Hi! Thank you for your attention to this patch!

On 16.03.2026 11:34, Андрей Зубков wrote:
Really it was "revocations", but I'm agree with Andrey that naming isn't clear. *_vm_cleared looks better, but talking about naming here "vm" meaning is not clear. I think it will be understood as visibility map, but it is "mark" really. Maybe "*_pages_marks_cleared" will be better?

Also a macro in pgstat.h:733 and pgstat.h:738 still holds "_rev_".
Good catch, fixed.

I think the docs description needs a little correction:

- visible_pages_vm_cleared. I think listing of possible DML operations
  is not needed here, also it seems a high rate of this counter has no
  direct relation to the index only scans because we can have very
  agressive vacuum on a table that will do the opposite. It will hold
  few pages without visibility marks constantly but with the cost
  of high visible_pages_vm_cleared rate. My proposition follows:

  Number of times the all-visible bit in the
   <link linkend="storage-vm">visibility map</link> was cleared for a
  pages of this table. The all-visible bit of a heap page is cleared
  every time backend process modifies a page previously marked
  all-visible by vacuum. Vacuum process must process page once again
  on the next run. A high rate of change of this counter means that
  vacuum should re-do its work on this table.

- frozen_pages_vm_cleared:

  Number of times the all-frozen bit in the
   <link linkend="storage-vm">visibility map</link> was cleared for a
  pages of this table.  The all-frozen bit of a heap page is cleared
  every time backend process modifies a page previously marked
  all-frozen by vacuum. Vacuum process must process page once again on
  the next freeze run on this table.
I agree, this description is clearer. Fixed.

-----------
Best regards,
Alena Rybakina
From bd7cbd4450512aaf640156e977faa28e5095d33a Mon Sep 17 00:00:00 2001
From: Alena Rybakina <[email protected]>
Date: Mon, 16 Mar 2026 14:55:45 +0300
Subject: [PATCH] Track table VM stability.

Add rev_all_visible_pages and rev_all_frozen_pages counters to
pg_stat_all_tables tracking the number of times the all-visible and
all-frozen bits are cleared in the visibility map. These bits are cleared by
backend processes during regular DML operations. Hence, the counters are placed
in table statistic entry.

A high rev_all_visible_pages rate relative to DML volume indicates
that modifications are scattered across previously-clean pages rather
than concentrated on already-dirty ones, causing index-only scans to
fall back to heap fetches.  A high rev_all_frozen_pages rate indicates
that vacuum's freezing work is being frequently undone by concurrent
DML.

Authors: Alena Rybakina <[email protected]>,
         Andrei Lepikhov <[email protected]>,
         Andrei Zubkov <[email protected]>
Reviewed-by: Dilip Kumar <[email protected]>,
         Masahiko Sawada <[email protected]>,
         Ilia Evdokimov <[email protected]>,
         Jian He <[email protected]>,
         Kirill Reshke <[email protected]>,
         Alexander Korotkov <[email protected]>,
         Jim Nasby <[email protected]>,
         Sami Imseih <[email protected]>,
         Karina Litskevich <[email protected]>
---
 doc/src/sgml/monitoring.sgml                  | 32 ++++++++
 src/backend/access/heap/visibilitymap.c       | 10 +++
 src/backend/catalog/system_views.sql          |  4 +-
 src/backend/utils/activity/pgstat_relation.c  |  2 +
 src/backend/utils/adt/pgstatfuncs.c           |  6 ++
 src/include/catalog/pg_proc.dat               | 10 +++
 src/include/pgstat.h                          | 17 ++++-
 .../expected/vacuum-extending-freeze.out      | 50 +++++++++++++
 src/test/isolation/isolation_schedule         |  1 +
 .../specs/vacuum-extending-freeze.spec        | 73 +++++++++++++++++++
 src/test/regress/expected/rules.out           | 12 ++-
 11 files changed, 212 insertions(+), 5 deletions(-)
 create mode 100644 src/test/isolation/expected/vacuum-extending-freeze.out
 create mode 100644 src/test/isolation/specs/vacuum-extending-freeze.spec

diff --git a/doc/src/sgml/monitoring.sgml b/doc/src/sgml/monitoring.sgml
index 9c5c6dc490f..0b27558686e 100644
--- a/doc/src/sgml/monitoring.sgml
+++ b/doc/src/sgml/monitoring.sgml
@@ -4258,6 +4258,38 @@ description | Waiting for a newly initialized WAL file 
to reach durable storage
       </para></entry>
      </row>
 
