox for [email protected]  
help / color / mirror / Atom feed
[PATCH] Fix compiler builtin usage in new pg_bitutils.c
87+ messages / 4 participants
[nested] [flat]

* [PATCH] Fix compiler builtin usage in new pg_bitutils.c
@ 2019-02-15 14:14  Alvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Alvaro Herrera @ 2019-02-15 14:14 UTC (permalink / raw)

Split out these new functions in three parts: one in a new file that
uses the compiler builtin and gets compiled with the -mpopcnt compiler
option if it exists; another one that uses the compiler builtin but not
the compiler option; and finally the fallback with open-coded
algorithms.

Split out the configure logic: in the original commit, it was selecting
to use the -mpopcnt compiler switch together with deciding to use the
compiler builtin.  However, some compilers implement the builtin even
though they don't have the compiler switch, so split both things.  Also,
expose whether the builtin exists to Makefile.global, so that src/port's
can decide whether to compile the special file.

Remove CPUID test for CTZ/CLZ.
---
 config/c-compiler.m4           |  22 +-
 configure                      | 105 +++++++--
 configure.in                   |   7 +-
 src/Makefile.global.in         |   3 +
 src/include/port/pg_bitutils.h | 176 +++++++++++++++-
 src/port/Makefile              |  16 +-
 src/port/pg_bitutils.c         | 374 ++++-----------------------------
 src/port/pg_bitutils_sse42.c   |  36 ++++
 8 files changed, 365 insertions(+), 374 deletions(-)
 create mode 100644 src/port/pg_bitutils_sse42.c

diff --git a/config/c-compiler.m4 b/config/c-compiler.m4
index 05fa82518f8..7c0d52b515f 100644
--- a/config/c-compiler.m4
+++ b/config/c-compiler.m4
@@ -381,22 +381,16 @@ fi])# PGAC_C_BUILTIN_OP_OVERFLOW
 # PGAC_C_BUILTIN_POPCOUNT
 # -------------------------
 AC_DEFUN([PGAC_C_BUILTIN_POPCOUNT],
-[define([Ac_cachevar], [AS_TR_SH([pgac_cv_popcount])])dnl
-AC_CACHE_CHECK([for __builtin_popcount], [Ac_cachevar],
-[pgac_save_CFLAGS=$CFLAGS
-CFLAGS="$pgac_save_CFLAGS -mpopcnt"
-AC_COMPILE_IFELSE([AC_LANG_SOURCE(
-[static int x = __builtin_popcount(255);])],
-[Ac_cachevar=yes],
-[Ac_cachevar=no])
-CFLAGS="$pgac_save_CFLAGS"])
-if test x"$Ac_cachevar" = x"yes"; then
-  CFLAGS_POPCNT="-mpopcnt"
+[AC_CACHE_CHECK([for __builtin_popcount], pgac_cv__builtin_popcount,
+[AC_COMPILE_IFELSE([AC_LANG_SOURCE(
+[static int x = __builtin_popcount(255);]
+)],
+[pgac_cv__builtin_popcount=yes],
+[pgac_cv__builtin_popcount=no])])
+if test x"$pgac_cv__builtin_popcount" = x"yes"; then
 AC_DEFINE(HAVE__BUILTIN_POPCOUNT, 1,
           [Define to 1 if your compiler understands __builtin_popcount.])
-fi
-undefine([Ac_cachevar])dnl
-])# PGAC_C_BUILTIN_POPCOUNT
+fi])# PGAC_C_BUILTIN_POPCOUNT
 
 
 
diff --git a/configure b/configure
index 73e9c235b69..fa0f1216a0a 100755
--- a/configure
+++ b/configure
@@ -651,7 +651,7 @@ CFLAGS_ARMV8_CRC32C
 CFLAGS_SSE42
 have_win32_dbghelp
 LIBOBJS
-CFLAGS_POPCNT
+have__builtin_popcount
 UUID_LIBS
 LDAP_LIBS_BE
 LDAP_LIBS_FE
@@ -733,6 +733,7 @@ CPP
 BITCODE_CXXFLAGS
 BITCODE_CFLAGS
 CFLAGS_VECTOR
+CFLAGS_POPCNT
 PERMIT_DECLARATION_AFTER_STATEMENT
 LLVM_BINPATH
 LLVM_CXXFLAGS
@@ -6581,6 +6582,87 @@ fi
 
 fi
 
+# Optimization flags and options for bit-twiddling
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether ${CC} supports -mpopcnt, for CFLAGS_POPCNT" >&5
+$as_echo_n "checking whether ${CC} supports -mpopcnt, for CFLAGS_POPCNT... " >&6; }
+if ${pgac_cv_prog_CC_cflags__mpopcnt+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  pgac_save_CFLAGS=$CFLAGS
+pgac_save_CC=$CC
+CC=${CC}
+CFLAGS="${CFLAGS_POPCNT} -mpopcnt"
+ac_save_c_werror_flag=$ac_c_werror_flag
+ac_c_werror_flag=yes
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"; then :
+  pgac_cv_prog_CC_cflags__mpopcnt=yes
+else
+  pgac_cv_prog_CC_cflags__mpopcnt=no
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+ac_c_werror_flag=$ac_save_c_werror_flag
+CFLAGS="$pgac_save_CFLAGS"
+CC="$pgac_save_CC"
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $pgac_cv_prog_CC_cflags__mpopcnt" >&5
+$as_echo "$pgac_cv_prog_CC_cflags__mpopcnt" >&6; }
+if test x"$pgac_cv_prog_CC_cflags__mpopcnt" = x"yes"; then
+  CFLAGS_POPCNT="${CFLAGS_POPCNT} -mpopcnt"
+fi
+
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether ${CC} supports -mpopcntb, for CFLAGS_POPCNT" >&5
+$as_echo_n "checking whether ${CC} supports -mpopcntb, for CFLAGS_POPCNT... " >&6; }
+if ${pgac_cv_prog_CC_cflags__mpopcntb+:} false; then :
+  $as_echo_n "(cached) " >&6
+else
+  pgac_save_CFLAGS=$CFLAGS
+pgac_save_CC=$CC
+CC=${CC}
+CFLAGS="${CFLAGS_POPCNT} -mpopcntb"
+ac_save_c_werror_flag=$ac_c_werror_flag
+ac_c_werror_flag=yes
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+int
+main ()
+{
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_compile "$LINENO"; then :
+  pgac_cv_prog_CC_cflags__mpopcntb=yes
+else
+  pgac_cv_prog_CC_cflags__mpopcntb=no
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
+ac_c_werror_flag=$ac_save_c_werror_flag
+CFLAGS="$pgac_save_CFLAGS"
+CC="$pgac_save_CC"
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $pgac_cv_prog_CC_cflags__mpopcntb" >&5
+$as_echo "$pgac_cv_prog_CC_cflags__mpopcntb" >&6; }
+if test x"$pgac_cv_prog_CC_cflags__mpopcntb" = x"yes"; then
+  CFLAGS_POPCNT="${CFLAGS_POPCNT} -mpopcntb"
+fi
+
+
+
+
 CFLAGS_VECTOR=$CFLAGS_VECTOR
 
 
@@ -14111,32 +14193,28 @@ $as_echo "#define HAVE__BUILTIN_CTZ 1" >>confdefs.h
 fi
 { $as_echo "$as_me:${as_lineno-$LINENO}: checking for __builtin_popcount" >&5
 $as_echo_n "checking for __builtin_popcount... " >&6; }
-if ${pgac_cv_popcount+:} false; then :
+if ${pgac_cv__builtin_popcount+:} false; then :
   $as_echo_n "(cached) " >&6
 else
-  pgac_save_CFLAGS=$CFLAGS
-CFLAGS="$pgac_save_CFLAGS -mpopcnt"
-cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
 /* end confdefs.h.  */
 static int x = __builtin_popcount(255);
+
 _ACEOF
 if ac_fn_c_try_compile "$LINENO"; then :
-  pgac_cv_popcount=yes
+  pgac_cv__builtin_popcount=yes
 else
-  pgac_cv_popcount=no
+  pgac_cv__builtin_popcount=no
 fi
 rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
-CFLAGS="$pgac_save_CFLAGS"
 fi
-{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $pgac_cv_popcount" >&5
-$as_echo "$pgac_cv_popcount" >&6; }
-if test x"$pgac_cv_popcount" = x"yes"; then
-  CFLAGS_POPCNT="-mpopcnt"
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $pgac_cv__builtin_popcount" >&5
+$as_echo "$pgac_cv__builtin_popcount" >&6; }
+if test x"$pgac_cv__builtin_popcount" = x"yes"; then
 
 $as_echo "#define HAVE__BUILTIN_POPCOUNT 1" >>confdefs.h
 
 fi
-
 { $as_echo "$as_me:${as_lineno-$LINENO}: checking for __builtin_unreachable" >&5
 $as_echo_n "checking for __builtin_unreachable... " >&6; }
 if ${pgac_cv__builtin_unreachable+:} false; then :
@@ -14654,6 +14732,7 @@ $as_echo "#define LOCALE_T_IN_XLOCALE 1" >>confdefs.h
 
 fi
 
+have__builtin_popcount=$pgac_cv__builtin_popcount
 
 
 # MSVC doesn't cope well with defining restrict to __restrict, the
diff --git a/configure.in b/configure.in
index 9c4d5f0691e..4a5f1fd62c1 100644
--- a/configure.in
+++ b/configure.in
@@ -547,6 +547,11 @@ elif test "$PORTNAME" = "hpux"; then
   PGAC_PROG_CXX_CFLAGS_OPT([+Olibmerrno])
 fi
 
+# Optimization flags and options for bit-twiddling
+PGAC_PROG_CC_VAR_OPT(CFLAGS_POPCNT, [-mpopcnt])
+PGAC_PROG_CC_VAR_OPT(CFLAGS_POPCNT, [-mpopcntb])
+AC_SUBST(CFLAGS_POPCNT)
+
 AC_SUBST(CFLAGS_VECTOR, $CFLAGS_VECTOR)
 
 # Determine flags used to emit bitcode for JIT inlining. Need to test
@@ -1506,7 +1511,7 @@ AC_TYPE_LONG_LONG_INT
 
 PGAC_TYPE_LOCALE_T
 
-AC_SUBST(CFLAGS_POPCNT)
+AC_SUBST(have__builtin_popcount, $pgac_cv__builtin_popcount)
 
 # MSVC doesn't cope well with defining restrict to __restrict, the
 # spelling it understands, because it conflicts with
diff --git a/src/Makefile.global.in b/src/Makefile.global.in
index aa16da3e0f2..0f4dd195845 100644
--- a/src/Makefile.global.in
+++ b/src/Makefile.global.in
@@ -517,6 +517,9 @@ WIN32_STACK_RLIMIT=4194304
 # Set if we have a working win32 crashdump header
 have_win32_dbghelp = @have_win32_dbghelp@
 
+# Set if __builtin_popcount() is supported by $(CC)
+have__builtin_popcount = @have__builtin_popcount@
+
 # Pull in platform-specific magic
 include $(top_builddir)/src/Makefile.port
 
diff --git a/src/include/port/pg_bitutils.h b/src/include/port/pg_bitutils.h
index 148c5550573..7b76a138f89 100644
--- a/src/include/port/pg_bitutils.h
+++ b/src/include/port/pg_bitutils.h
@@ -10,17 +10,177 @@
  *
  *------------------------------------------------------------------------ -
  */
-
 #ifndef PG_BITUTILS_H
 #define PG_BITUTILS_H
 
-extern int (*pg_popcount32) (uint32 word);
-extern int (*pg_popcount64) (uint64 word);
-extern int (*pg_rightmost_one32) (uint32 word);
-extern int (*pg_rightmost_one64) (uint64 word);
-extern int (*pg_leftmost_one32) (uint32 word);
-extern int (*pg_leftmost_one64) (uint64 word);
-
+extern int	(*pg_popcount32) (uint32 word);
+extern int	(*pg_popcount64) (uint64 word);
 extern uint64 pg_popcount(const char *buf, int bytes);
 
+/* in pg_popcount.c */
+extern int	pg_popcount32_sse42(uint32 word);
+extern int	pg_popcount64_sse42(uint64 word);
+
+
+#ifndef HAVE__BUILTIN_CTZ
+/*
+ * Array marking the position of the right-most set bit for each value of
+ * 1-255.  We count the right-most position as the 0th bit, and the
+ * left-most the 7th bit.  The 0th index of the array must not be used.
+ */
+static const uint8 rightmost_one_pos[256] = {
+	0, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+	4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+	5, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+	4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+	6, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+	4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+	5, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+	4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+	7, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+	4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+	5, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+	4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+	6, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+	4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+	5, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
+	4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0
+};
+#endif							/* !HAVE__BUILTIN_CTZ */
+
+/*
+ * pg_rightmost_one32
+ *		Returns the number of trailing 0-bits in word, starting at the least
+ *		significant bit position. word must not be 0.
+ */
+static inline int
+pg_rightmost_one32(uint32 word)
+{
+	int			result = 0;
+
+	Assert(word != 0);
+
+#ifdef HAVE__BUILTIN_CTZ
+	result = __builtin_ctz(word);
+#else
+	while ((word & 255) == 0)
+	{
+		word >>= 8;
+		result += 8;
+	}
+	result += rightmost_one_pos[word & 255];
+#endif							/* HAVE__BUILTIN_CTZ */
+
+	return result;
+}
+
+/*
+ * pg_rightmost_one64
+ *		Returns the number of trailing 0-bits in word, starting at the least
+ *		significant bit position. word must not be 0.
+ */
+static inline int
+pg_rightmost_one64(uint64 word)
+{
+	int			result = 0;
+
+	Assert(word != 0);
+#ifdef HAVE__BUILTIN_CTZ
+#if defined(HAVE_LONG_INT_64)
+	return __builtin_ctzl(word);
+#elif defined(HAVE_LONG_LONG_INT_64)
+	return __builtin_ctzll(word);
+#else
+#error must have a working 64-bit integer datatype
+#endif
+#else							/* HAVE__BUILTIN_CTZ */
+	while ((word & 255) == 0)
+	{
+		word >>= 8;
+		result += 8;
+	}
+	result += rightmost_one_pos[word & 255];
+#endif
+
+	return result;
+}
+
+#ifndef HAVE__BUILTIN_CLZ
+/*
+ * Array marking the position of the left-most set bit for each value of
+ * 1-255.  We count the right-most position as the 0th bit, and the
+ * left-most the 7th bit.  The 0th index of the array must not be used.
+ */
+static const uint8 leftmost_one_pos[256] = {
+	0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3,
+	4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
+	5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+	5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
+	6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
+	6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
+	6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
+	6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
+	7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+	7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+	7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+	7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+	7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+	7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+	7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+	7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7
+};
+#endif							/* !HAVE_BUILTIN_CLZ */
+
+/*
+ * pg_leftmost_one32
+ *		Returns the 0-based position of the most significant set bit in word
+ *		measured from the least significant bit.  word must not be 0.
+ */
+static inline int
+pg_leftmost_one32(uint32 word)
+{
+#ifdef HAVE__BUILTIN_CLZ
+	Assert(word != 0);
+
+	return 31 - __builtin_clz(word);
+#else
+	int			shift = 32 - 8;
+
+	Assert(word != 0);
+
+	while ((word >> shift) == 0)
+		shift -= 8;
+
+	return shift + leftmost_one_pos[(word >> shift) & 255];
+#endif							/* HAVE__BUILTIN_CLZ */
+}
+
+/*
+ * pg_leftmost_one64
+ *		Returns the 0-based position of the most significant set bit in word
+ *		measured from the least significant bit.  word must not be 0.
+ */
+static inline int
+pg_leftmost_one64(uint64 word)
+{
+#ifdef HAVE__BUILTIN_CLZ
+	Assert(word != 0);
+#if defined(HAVE_LONG_INT_64)
+	return 63 - __builtin_clzl(word);
+#elif defined(HAVE_LONG_LONG_INT_64)
+	return 63 - __builtin_clzll(word);
+#else
+#error must have a working 64-bit integer datatype
+#endif
+#else							/* HAVE__BUILTIN_CLZ */
+	int			shift = 64 - 8;
+
+	Assert(word != 0);
+	while ((word >> shift) == 0)
+		shift -= 8;
+
+	return shift + leftmost_one_pos[(word >> shift) & 255];
+#endif							/* !HAVE__BUIILTIN_CLZ */
+}
+
 #endif							/* PG_BITUTILS_H */
diff --git a/src/port/Makefile b/src/port/Makefile
index 2da73260a13..237cc625e19 100644
--- a/src/port/Makefile
+++ b/src/port/Makefile
@@ -41,6 +41,14 @@ OBJS = $(LIBOBJS) $(PG_CRC32C_OBJS) chklocale.o erand48.o inet_net_ntop.o \
 	qsort.o qsort_arg.o quotes.o snprintf.o sprompt.o strerror.o \
 	tar.o thread.o
 