+     <row>
+      <entry role="catalog_table_entry"><para role="column_definition">
+       <structfield>visible_pages_cleared</structfield> <type>bigint</type>
+      </para>
+      <para>
+       Number of times the all-visible bit in the
+       <link linkend="storage-vm">visibility map</link> was cleared for
+       pages of this table.  The all-visible bit of a heap page is
+       cleared whenever a backend process modifies a page that was
+       previously marked all-visible by <command>VACUUM</command>.  The
+       page must then be processed again by <command>VACUUM</command> on a
+       subsequent run.  A high rate of change in this counter means that
+       <command>VACUUM</command> has to repeatedly re-process pages of this
+       table.
+      </para></entry>
+     </row>
+
+     <row>
+      <entry role="catalog_table_entry"><para role="column_definition">
+       <structfield>frozen_pages_cleared</structfield> <type>bigint</type>
+      </para>
+      <para>
+       Number of times the all-frozen bit in the
+       <link linkend="storage-vm">visibility map</link> was cleared for
+       pages of this table.  The all-frozen bit of a heap page is cleared
+       whenever a backend process modifies a page that was previously
+       marked all-frozen by <command>VACUUM</command>.  The page must then
+       be processed again by <command>VACUUM</command> on the next freeze
+       run for this table.
+      </para></entry>
+     </row>
+
      <row>
       <entry role="catalog_table_entry"><para role="column_definition">
        <structfield>last_vacuum</structfield> <type>timestamp with time 
zone</type>
diff --git a/src/backend/access/heap/visibilitymap.c 
b/src/backend/access/heap/visibilitymap.c
index e21b96281a6..7b3ab6244d0 100644
--- a/src/backend/access/heap/visibilitymap.c
+++ b/src/backend/access/heap/visibilitymap.c
@@ -92,6 +92,7 @@
 #include "access/xloginsert.h"
 #include "access/xlogutils.h"
 #include "miscadmin.h"
+#include "pgstat.h"
 #include "port/pg_bitutils.h"
 #include "storage/bufmgr.h"
 #include "storage/smgr.h"
@@ -163,6 +164,15 @@ visibilitymap_clear(Relation rel, BlockNumber heapBlk, 
Buffer vmbuf, uint8 flags
 
        if (map[mapByte] & mask)
        {
+               /*
+                * Track how often all-visible or all-frozen bits are cleared 
in the
+                * visibility map.
+                */
+               if (map[mapByte] & ((flags & VISIBILITYMAP_ALL_VISIBLE) << 
mapOffset))
+                       pgstat_count_visible_pages_cleared(rel);
+               if (map[mapByte] & ((flags & VISIBILITYMAP_ALL_FROZEN) << 
mapOffset))
+                       pgstat_count_frozen_pages_cleared(rel);
+
                map[mapByte] &= ~mask;
 
                MarkBufferDirty(vmbuf);
diff --git a/src/backend/catalog/system_views.sql 
b/src/backend/catalog/system_views.sql
index 90d48bc9c80..9ff013ac797 100644
--- a/src/backend/catalog/system_views.sql
+++ b/src/backend/catalog/system_views.sql
@@ -741,7 +741,9 @@ CREATE VIEW pg_stat_all_tables AS
             pg_stat_get_total_autovacuum_time(C.oid) AS total_autovacuum_time,
             pg_stat_get_total_analyze_time(C.oid) AS total_analyze_time,
             pg_stat_get_total_autoanalyze_time(C.oid) AS 
total_autoanalyze_time,
-            pg_stat_get_stat_reset_time(C.oid) AS stats_reset
+            pg_stat_get_stat_reset_time(C.oid) AS stats_reset,
+            pg_stat_get_visible_pages_cleared(C.oid) AS visible_pages_cleared,
+            pg_stat_get_frozen_pages_cleared(C.oid) AS frozen_pages_cleared
     FROM pg_class C LEFT JOIN
          pg_index I ON C.oid = I.indrelid
          LEFT JOIN pg_namespace N ON (N.oid = C.relnamespace)
diff --git a/src/backend/utils/activity/pgstat_relation.c 
b/src/backend/utils/activity/pgstat_relation.c
index bc8c43b96aa..78936aca82e 100644
--- a/src/backend/utils/activity/pgstat_relation.c
+++ b/src/backend/utils/activity/pgstat_relation.c
@@ -879,6 +879,8 @@ pgstat_relation_flush_cb(PgStat_EntryRef *entry_ref, bool 
nowait)
 