+# If the compiler supports a special flag for the POPCOUNT instruction and it
+# has __builtin_popcount, add pg_bitutils_sse42.o.
+ifneq ($(CFLAGS_POPCNT),)
+ifeq ($(have__builtin_popcount),yes)
+OBJS += pg_bitutils_sse42.o
+endif
+endif
+
 # libpgport.a, libpgport_shlib.a, and libpgport_srv.a contain the same files
 # foo.o, foo_shlib.o, and foo_srv.o are all built from foo.c
 OBJS_SHLIB = $(OBJS:%.o=%_shlib.o)
@@ -78,10 +86,10 @@ pg_crc32c_armv8.o: CFLAGS+=$(CFLAGS_ARMV8_CRC32C)
 pg_crc32c_armv8_shlib.o: CFLAGS+=$(CFLAGS_ARMV8_CRC32C)
 pg_crc32c_armv8_srv.o: CFLAGS+=$(CFLAGS_ARMV8_CRC32C)
 
-# pg_bitutils.c needs CFLAGS_POPCNT
-pg_bitutils.o: CFLAGS+=$(CFLAGS_POPCNT)
-pg_bitutils_shlib.o: CFLAGS+=$(CFLAGS_POPCNT)
-pg_bitutils_srv.o: CFLAGS+=$(CFLAGS_POPCNT)
+# pg_popcount.c needs CFLAGS_POPCNT
+pg_popcount.o: CFLAGS+=$(CFLAGS_POPCNT)
+pg_popcount_shlib.o: CFLAGS+=$(CFLAGS_POPCNT)
+pg_popcount_srv.o: CFLAGS+=$(CFLAGS_POPCNT)
 
 #
 # Shared library versions of object files
diff --git a/src/port/pg_bitutils.c b/src/port/pg_bitutils.c
index aac394fe927..3d26883111f 100644
--- a/src/port/pg_bitutils.c
+++ b/src/port/pg_bitutils.c
@@ -10,7 +10,6 @@
  *
  *-------------------------------------------------------------------------
  */
-
 #include "postgres.h"
 
 #ifdef HAVE__GET_CPUID
@@ -23,61 +22,21 @@
 
 #include "port/pg_bitutils.h"
 
-#if defined(HAVE__BUILTIN_POPCOUNT) && defined(HAVE__GET_CPUID)
+#ifdef HAVE__BUILTIN_POPCOUNT
 static bool pg_popcount_available(void);
-static int pg_popcount32_choose(uint32 word);
-static int pg_popcount32_sse42(uint32 word);
-static int pg_popcount64_choose(uint64 word);
-static int pg_popcount64_sse42(uint64 word);
-#endif
-static int pg_popcount32_slow(uint32 word);
-static int pg_popcount64_slow(uint64 word);
-
-#if defined(HAVE__GET_CPUID) && (defined(HAVE__BUILTIN_CTZ) || defined(HAVE__BUILTIN_CLZ))
-static bool pg_lzcnt_available(void);
-#endif
-
-#if defined(HAVE__BUILTIN_CTZ) && defined(HAVE__GET_CPUID)
-static int pg_rightmost_one32_choose(uint32 word);
-static int pg_rightmost_one32_abm(uint32 word);
-static int pg_rightmost_one64_choose(uint64 word);
-static int pg_rightmost_one64_abm(uint64 word);
-#endif
-static int pg_rightmost_one32_slow(uint32 word);
-static int pg_rightmost_one64_slow(uint64 word);
-
-#if defined(HAVE__BUILTIN_CLZ) && defined(HAVE__GET_CPUID)
-static int pg_leftmost_one32_choose(uint32 word);
-static int pg_leftmost_one32_abm(uint32 word);
-static int pg_leftmost_one64_choose(uint64 word);
-static int pg_leftmost_one64_abm(uint64 word);
-#endif
-static int pg_leftmost_one32_slow(uint32 word);
-static int pg_leftmost_one64_slow(uint64 word);
-
-#if defined(HAVE__BUILTIN_POPCOUNT) && defined(HAVE__GET_CPUID)
-int (*pg_popcount32) (uint32 word) = pg_popcount32_choose;
-int (*pg_popcount64) (uint64 word) = pg_popcount64_choose;
+static int	pg_popcount32_choose(uint32 word);
+static int	pg_popcount32_builtin(uint32 word);
+static int	pg_popcount64_choose(uint64 word);
+static int	pg_popcount64_builtin(uint64 word);
+int			(*pg_popcount32) (uint32 word) = pg_popcount32_choose;
+int			(*pg_popcount64) (uint64 word) = pg_popcount64_choose;
 #else
-int (*pg_popcount32) (uint32 word) = pg_popcount32_slow;
-int (*pg_popcount64) (uint64 word) = pg_popcount64_slow;
-#endif
+static int	pg_popcount32_slow(uint32 word);
+static int	pg_popcount64_slow(uint64 word);
+int			(*pg_popcount32) (uint32 word) = pg_popcount32_slow;
+int			(*pg_popcount64) (uint64 word) = pg_popcount64_slow;
+#endif							/* !HAVE_BUILTIN_POPCOUNT */
 
-#if defined(HAVE__BUILTIN_CTZ) && defined(HAVE__GET_CPUID)
-int (*pg_rightmost_one32) (uint32 word) = pg_rightmost_one32_choose;
-int (*pg_rightmost_one64) (uint64 word) = pg_rightmost_one64_choose;
-#else
-int (*pg_rightmost_one32) (uint32 word) = pg_rightmost_one32_slow;
-int (*pg_rightmost_one64) (uint64 word) = pg_rightmost_one64_slow;
-#endif
-
-#if defined(HAVE__BUILTIN_CLZ) && defined(HAVE__GET_CPUID)
-int (*pg_leftmost_one32) (uint32 word) = pg_leftmost_one32_choose;
-int (*pg_leftmost_one64) (uint64 word) = pg_leftmost_one64_choose;
-#else
-int (*pg_leftmost_one32) (uint32 word) = pg_leftmost_one32_slow;
-int (*pg_leftmost_one64) (uint64 word) = pg_leftmost_one64_slow;
-#endif
 
 /* Array marking the number of 1-bits for each value of 0-255. */
 static const uint8 number_of_ones[256] = {
@@ -100,77 +59,33 @@ static const uint8 number_of_ones[256] = {
 };
 
 /*
- * Array marking the position of the right-most set bit for each value of
- * 1-255.  We count the right-most position as the 0th bit, and the
- * left-most the 7th bit.  The 0th index of the array must not be used.
+ * Return true iff we have CPUID support and it indicates that the POPCNT
+ * instruction is available.
  */
-static const uint8 rightmost_one_pos[256] = {
-	0, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
-	4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
-	5, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
-	4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
-	6, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
-	4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
-	5, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
-	4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
-	7, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
-	4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
-	5, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
-	4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
-	6, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
-	4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
-	5, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0,
-	4, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, 0, 1, 0
-};
-
-/*
- * Array marking the position of the left-most set bit for each value of
- * 1-255.  We count the right-most position as the 0th bit, and the
- * left-most the 7th bit.  The 0th index of the array must not be used.
- */
-static const uint8 leftmost_one_pos[256] = {
-	0, 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3,
-	4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
-	5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
-	5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
-	6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
-	6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
-	6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
-	6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
-	7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
-	7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
-	7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
-	7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
-	7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
-	7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
-	7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
-	7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7
-};
-
-#if defined(HAVE__GET_CPUID) && defined(HAVE__BUILTIN_POPCOUNT)
-
 static bool
 pg_popcount_available(void)
 {
-	unsigned int exx[4] = { 0, 0, 0, 0 };
+#if defined(HAVE__GET_CPUID) || defined(HAVE__CPUID)
+	unsigned int exx[4] = {0, 0, 0, 0};
 
 #if defined(HAVE__GET_CPUID)
 	__get_cpuid(1, &exx[0], &exx[1], &exx[2], &exx[3]);
 #elif defined(HAVE__CPUID)
 	__cpuid(exx, 1);
-#else
-#error cpuid instruction not available
 #endif
 
 	return (exx[2] & (1 << 23)) != 0;	/* POPCNT */
-}
+#else							/* HAVE__GET_CPUID || HAVE__CPUID */
+
+	return false;
 #endif
+}
 
-#if defined(HAVE__GET_CPUID) && defined(HAVE__BUILTIN_POPCOUNT)
-
+#ifdef HAVE__BUILTIN_POPCOUNT
 /*
- * This gets called on the first call. It replaces the function pointer
- * so that subsequent calls are routed directly to the chosen implementation.
+ * This gets called on the first call to pg_popcount32. It replaces the
+ * function pointer so that subsequent calls are routed directly to the chosen
+ * implementation.
  */
 static int
 pg_popcount32_choose(uint32 word)
@@ -178,18 +93,17 @@ pg_popcount32_choose(uint32 word)
 	if (pg_popcount_available())
 		pg_popcount32 = pg_popcount32_sse42;
 	else
-		pg_popcount32 = pg_popcount32_slow;
+		pg_popcount32 = pg_popcount32_builtin;
 
 	return pg_popcount32(word);
 }
 
 static int
-pg_popcount32_sse42(uint32 word)
+pg_popcount32_builtin(uint32 word)
 {
 	return __builtin_popcount(word);
 }
-#endif
-
+#else							/* HAVE__BUILTIN_POPCOUNT */
 /*
  * pg_popcount32_slow
  *		Return the number of 1 bits set in word
@@ -197,7 +111,7 @@ pg_popcount32_sse42(uint32 word)
 static int
 pg_popcount32_slow(uint32 word)
 {
-	int result = 0;
+	int			result = 0;
 
 	while (word != 0)
 	{
@@ -207,6 +121,7 @@ pg_popcount32_slow(uint32 word)
 
 	return result;
 }
+#endif
 
 /*
  * pg_popcount
@@ -215,13 +130,13 @@ pg_popcount32_slow(uint32 word)
 uint64
 pg_popcount(const char *buf, int bytes)
 {
-	uint64 popcnt = 0;
+	uint64		popcnt = 0;
 
 #if SIZEOF_VOID_P >= 8
 	/* Process in 64-bit chunks if the buffer is aligned. */
 	if (buf == (char *) TYPEALIGN(8, buf))
 	{
-		uint64 *words = (uint64 *) buf;
+		uint64	   *words = (uint64 *) buf;
 
 		while (bytes >= 8)
 		{
@@ -235,7 +150,7 @@ pg_popcount(const char *buf, int bytes)
 	/* Process in 32-bit chunks if the buffer is aligned. */
 	if (buf == (char *) TYPEALIGN(4, buf))
 	{
-		uint32 *words = (uint32 *) buf;
+		uint32	   *words = (uint32 *) buf;
 
 		while (bytes >= 4)
 		{
@@ -254,11 +169,11 @@ pg_popcount(const char *buf, int bytes)
 	return popcnt;
 }
 
-#if defined(HAVE__GET_CPUID) && defined(HAVE__BUILTIN_POPCOUNT)
-
+#ifdef HAVE__BUILTIN_POPCOUNT
 /*
- * This gets called on the first call. It replaces the function pointer
- * so that subsequent calls are routed directly to the chosen implementation.
+ * This gets called on the first call to pg_popcount64. It replaces the
+ * function pointer so that subsequent calls are routed directly to the chosen
+ * implementation.
  */
 static int
 pg_popcount64_choose(uint64 word)
@@ -266,26 +181,24 @@ pg_popcount64_choose(uint64 word)
 	if (pg_popcount_available())
 		pg_popcount64 = pg_popcount64_sse42;
 	else
-		pg_popcount64 = pg_popcount64_slow;
+		pg_popcount64 = pg_popcount64_builtin;
 
 	return pg_popcount64(word);
 }
 
 static int
-pg_popcount64_sse42(uint64 word)
+pg_popcount64_builtin(uint64 word)
 {
 #if defined(HAVE_LONG_INT_64)
 	return __builtin_popcountl(word);
 #elif defined(HAVE_LONG_LONG_INT_64)
 	return __builtin_popcountll(word);
 #else
-	/* shouldn't happen */
 #error must have a working 64-bit integer datatype
 #endif
 }
 
-#endif
-
+#else							/* HAVE__BUILTIN_POPCOUNT */
 /*
  * pg_popcount64_slow
  *		Return the number of 1 bits set in word
@@ -293,7 +206,7 @@ pg_popcount64_sse42(uint64 word)
 static int
 pg_popcount64_slow(uint64 word)
 {
-	int result = 0;
+	int			result = 0;
 
 	while (word != 0)
 	{
@@ -303,211 +216,4 @@ pg_popcount64_slow(uint64 word)
 