        tabentry->blocks_fetched += lstats->counts.blocks_fetched;
        tabentry->blocks_hit += lstats->counts.blocks_hit;
+       tabentry->visible_pages_cleared += lstats->counts.visible_pages_cleared;
+       tabentry->frozen_pages_cleared += lstats->counts.frozen_pages_cleared;
 
        /* Clamp live_tuples in case of negative delta_live_tuples */
        tabentry->live_tuples = Max(tabentry->live_tuples, 0);
diff --git a/src/backend/utils/adt/pgstatfuncs.c 
b/src/backend/utils/adt/pgstatfuncs.c
index 5ac022274a7..d50b7233c0e 100644
--- a/src/backend/utils/adt/pgstatfuncs.c
+++ b/src/backend/utils/adt/pgstatfuncs.c
@@ -107,6 +107,12 @@ PG_STAT_GET_RELENTRY_INT64(tuples_updated)
 /* pg_stat_get_vacuum_count */
 PG_STAT_GET_RELENTRY_INT64(vacuum_count)
 
+/* pg_stat_get_visible_pages_cleared */
+PG_STAT_GET_RELENTRY_INT64(visible_pages_cleared)
+
+/* pg_stat_get_frozen_pages_cleared */
+PG_STAT_GET_RELENTRY_INT64(frozen_pages_cleared)
+
 #define PG_STAT_GET_RELENTRY_FLOAT8(stat)                                      
        \
 Datum                                                                          
                                        \
 CppConcat(pg_stat_get_,stat)(PG_FUNCTION_ARGS)                                 
\
diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat
index 361e2cfffeb..b52e463e63f 100644
--- a/src/include/catalog/pg_proc.dat
+++ b/src/include/catalog/pg_proc.dat
@@ -12833,4 +12833,14 @@
   proname => 'hashoid8extended', prorettype => 'int8',
   proargtypes => 'oid8 int8', prosrc => 'hashoid8extended' },
 
+{ oid => '8002',
+  descr => 'statistics: number of times the all-visible pages in the 
visibility map were cleared for pages of this table',
+  proname => 'pg_stat_get_visible_pages_cleared', provolatile => 's',
+  proparallel => 'r', prorettype => 'int8', proargtypes => 'oid',
+  prosrc => 'pg_stat_get_visible_pages_cleared' },
+{ oid => '8003',
+  descr => 'statistics: number of times the all-frozen pages in the visibility 
map were cleared for pages of this table',
+  proname => 'pg_stat_get_frozen_pages_cleared', provolatile => 's',
+  proparallel => 'r', prorettype => 'int8', proargtypes => 'oid',
+  prosrc => 'pg_stat_get_frozen_pages_cleared' },
 ]
diff --git a/src/include/pgstat.h b/src/include/pgstat.h
index 216b93492ba..8116d0959de 100644
--- a/src/include/pgstat.h
+++ b/src/include/pgstat.h
@@ -159,6 +159,8 @@ typedef struct PgStat_TableCounts
 
        PgStat_Counter blocks_fetched;
        PgStat_Counter blocks_hit;
+       PgStat_Counter visible_pages_cleared;
+       PgStat_Counter frozen_pages_cleared;
 } PgStat_TableCounts;
 
 /* ----------
@@ -217,7 +219,7 @@ typedef struct PgStat_TableXactStatus
  * ------------------------------------------------------------
  */
 
-#define PGSTAT_FILE_FORMAT_ID  0x01A5BCBB
+#define PGSTAT_FILE_FORMAT_ID  0x01A5BCBC
 
 typedef struct PgStat_ArchiverStats
 {
@@ -450,6 +452,8 @@ typedef struct PgStat_StatTabEntry
 