 	return result;
 }
-
-#if defined(HAVE__GET_CPUID) && (defined(HAVE__BUILTIN_CTZ) || defined(HAVE__BUILTIN_CLZ))
-
-static bool
-pg_lzcnt_available(void)
-{
-
-	unsigned int exx[4] = { 0, 0, 0, 0 };
-
-#if defined(HAVE__GET_CPUID)
-	__get_cpuid(0x80000001, &exx[0], &exx[1], &exx[2], &exx[3]);
-#elif defined(HAVE__CPUID)
-	__cpuid(exx, 0x80000001);
-#else
-#error cpuid instruction not available
 #endif
-
-	return (exx[2] & (1 << 5)) != 0;	/* LZCNT */
-}
-#endif
-
-#if defined(HAVE__GET_CPUID) && defined(HAVE__BUILTIN_CTZ)
-/*
- * This gets called on the first call. It replaces the function pointer
- * so that subsequent calls are routed directly to the chosen implementation.
- */
-static int
-pg_rightmost_one32_choose(uint32 word)
-{
-	if (pg_lzcnt_available())
-		pg_rightmost_one32 = pg_rightmost_one32_abm;
-	else
-		pg_rightmost_one32 = pg_rightmost_one32_slow;
-
-	return pg_rightmost_one32(word);
-}
-
-static int
-pg_rightmost_one32_abm(uint32 word)
-{
-	return __builtin_ctz(word);
-}
-
-#endif
-
-/*
- * pg_rightmost_one32_slow
- *		Returns the number of trailing 0-bits in word, starting at the least
- *		significant bit position. word must not be 0.
- */
-static int
-pg_rightmost_one32_slow(uint32 word)
-{
-	int result = 0;
-
-	Assert(word != 0);
-
-	while ((word & 255) == 0)
-	{
-		word >>= 8;
-		result += 8;
-	}
-	result += rightmost_one_pos[word & 255];
-
-	return result;
-}
-
-#if defined(HAVE__GET_CPUID) && defined(HAVE__BUILTIN_CTZ)
-/*
- * This gets called on the first call. It replaces the function pointer
- * so that subsequent calls are routed directly to the chosen implementation.
- */
-static int
-pg_rightmost_one64_choose(uint64 word)
-{
-	if (pg_lzcnt_available())
-		pg_rightmost_one64 = pg_rightmost_one64_abm;
-	else
-		pg_rightmost_one64 = pg_rightmost_one64_slow;
-
-	return pg_rightmost_one64(word);
-}
-
-static int
-pg_rightmost_one64_abm(uint64 word)
-{
-#if defined(HAVE_LONG_INT_64)
-	return __builtin_ctzl(word);
-#elif defined(HAVE_LONG_LONG_INT_64)
-	return __builtin_ctzll(word);
-#else
-	/* shouldn't happen */
-#error must have a working 64-bit integer datatype
-#endif
-}
-#endif
-
-/*
- * pg_rightmost_one64_slow
- *		Returns the number of trailing 0-bits in word, starting at the least
- *		significant bit position. word must not be 0.
- */
-static int
-pg_rightmost_one64_slow(uint64 word)
-{
-	int result = 0;
-
-	Assert(word != 0);
-
-	while ((word & 255) == 0)
-	{
-		word >>= 8;
-		result += 8;
-	}
-	result += rightmost_one_pos[word & 255];
-
-	return result;
-}
-
-#if defined(HAVE__GET_CPUID) && defined(HAVE__BUILTIN_CLZ)
-/*
- * This gets called on the first call. It replaces the function pointer
- * so that subsequent calls are routed directly to the chosen implementation.
- */
-static int
-pg_leftmost_one32_choose(uint32 word)
-{
-	if (pg_lzcnt_available())
-		pg_leftmost_one32 = pg_leftmost_one32_abm;
-	else
-		pg_leftmost_one32 = pg_leftmost_one32_slow;
-
-	return pg_leftmost_one32(word);
-}
-
-static int
-pg_leftmost_one32_abm(uint32 word)
-{
-	return 31 - __builtin_clz(word);
-}
-#endif
-
-/*
- * pg_leftmost_one32_slow
- *		Returns the 0-based position of the most significant set bit in word
- *		measured from the least significant bit.  word must not be 0.
- */
-static int
-pg_leftmost_one32_slow(uint32 word)
-{
-	int			shift = 32 - 8;
-
-	Assert(word != 0);
-
-	while ((word >> shift) == 0)
-		shift -= 8;
-
-	return shift + leftmost_one_pos[(word >> shift) & 255];
-}
-
-#if defined(HAVE__GET_CPUID) && defined(HAVE__BUILTIN_CLZ)
-/*
- * This gets called on the first call. It replaces the function pointer
- * so that subsequent calls are routed directly to the chosen implementation.
- */
-static int
-pg_leftmost_one64_choose(uint64 word)
-{
-	if (pg_lzcnt_available())
-		pg_leftmost_one64 = pg_leftmost_one64_abm;
-	else
-		pg_leftmost_one64 = pg_leftmost_one64_slow;
-
-	return pg_leftmost_one64(word);
-}
-
-static int
-pg_leftmost_one64_abm(uint64 word)
-{
-#if defined(HAVE_LONG_INT_64)
-	return 63 - __builtin_clzl(word);
-#elif defined(HAVE_LONG_LONG_INT_64)
-	return 63 - __builtin_clzll(word);
-#else
-	/* shouldn't happen */
-#error must have a working 64-bit integer datatype
-#endif
-
-}
-#endif
-
-/*
- * pg_leftmost_one64_slow
- *		Returns the 0-based position of the most significant set bit in word
- *		measured from the least significant bit.  word must not be 0.
- */
-static int
-pg_leftmost_one64_slow(uint64 word)
-{
-	int			shift = 64 - 8;
-
-	Assert(word != 0);
-
-	while ((word >> shift) == 0)
-		shift -= 8;
-
-	return shift + leftmost_one_pos[(word >> shift) & 255];
-}
diff --git a/src/port/pg_bitutils_sse42.c b/src/port/pg_bitutils_sse42.c
new file mode 100644
index 00000000000..9945e5c103d
--- /dev/null
+++ b/src/port/pg_bitutils_sse42.c
@@ -0,0 +1,36 @@
+/*-------------------------------------------------------------------------
+ *
+ * pg_popcount.c
+ *	  CPU-optimized implementation of pg_popcount
+ *
+ * This file must be compiled with a compiler-specific flag to enable the
+ * POPCNT instruction.
+ *
+ * Portions Copyright (c) 2019, PostgreSQL Global Development Group
+ *
+ * IDENTIFICATION
+ *	  src/port/pg_popcount.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#include "port/pg_bitutils.h"
+
+int
+pg_popcount32_sse42(uint32 word)
+{
+	return __builtin_popcount(word);
+}
+
+int
+pg_popcount64_sse42(uint64 word)
+{
+#if defined(HAVE_LONG_INT_64)
+	return __builtin_popcountl(word);
+#elif defined(HAVE_LONG_LONG_INT_64)
+	return __builtin_popcountll(word);
+#else
+#error must have a working 64-bit integer datatype
+#endif
+}
-- 
2.17.1


--r5Pyd7+fXNt84Ff3--




^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v1 2/2] WIP: Fix EEOP_JSON_CONSTRUCTOR and EEOP_JSONEXPR size.
@ 2022-06-17 01:33  Andres Freund <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Andres Freund @ 2022-06-17 01:33 UTC (permalink / raw)

Author:
Reviewed-By:
Discussion: https://postgr.es/m/
Backpatch:
---
 src/include/executor/execExpr.h | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/src/include/executor/execExpr.h b/src/include/executor/execExpr.h
index 2fba1450c65..5bfa7e9bdd4 100644
--- a/src/include/executor/execExpr.h
+++ b/src/include/executor/execExpr.h
@@ -22,6 +22,8 @@ struct ExprEvalStep;
 struct SubscriptingRefState;
 struct ScalarArrayOpExprHashTable;
 struct JsonbValue;
+struct JsonExprState;
+struct JsonConstructorExprState;
 
 /* Bits in ExprState->flags (see also execnodes.h for public flag bits): */
 /* expression's interpreter has been initialized */
-- 
2.35.1.677.gabf474a5dd


--rftf6si7y32exfgc--





^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v4 01/12] Change section heading to better reflect saving a result in variable(s)
@ 2023-09-24 20:49  Karl O. Pinc <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Karl O. Pinc @ 2023-09-24 20:49 UTC (permalink / raw)

The current section title of "Executing a Command with a Single-Row
Result" does not reflect what the section is really about.  Other
sections make clear how to _execute_ commands, single-row result or not.
What this section is about is how to _save_ a single row of results into
variable(s).