        PgStat_Counter blocks_fetched;
        PgStat_Counter blocks_hit;
+       PgStat_Counter visible_pages_cleared;
+       PgStat_Counter frozen_pages_cleared;
 
        TimestampTz last_vacuum_time;   /* user initiated vacuum */
        PgStat_Counter vacuum_count;
@@ -725,6 +729,17 @@ extern void pgstat_report_analyze(Relation rel,
                if (pgstat_should_count_relation(rel))                          
                \
                        (rel)->pgstat_info->counts.blocks_hit++;                
                \
        } while (0)
+/* count revocations of all-visible and all-frozen bits in visibility map */
+#define pgstat_count_visible_pages_cleared(rel)                                
                \
+       do {                                                                    
                                                \
+               if (pgstat_should_count_relation(rel))                          
                \
+                       (rel)->pgstat_info->counts.visible_pages_cleared++;     
\
+       } while (0)
+#define pgstat_count_frozen_pages_cleared(rel)                                 
        \
+       do {                                                                    
                                                \
+               if (pgstat_should_count_relation(rel))                          
                \
+                       (rel)->pgstat_info->counts.frozen_pages_cleared++;      
\
+       } while (0)
 
 extern void pgstat_count_heap_insert(Relation rel, PgStat_Counter n);
 extern void pgstat_count_heap_update(Relation rel, bool hot, bool newpage);
diff --git a/src/test/isolation/expected/vacuum-extending-freeze.out 
b/src/test/isolation/expected/vacuum-extending-freeze.out
new file mode 100644
index 00000000000..58b51570e5e
--- /dev/null
+++ b/src/test/isolation/expected/vacuum-extending-freeze.out
@@ -0,0 +1,50 @@
+Parsed test spec with 2 sessions
+
+starting permutation: s1_initial_vacuum s2_vacuum s1_get_set_vm_flags_stats 
s1_update_table s1_get_cleared_vm_flags_stats
+pg_stat_force_next_flush
+------------------------
+                        
+(1 row)
+
+step s1_initial_vacuum: 
+    SELECT pg_stat_force_next_flush();
+
+pg_stat_force_next_flush
+------------------------
+                        
+(1 row)
+
+step s2_vacuum: 
+    VACUUM vestat;
+
+step s1_get_set_vm_flags_stats: 
+    SELECT relallfrozen > 0 AS relallfrozen_pos,
+           relallvisible > 0 AS relallvisible_pos
+    FROM pg_class c
+    WHERE c.relname = 'vestat';
+
+relallfrozen_pos|relallvisible_pos
+----------------+-----------------
+t               |t                
+(1 row)
+
+step s1_update_table: 
+    UPDATE vestat SET x = x + 1001;
+    SELECT pg_stat_force_next_flush();
+
+pg_stat_force_next_flush
+------------------------
+                        
+(1 row)
+
+step s1_get_cleared_vm_flags_stats: 
+    SELECT visible_pages_cleared > 0 AS visible_pages_cleared,
+           frozen_pages_cleared > 0 AS frozen_pages_cleared
+    FROM pg_stat_all_tables
+    WHERE relname = 'vestat';
+
+visible_pages_cleared|frozen_pages_cleared
+---------------------+--------------------
+t                    |t                   
+(1 row)
+
diff --git a/src/test/isolation/isolation_schedule 
b/src/test/isolation/isolation_schedule
index 4e466580cd4..81e68f85d88 100644
--- a/src/test/isolation/isolation_schedule
+++ b/src/test/isolation/isolation_schedule
@@ -124,3 +124,4 @@ test: serializable-parallel-2
 test: serializable-parallel-3
 test: matview-write-skew
 test: lock-nowait
+test: vacuum-extending-freeze