It would be nice to talk about saving results into variables in the
section heading but I couldn't come up with anything pithy.  "Saving a
Single-Row of a Command's Result" seems good enough, especially since
there's few other places to save results other than in variables.
---
 doc/src/sgml/plpgsql.sgml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/doc/src/sgml/plpgsql.sgml b/doc/src/sgml/plpgsql.sgml
index f55e901c7e..8747e84245 100644
--- a/doc/src/sgml/plpgsql.sgml
+++ b/doc/src/sgml/plpgsql.sgml
@@ -1126,7 +1126,7 @@ PERFORM create_mv('cs_session_page_requests_mv', my_query);
    </sect2>
 
    <sect2 id="plpgsql-statements-sql-onerow">
-    <title>Executing a Command with a Single-Row Result</title>
+    <title>Saving a Single-Row of a Command's Result</title>
 
     <indexterm zone="plpgsql-statements-sql-onerow">
      <primary>SELECT INTO</primary>
-- 
2.30.2


--MP_/OOXZvOwbpccKfGOtE9/SwX6
Content-Type: text/x-patch
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
 filename=v4-0002-Change-section-heading-to-better-describe-referen.patch



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread

* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific
@ 2026-04-06 22:07  Álvaro Herrera <[email protected]>
  0 siblings, 0 replies; 87+ messages in thread

From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw)

By default, the logical decoding assumes access to shared catalogs, so
the snapshot builder needs to consider cluster-wide XIDs during startup.
That in turn means that, if any transaction is already running (and has
XID assigned), the snapshot builder needs to wait for its completion, as
it does not know if that transaction performed catalog changes earlier.

A possible problem with this concept is that if REPACK (CONCURRENTLY) is
running in some database, backends running the same command in other
databases get stuck until the first one has committed. Thus only a
single backend in the cluster can run REPACK (CONCURRENTLY) at any time.
Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf
of subscriptions throughout the cluster.

This patch adds a new option to logical replication output plugin, to
declare that it does not use shared catalogs (i.e. catalogs that can be
changed by transactions running in other databases in the cluster). In
that case, no snapshot the backend will use during the decoding needs to
contain information about transactions running in other databases. Thus
the snapshot builder only needs to wait for completion of transactions
in the current database.

Currently we only use this option in the REPACK background worker. It
could possibly be used in the plugin for logical replication too,
however that would need thorough analysis of that plugin.

Bump WAL version number, due to a new field in xl_running_xacts.

Author: Antonin Houska <[email protected]>
Reviewed-by: Álvaro Herrera <[email protected]>
Discussion: https://postgr.es/m/90475.1775218118@localhost
---
 contrib/pg_visibility/pg_visibility.c       |  4 +-
 doc/src/sgml/logicaldecoding.sgml           |  4 ++
 src/backend/access/index/genam.c            |  8 +++
 src/backend/access/rmgrdesc/standbydesc.c   |  2 +
 src/backend/access/transam/xlog.c           |  2 +-
 src/backend/access/transam/xlogfuncs.c      |  2 +-
 src/backend/postmaster/bgwriter.c           |  2 +-
 src/backend/replication/logical/decode.c    | 17 +++++-
 src/backend/replication/logical/logical.c   |  3 +
 src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++-
 src/backend/replication/pgrepack/pgrepack.c |  7 +++
 src/backend/replication/slot.c              |  2 +-
 src/backend/storage/ipc/procarray.c         | 23 +++++++-
 src/backend/storage/ipc/standby.c           | 24 +++++++-
 src/include/access/genam.h                  |  1 +
 src/include/access/xlog_internal.h          |  2 +-
 src/include/miscadmin.h                     |  9 +++
 src/include/replication/output_plugin.h     |  1 +
 src/include/replication/snapbuild.h         |  3 +-
 src/include/storage/procarray.h             |  2 +-
 src/include/storage/standby.h               |  3 +-
 src/include/storage/standbydefs.h           |  1 +
 22 files changed, 167 insertions(+), 18 deletions(-)

diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c
index dfab0b64cf5..d564bd2a00c 100644
--- a/contrib/pg_visibility/pg_visibility.c
+++ b/contrib/pg_visibility/pg_visibility.c
@@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 	else if (rel == NULL || rel->rd_rel->relisshared)
 	{
 		/* Shared relation: take into account all running xids */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestRunningXid;
@@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel)
 		 * Normal relation: take into account xids running within the current
 		 * database
 		 */
-		runningTransactions = GetRunningTransactionData();
+		runningTransactions = GetRunningTransactionData(InvalidOid);
 		LWLockRelease(ProcArrayLock);
 		LWLockRelease(XidGenLock);
 		return runningTransactions->oldestDatabaseRunningXid;
diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml
index 6dc49108997..9b1d68d0de6 100644
--- a/doc/src/sgml/logicaldecoding.sgml
+++ b/doc/src/sgml/logicaldecoding.sgml
@@ -959,6 +959,7 @@ typedef struct OutputPluginOptions
 {
     OutputPluginOutputType output_type;
     bool        receive_rewrites;
+    bool        need_shared_catalogs;
 } OutputPluginOptions;
 </programlisting>
       <literal>output_type</literal> has to either be set to
@@ -969,6 +970,9 @@ typedef struct OutputPluginOptions
       also be called for changes made by heap rewrites during certain DDL
       operations.  These are of interest to plugins that handle DDL
       replication, but they require special handling.
+      <literal>need_shared_catalogs</literal> can be set to false if you are
+      certain the plugin functions do not access shared system catalogs.
+      Doing so can speed up creation of replication slots that use this plugin.
      </para>
 
      <para>
diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c
index 1408989c568..97d44b84622 100644
--- a/src/backend/access/index/genam.c
+++ b/src/backend/access/index/genam.c
@@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation,
 	SysScanDesc sysscan;
 	Relation	irel;
 
+	/*
+	 * If this backend promised that it won't access shared catalogs during
+	 * logical decoding, this it the right place to verify.
+	 */
+	Assert(!HistoricSnapshotActive() ||
+		   accessSharedCatalogsInDecoding ||
+		   !heapRelation->rd_rel->relisshared);
+
 	if (indexOK &&
 		!IgnoreSystemIndexes &&
 		!ReindexIsProcessingIndex(indexId))
diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c
index 0a291354ae2..685d1bdb024 100644
--- a/src/backend/access/rmgrdesc/standbydesc.c
+++ b/src/backend/access/rmgrdesc/standbydesc.c
@@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec)
 		for (i = 0; i < xlrec->subxcnt; i++)
 			appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]);
 	}
+
+	appendStringInfo(buf, "; dbid: %u", xlrec->dbid);
 }
 
 void
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b82af9a85c0..3f08a832ca6 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags)
 	 * recovery we don't need to write running xact data.
 	 */
 	if (!shutdown && XLogStandbyInfoActive())
-		LogStandbySnapshot();
+		LogStandbySnapshot(InvalidOid);
 
 	START_CRIT_SECTION();
 
diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c
index 65bbaeda59c..0f5979691e6 100644
--- a/src/backend/access/transam/xlogfuncs.c
+++ b/src/backend/access/transam/xlogfuncs.c
@@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 				 errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\"")));
 
-	recptr = LogStandbySnapshot();
+	recptr = LogStandbySnapshot(InvalidOid);
 