diff --git a/src/test/isolation/specs/vacuum-extending-freeze.spec 
b/src/test/isolation/specs/vacuum-extending-freeze.spec
new file mode 100644
index 00000000000..b8f8c177595
--- /dev/null
+++ b/src/test/isolation/specs/vacuum-extending-freeze.spec
@@ -0,0 +1,73 @@
+# In short, this test validates the correctness and stability of cumulative
+# vacuum statistics accounting around freezing, visibility, and revision
+# tracking across VACUUM and backend operations.
+
+setup
+{
+    CREATE TABLE vestat (x int)
+        WITH (autovacuum_enabled = off, fillfactor = 70);
+
+    INSERT INTO vestat
+        SELECT i FROM generate_series(1, 5000) AS g(i);
+
+    ANALYZE vestat;
+
+    -- Ensure stats are flushed before starting the scenario
+    SELECT pg_stat_force_next_flush();
+}
+
+teardown
+{
+    DROP TABLE IF EXISTS vestat;
+    RESET vacuum_freeze_min_age;
+    RESET vacuum_freeze_table_age;
+
+}
+
+session s1
+
+step s1_initial_vacuum
+{
+    SELECT pg_stat_force_next_flush();
+}
+
+step s1_get_set_vm_flags_stats
+{
+    SELECT relallfrozen > 0 AS relallfrozen_pos,
+           relallvisible > 0 AS relallvisible_pos
+    FROM pg_class c
+    WHERE c.relname = 'vestat';
+}
+
+step s1_get_cleared_vm_flags_stats
+{
+    SELECT visible_pages_cleared > 0 AS visible_pages_cleared,
+           frozen_pages_cleared > 0 AS frozen_pages_cleared
+    FROM pg_stat_all_tables
+    WHERE relname = 'vestat';
+}
+
+session s2
+setup
+{
+    -- Configure aggressive freezing vacuum behavior
+    SET vacuum_freeze_min_age = 0;
+    SET vacuum_freeze_table_age = 0;
+}
+step s2_vacuum
+{
+    VACUUM vestat;
+}
+
+step s1_update_table
+{
+    UPDATE vestat SET x = x + 1001;
+    SELECT pg_stat_force_next_flush();
+}
+
+permutation
+    s1_initial_vacuum
+    s2_vacuum
+    s1_get_set_vm_flags_stats
+    s1_update_table
+    s1_get_cleared_vm_flags_stats
diff --git a/src/test/regress/expected/rules.out 
b/src/test/regress/expected/rules.out
index 71d7262049e..b36b551d877 100644
--- a/src/test/regress/expected/rules.out
+++ b/src/test/regress/expected/rules.out
@@ -1846,7 +1846,9 @@ pg_stat_all_tables| SELECT c.oid AS relid,
     pg_stat_get_total_autovacuum_time(c.oid) AS total_autovacuum_time,
     pg_stat_get_total_analyze_time(c.oid) AS total_analyze_time,
     pg_stat_get_total_autoanalyze_time(c.oid) AS total_autoanalyze_time,
-    pg_stat_get_stat_reset_time(c.oid) AS stats_reset
+    pg_stat_get_stat_reset_time(c.oid) AS stats_reset,
+    pg_stat_get_visible_pages_cleared(c.oid) AS visible_pages_cleared,
+    pg_stat_get_frozen_pages_cleared(c.oid) AS frozen_pages_cleared
    FROM ((pg_class c
      LEFT JOIN pg_index i ON ((c.oid = i.indrelid)))
      LEFT JOIN pg_namespace n ON ((n.oid = c.relnamespace)))
@@ -2298,7 +2300,9 @@ pg_stat_sys_tables| SELECT relid,
     total_autovacuum_time,
     total_analyze_time,
     total_autoanalyze_time,
-    stats_reset
+    stats_reset,
+    visible_pages_cleared,
+    frozen_pages_cleared
    FROM pg_stat_all_tables
   WHERE ((schemaname = ANY (ARRAY['pg_catalog'::name, 
'information_schema'::name])) OR (schemaname ~ '^pg_toast'::text));
 pg_stat_user_functions| SELECT p.oid AS funcid,
@@ -2353,7 +2357,9 @@ pg_stat_user_tables| SELECT relid,
     total_autovacuum_time,
     total_analyze_time,
     total_autoanalyze_time,
-    stats_reset
+    stats_reset,
+    visible_pages_cleared,
+    frozen_pages_cleared
    FROM pg_stat_all_tables
   WHERE ((schemaname <> ALL (ARRAY['pg_catalog'::name, 
'information_schema'::name])) AND (schemaname !~ '^pg_toast'::text));
 pg_stat_wal| SELECT wal_records,
-- 
2.39.5 (Apple Git-154)

Reply via email to