 	/*
 	 * As a convenience, return the WAL location of the last inserted record
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index 1d8947774a9..a30de4262eb 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len)
 			if (now >= timeout &&
 				last_snapshot_lsn <= GetLastImportantRecPtr())
 			{
-				last_snapshot_lsn = LogStandbySnapshot();
+				last_snapshot_lsn = LogStandbySnapshot(InvalidOid);
 				last_snapshot_ts = now;
 			}
 		}
diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c
index c9fea8cad28..38c5a4f5540 100644
--- a/src/backend/replication/logical/decode.c
+++ b/src/backend/replication/logical/decode.c
@@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 			{
 				xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r);
 
-				SnapBuildProcessRunningXacts(builder, buf->origptr, running);
+				/*
+				 * Update this decoder's idea of transactions currently
+				 * running.  In doing so we will determine whether we have
+				 * reached consistent status.
+				 *
+				 * If the output plugin doesn't need access to shared
+				 * catalogs, we can ignore transactions in other databases.
+				 */
+				SnapBuildProcessRunningXacts(builder, buf->origptr, running,
+											 !ctx->options.need_shared_catalogs);
 
 				/*
 				 * Abort all transactions that we keep track of, that are
@@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
 				 * all running transactions which includes prepared ones,
 				 * while shutdown checkpoints just know that no non-prepared
 				 * transactions are in progress.
+				 *
+				 * The database-specific records might work here too, but it's
+				 * not their purpose.
 				 */
-				ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
+				if (!OidIsValid(running->dbid))
+					ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid);
 			}
 			break;
 		case XLOG_STANDBY_LOCK:
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index f20d0c542f3..8ceaf64d164 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options,
 	ctx->write = do_write;
 	ctx->update_progress = update_progress;
 
+	/* Assume shared catalog access. The startup callback can change it. */
+	ctx->options.need_shared_catalogs = true;
+
 	ctx->output_plugin_options = output_plugin_options;
 
 	ctx->fast_forward = fast_forward;
diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c
index b4269a3b102..ffb4ab2cf2a 100644
--- a/src/backend/replication/logical/snapbuild.c
+++ b/src/backend/replication/logical/snapbuild.c
@@ -154,6 +154,14 @@
 static ResourceOwner SavedResourceOwnerDuringExport = NULL;
 static bool ExportInProgress = false;
 
+/*
+ * If a backend is going to do logical decoding and the output plugin does
+ * not need to access shared catalogs, setting this variable to false can make
+ * the decoding startup faster. In particular, the backend will not need to
+ * wait for completion of already running transactions in other databases.
+ */
+bool		accessSharedCatalogsInDecoding = true;
+
 /* ->committed and ->catchange manipulation */
 static void SnapBuildPurgeOlderTxn(SnapBuild *builder);
 
@@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction
 												 uint32 xinfo);
 
 /* xlog reading helper functions for SnapBuildProcessRunningXacts */
-static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running);
+static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn,
+								  xl_running_xacts *running);
 static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff);
 
 /* serialization functions */
@@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder,
 
 	MemoryContextSwitchTo(oldcontext);
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	return builder;
 }
 
@@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder)
 		builder->snapshot = NULL;
 	}
 
+	/* The default is that shared catalog are used. */
+	accessSharedCatalogsInDecoding = true;
+
 	/* other resources are deallocated via memory context reset */
 	MemoryContextDelete(context);
 }
@@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid,
  * anymore.
  */
 void
-SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running)
+SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running,
+							 bool db_specific)
 {
 	ReorderBufferTXN *txn;
 	TransactionId xmin;
@@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	 */
 	if (builder->state < SNAPBUILD_CONSISTENT)
 	{
+		/*
+		 * To reduce the potential for unnecessarily waiting for completion of
+		 * unrelated transactions, the caller can declare that only
+		 * transactions of the current database are relevant at this stage.
+		 */
+		if (db_specific)
+		{
+			/*
+			 * If we must only keep track of transactions running in the
+			 * current database, we need transaction info from exactly that
+			 * database.
+			 */
+			if (running->dbid != MyDatabaseId)
+			{
+				LogStandbySnapshot(MyDatabaseId);
+
+				return;
+			}
+
+			/*
+			 * We'd better be able to check during scan if the plugin does not
+			 * lie.
+			 */
+			if (accessSharedCatalogsInDecoding)
+				accessSharedCatalogsInDecoding = false;
+		}
+
 		/* returns false if there's no point in performing cleanup just yet */
 		if (!SnapBuildFindSnapshot(builder, lsn, running))
 			return;
@@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact
 	else
 		SnapBuildSerialize(builder, lsn);
 
+	/*
+	 * Database specific transaction info may exist to reach CONSISTENT state
+	 * faster, however the code below makes no use of it. Moreover, such
+	 * record might cause problems because the following normal (cluster-wide)
+	 * record can have lower value of oldestRunningXid. In that case, let's
+	 * wait with the cleanup for the next regular cluster-wide record.
+	 */
+	if (OidIsValid(running->dbid))
+		return;
+
 	/*
 	 * Update range of interesting xids based on the running xacts
 	 * information. We don't increase ->xmax using it, because once we are in
@@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff)
 	 */
 	if (!RecoveryInProgress())
 	{
-		LogStandbySnapshot();
+		/*
+		 * If the last transaction info was about specific database, so needs
+		 * to be the next one - at least until we're in the CONSISTENT state.
+		 */
+		LogStandbySnapshot(running->dbid);
 	}
 }
 
diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c
index 457b6518a8e..4da66aa0f44 100644
--- a/src/backend/replication/pgrepack/pgrepack.c
+++ b/src/backend/replication/pgrepack/pgrepack.c
@@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt,
 	/* Probably unnecessary, as we don't use the SQL interface ... */
 	opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT;
 
+	/*
+	 * REPACK doesn't need access to shared catalogs, so we can speed up the
+	 * historic snapshot creation by setting this flag.  We'll only have to
+	 * wait for transactions in our database.
+	 */
+	opt->need_shared_catalogs = false;
+
 	if (ctx->output_plugin_options != NIL)
 	{
 		ereport(ERROR,
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 21a213a0ebf..a1f37e59dbc 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void)
 		XLogRecPtr	flushptr;
 
 		/* make sure we have enough information to start */
-		flushptr = LogStandbySnapshot();
+		flushptr = LogStandbySnapshot(InvalidOid);
 
 		/* and make sure it's fsynced to disk */
 		XLogFlush(flushptr);
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index f540bb6b23f..9299bcebbda 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc)
  *
  * Note that if any transaction has overflowed its cached subtransactions
  * then there is no real need include any subtransactions.
+ *
+ * If 'dbid' is valid, only gather transactions running in that database.
  */
 RunningTransactions
-GetRunningTransactionData(void)
+GetRunningTransactionData(Oid dbid)
 {
 	/* result workspace */
 	static RunningTransactionsData CurrentRunningXactsData;
@@ -2700,6 +2702,18 @@ GetRunningTransactionData(void)
 		if (!TransactionIdIsValid(xid))
 			continue;
 
+		/*
+		 * Filter by database OID if requested.
+		 */
+		if (OidIsValid(dbid))
+		{
+			int			pgprocno = arrayP->pgprocnos[index];
+			PGPROC	   *proc = &allProcs[pgprocno];
+
+			if (proc->databaseId != dbid)
+				continue;
+		}
+
 		/*
 		 * Be careful not to exclude any xids before calculating the values of
 		 * oldestRunningXid and suboverflowed, since these are used to clean
@@ -2750,6 +2764,12 @@ GetRunningTransactionData(void)
 			PGPROC	   *proc = &allProcs[pgprocno];
 			int			nsubxids;
 
+			/*
+			 * Filter by database OID if requested.
+			 */
+			if (OidIsValid(dbid) && proc->databaseId != dbid)
+				continue;
+
 			/*
 			 * Save subtransaction XIDs. Other backends can't add or remove
 			 * entries while we're holding XidGenLock.
@@ -2783,6 +2803,7 @@ GetRunningTransactionData(void)
 	 * increases if slots do.
 	 */
 
+	CurrentRunningXacts->dbid = dbid;
 	CurrentRunningXacts->xcnt = count - subcount;
 	CurrentRunningXacts->subxcnt = subcount;
 	CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY;
diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c
index de9092fdf5b..29af7733948 100644
--- a/src/backend/storage/ipc/standby.c
+++ b/src/backend/storage/ipc/standby.c
@@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record)
 		xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record);
 		RunningTransactionsData running;
 
+		/*
+		 * Records issued for specific database are not suitable for physical
+		 * replication because that affects the whole cluster. In particular,
+		 * the list of XID is probably incomplete here.
+		 */
+		if (OidIsValid(xlrec->dbid))
+			return;
+
 		running.xcnt = xlrec->xcnt;
 		running.subxcnt = xlrec->subxcnt;
 		running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY;
@@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record)
  * as there's no independent knob to just enable logical decoding. For
  * details of how this is used, check snapbuild.c's introductory comment.
  *
+ * If 'dbid' is valid, only gather transactions running in that
+ * database. snapbuild.c can use such running xacts information for faster
+ * startup, but it still needs normal (cluster-wide) during the actual
+ * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for
+ * details. Other processes (e.g. checkpointer) issue the cluster-wide records
+ * whether logical decoding is active or not.
+ *
+ * Please be careful about using this argument for other purposes. In
+ * particular, physical replication *must* ignore the database-specific
+ * records, exactly because they do not cover the whole cluster - see
+ * standby_redo().
  *
  * Returns the RecPtr of the last inserted record.
  */
 XLogRecPtr
-LogStandbySnapshot(void)
+LogStandbySnapshot(Oid dbid)
 {
 	XLogRecPtr	recptr;
 	RunningTransactions running;
@@ -1314,7 +1333,7 @@ LogStandbySnapshot(void)
 	 * Log details of all in-progress transactions. This should be the last
 	 * record we write, because standby will open up when it sees this.
 	 */
-	running = GetRunningTransactionData();
+	running = GetRunningTransactionData(dbid);
 
 	/*
 	 * GetRunningTransactionData() acquired ProcArrayLock, we must release it.
@@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts)
 	xl_running_xacts xlrec;
 	XLogRecPtr	recptr;
 
+	xlrec.dbid = CurrRunningXacts->dbid;
 	xlrec.xcnt = CurrRunningXacts->xcnt;
 	xlrec.subxcnt = CurrRunningXacts->subxcnt;
 	xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY);
diff --git a/src/include/access/genam.h b/src/include/access/genam.h
index b69320a7fc8..de55821e414 100644
--- a/src/include/access/genam.h
+++ b/src/include/access/genam.h
@@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance
 	bool		isnull;
 } IndexOrderByDistance;
 
+
 /*
  * generalized index_ interface routines (in indexam.c)
  */
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 10c18d39ff8..13ae3ad4fbb 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -32,7 +32,7 @@
 /*
  * Each page of XLOG file has a header like this:
  */
-#define XLOG_PAGE_MAGIC 0xD11E	/* can be used as WAL version indicator */
+#define XLOG_PAGE_MAGIC 0xD11F	/* can be used as WAL version indicator */
 
 typedef struct XLogPageHeaderData
 {
diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h
index 2e10e3c814d..93b7816c09c 100644
--- a/src/include/miscadmin.h
+++ b/src/include/miscadmin.h
@@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname);
 extern void PreventCommandIfParallelMode(const char *cmdname);
 extern void PreventCommandDuringRecovery(const char *cmdname);
 
+/* in replication/snapbuild.c */
+
+/*
+ * Keep track of whether logical decoding in this backend promised not to
+ * access shared catalogs, as a safety check.  This is checked by genam.c when
+ * a catalog scan takes place to verify that no shared catalogs are accessed.
+ */
+extern bool accessSharedCatalogsInDecoding;
+
 /*****************************************************************************
  *	  pdir.h --																 *
  *			POSTGRES directory path definitions.                             *
diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h
index 842fcde67f9..917f3cff232 100644
--- a/src/include/replication/output_plugin.h
+++ b/src/include/replication/output_plugin.h
@@ -27,6 +27,7 @@ typedef struct OutputPluginOptions
 {
 	OutputPluginOutputType output_type;
 	bool		receive_rewrites;
+	bool		need_shared_catalogs;
 } OutputPluginOptions;
 
 /*
diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h
index a22a83a2f23..d02530a912a 100644
--- a/src/include/replication/snapbuild.h
+++ b/src/include/replication/snapbuild.h
@@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid,
 								   XLogRecPtr lsn,
 								   xl_heap_new_cid *xlrec);
 extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn,
-										 xl_running_xacts *running);
+										 xl_running_xacts *running,
+										 bool db_specific);
 extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn);
 
 extern bool SnapBuildSnapshotExists(XLogRecPtr lsn);
diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h
index d718a5b542f..ec89c448220 100644
--- a/src/include/storage/procarray.h
+++ b/src/include/storage/procarray.h
@@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin,
 										 VirtualTransactionId *sourcevxid);
 extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc);
 
-extern RunningTransactions GetRunningTransactionData(void);
+extern RunningTransactions GetRunningTransactionData(Oid dbid);
 
 extern bool TransactionIdIsInProgress(TransactionId xid);
 extern TransactionId GetOldestNonRemovableTransactionId(Relation rel);
diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h
index 6a314c693cd..8715c08e94f 100644
--- a/src/include/storage/standby.h
+++ b/src/include/storage/standby.h
@@ -126,6 +126,7 @@ typedef enum
 
 typedef struct RunningTransactionsData
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	subxids_array_status subxid_status;
@@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions;
 extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid);
 extern void LogAccessExclusiveLockPrepare(void);
 
-extern XLogRecPtr LogStandbySnapshot(void);
+extern XLogRecPtr LogStandbySnapshot(Oid dbid);
 extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs,
 									bool relcacheInitFileInval);
 
diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h
index 231d251fd51..e75b7078766 100644
--- a/src/include/storage/standbydefs.h
+++ b/src/include/storage/standbydefs.h
@@ -46,6 +46,7 @@ typedef struct xl_standby_locks
  */
 typedef struct xl_running_xacts
 {
+	Oid			dbid;			/* only track xacts in this database */
 	int			xcnt;			/* # of xact ids in xids[] */
 	int			subxcnt;		/* # of subxact ids in xids[] */
 	bool		subxid_overflow;	/* snapshot overflowed, subxids missing */
-- 
2.47.3


--qs77q6fpwolne4ds
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch"



^ permalink  raw  reply  [nested|flat] 87+ messages in thread


end of thread, other threads:[~2026-04-06 22:07 UTC | newest]

Thread overview: 87+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2019-02-15 14:14 [PATCH] Fix compiler builtin usage in new pg_bitutils.c Alvaro Herrera <[email protected]>
2022-06-17 01:33 [PATCH v1 2/2] WIP: Fix EEOP_JSON_CONSTRUCTOR and EEOP_JSONEXPR size. Andres Freund <[email protected]>
2023-09-24 20:49 [PATCH v4 01/12] Change section heading to better reflect saving a result in variable(s) Karl O. Pinc <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>
2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]>

This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox