v1-0003-matview-use-buffered-insert-API-for-CMV-and-RMV.patch

application/octet-stream

Filename: v1-0003-matview-use-buffered-insert-API-for-CMV-and-RMV.patch
Type: application/octet-stream
Part: 3
Message: Re: Introduce new multi insert Table AM and improve performance of various SQL commands with it for Heap AM

Patch

Same data as JSON: GET /api/v1/attachments/:id/patch the parsed metadata as JSON — format, series position, per-file stats; never the diff bytes. API reference →
Format: format-patch
Series: patch v1-0003
Subject: matview: use buffered-insert API for CMV and RMV
File+
src/backend/commands/matview.c 80 19
src/test/regress/expected/matview.out 152 0
src/test/regress/sql/matview.sql 59 0
From 0b810edcdc5e80ab4ec7d77a8d6687fdedd73b32 Mon Sep 17 00:00:00 2001
From: Haibo Yan <haibo.yan@apple.com>
Date: Thu, 23 Apr 2026 12:43:31 -0700
Subject: [PATCH v1 3/5] matview: use buffered-insert API for CMV and RMV

Adopt the buffered-insert lifecycle API in the transientrel datafill
path used by CREATE MATERIALIZED VIEW and REFRESH MATERIALIZED VIEW.

The path uses table_buffered_insert_begin()/put()/end() with a NULL
flush callback and falls back to the existing single-row path when
needed.

Retain the volatile-function check conservatively for the initial patch
series. Concurrent refresh is unchanged.

Add focused CMV/RMV regression tests.
---
 src/backend/commands/matview.c        |  99 +++++++++++++----
 src/test/regress/expected/matview.out | 152 ++++++++++++++++++++++++++
 src/test/regress/sql/matview.sql      |  59 ++++++++++
 3 files changed, 291 insertions(+), 19 deletions(-)

diff --git a/src/backend/commands/matview.c b/src/backend/commands/matview.c
index f7d8007f796..7de4acbdfb1 100644
--- a/src/backend/commands/matview.c
+++ b/src/backend/commands/matview.c
@@ -31,6 +31,7 @@
 #include "executor/executor.h"
 #include "executor/spi.h"
 #include "miscadmin.h"
+#include "optimizer/optimizer.h"
 #include "pgstat.h"
 #include "rewrite/rewriteHandler.h"
 #include "storage/lmgr.h"
@@ -50,7 +51,9 @@ typedef struct
 	Relation	transientrel;	/* relation to write to */
 	CommandId	output_cid;		/* cmin to insert in output tuples */
 	uint32		ti_options;		/* table_tuple_insert performance options */
-	BulkInsertState bistate;	/* bulk insert state */
+	BulkInsertState bistate;	/* bulk insert state (fallback path only) */
+	TableBufferedInsertState buffered_state;	/* buffered-insert state, or NULL */
+	bool		use_buffered_insert;	/* true if buffered path is eligible */
 } DR_transientrel;
 
 static int	matview_maintenance_depth = 0;
@@ -427,6 +430,24 @@ refresh_matview_datafill(DestReceiver *dest, Query *query,
 	/* Plan the query which will generate data for the refresh. */
 	plan = pg_plan_query(query, queryString, CURSOR_OPT_PARALLEL_OK, NULL, NULL);
 
+	/*
+	 * Conservative implementation choice: disable the buffered-insert path
+	 * if the planned target list contains volatile functions.
+	 *
+	 * The buffered-insert API contract does not require this check — the
+	 * matview's defining query was parsed at creation time and cannot
+	 * reference the transient target table.  This guard is retained for the
+	 * initial patch series as a caller-local conservatism and can be relaxed
+	 * after validation without any API change.
+	 */
+	{
+		DR_transientrel *myState = (DR_transientrel *) dest;
+
+		myState->use_buffered_insert =
+			!contain_volatile_functions_after_planning(
+				(Expr *) plan->planTree->targetlist);
+	}
+
 	/*
 	 * Use a snapshot with an updated command ID to ensure this query sees
 	 * results of any previously executed queries.  (This could only matter if
@@ -492,7 +513,40 @@ transientrel_startup(DestReceiver *self, int operation, TupleDesc typeinfo)
 	myState->transientrel = transientrel;
 	myState->output_cid = GetCurrentCommandId(true);
 	myState->ti_options = TABLE_INSERT_SKIP_FSM | TABLE_INSERT_FROZEN;
-	myState->bistate = GetBulkInsertState();
+
+	if (myState->use_buffered_insert)
+	{
+		/*
+		 * Try the buffered-insert path.  Pass NULL flush callback — the
+		 * transient table has no indexes, triggers, or per-tuple post-insert
+		 * work during the datafill phase.
+		 */
+		myState->ti_options |= TABLE_INSERT_BAS_BULKWRITE;
+		myState->buffered_state =
+			table_buffered_insert_begin(transientrel,
+										myState->output_cid,
+										myState->ti_options,
+										NULL, NULL);
+
+		if (myState->buffered_state != NULL)
+		{
+			myState->bistate = NULL;
+		}
+		else
+		{
+			/* AM does not support buffered inserts; fall back */
+			myState->bistate = GetBulkInsertState();
+		}
+	}
+	else
+	{
+		/*
+		 * Buffered insertion not selected for this datafill.  Currently this
+		 * is reached when the conservative volatile-function guard fires.
+		 */
+		myState->buffered_state = NULL;
+		myState->bistate = GetBulkInsertState();
+	}
 
 	/*
 	 * Valid smgr_targblock implies something already wrote to the relation.
@@ -509,20 +563,19 @@ transientrel_receive(TupleTableSlot *slot, DestReceiver *self)
 {
 	DR_transientrel *myState = (DR_transientrel *) self;
 
-	/*
-	 * Note that the input slot might not be of the type of the target
-	 * relation. That's supported by table_tuple_insert(), but slightly less
-	 * efficient than inserting with the right slot - but the alternative
-	 * would be to copy into a slot of the right type, which would not be
-	 * cheap either. This also doesn't allow accessing per-AM data (say a
-	 * tuple's xmin), but since we don't do that here...
-	 */
-
-	table_tuple_insert(myState->transientrel,
-					   slot,
-					   myState->output_cid,
-					   myState->ti_options,
-					   myState->bistate);
+	/* Both paths accept the caller-provided slot directly. */
+	if (myState->buffered_state != NULL)
+	{
+		table_buffered_insert_put(myState->buffered_state, slot);
+	}
+	else
+	{
+		table_tuple_insert(myState->transientrel,
+						   slot,
+						   myState->output_cid,
+						   myState->ti_options,
+						   myState->bistate);
+	}
 
 	/* We know this is a newly created relation, so there are no indexes */
 
@@ -537,9 +590,17 @@ transientrel_shutdown(DestReceiver *self)
 {
 	DR_transientrel *myState = (DR_transientrel *) self;
 
-	FreeBulkInsertState(myState->bistate);
-
-	table_finish_bulk_insert(myState->transientrel, myState->ti_options);
+	if (myState->buffered_state != NULL)
+	{
+		/* end() flushes remaining tuples and subsumes finish_bulk_insert */
+		table_buffered_insert_end(myState->buffered_state);
+		myState->buffered_state = NULL;
+	}
+	else
+	{
+		FreeBulkInsertState(myState->bistate);
+		table_finish_bulk_insert(myState->transientrel, myState->ti_options);
+	}
 
 	/* close transientrel, but keep lock until commit */
 	table_close(myState->transientrel, NoLock);
diff --git a/src/test/regress/expected/matview.out b/src/test/regress/expected/matview.out
index 0355720dfc6..330ea4989f2 100644
--- a/src/test/regress/expected/matview.out
+++ b/src/test/regress/expected/matview.out
@@ -699,3 +699,155 @@ NOTICE:  relation "matview_ine_tab" already exists, skipping
 (0 rows)
 
 DROP MATERIALIZED VIEW matview_ine_tab;
+--
+-- Tests for CMV/RMV with buffered-insert path.
+--
+-- These tests verify correctness under both the buffered path (no volatile
+-- functions) and the fallback single-row path (volatile functions present).
+-- Path selection is not directly observable in SQL output.
+--
+-- CMV: basic correctness
+CREATE MATERIALIZED VIEW mv_buffered_1 AS
+  SELECT g AS a, g * 10 AS b FROM generate_series(1, 5) g;
+SELECT count(*) FROM mv_buffered_1;
+ count 
+-------
+     5
+(1 row)
+
+SELECT * FROM mv_buffered_1 ORDER BY a;
+ a | b  
+---+----
+ 1 | 10
+ 2 | 20
+ 3 | 30
+ 4 | 40
+ 5 | 50
+(5 rows)
+
+-- RMV: refresh repopulates correctly
+REFRESH MATERIALIZED VIEW mv_buffered_1;
+SELECT count(*) FROM mv_buffered_1;
+ count 
+-------
+     5
+(1 row)
+
+SELECT * FROM mv_buffered_1 ORDER BY a;
+ a | b  
+---+----
+ 1 | 10
+ 2 | 20
+ 3 | 30
+ 4 | 40
+ 5 | 50
+(5 rows)
+
+DROP MATERIALIZED VIEW mv_buffered_1;
+-- CMV + RMV: bulk case to exercise auto-flush (>1000 rows)
+CREATE MATERIALIZED VIEW mv_buffered_bulk AS
+  SELECT g AS a FROM generate_series(1, 2500) g;
+SELECT count(*) FROM mv_buffered_bulk;
+ count 
+-------
+  2500
+(1 row)
+
+SELECT min(a), max(a) FROM mv_buffered_bulk;
+ min | max  
+-----+------
+   1 | 2500
+(1 row)
+
+REFRESH MATERIALIZED VIEW mv_buffered_bulk;
+SELECT count(*) FROM mv_buffered_bulk;
+ count 
+-------
+  2500
+(1 row)
+
+SELECT min(a), max(a) FROM mv_buffered_bulk;
+ min | max  
+-----+------
+   1 | 2500
+(1 row)
+
+DROP MATERIALIZED VIEW mv_buffered_bulk;
+-- Wide tuples: verify no regression
+CREATE MATERIALIZED VIEW mv_buffered_wide AS
+  SELECT g AS id,
+         repeat('x', 200) AS col1,
+         repeat('y', 200) AS col2,
+         repeat('z', 200) AS col3
+  FROM generate_series(1, 100) g;
+SELECT count(*) FROM mv_buffered_wide;
+ count 
+-------
+   100
+(1 row)
+
+SELECT id, length(col1), length(col2), length(col3)
+  FROM mv_buffered_wide WHERE id IN (1, 50, 100) ORDER BY id;
+ id  | length | length | length 
+-----+--------+--------+--------
+   1 |    200 |    200 |    200
+  50 |    200 |    200 |    200
+ 100 |    200 |    200 |    200
+(3 rows)
+
+DROP MATERIALIZED VIEW mv_buffered_wide;
+-- Volatile-function fallback: random() triggers conservative guard.
+-- Result must still be correct through the single-row path.
+CREATE MATERIALIZED VIEW mv_volatile AS
+  SELECT g AS a, random() AS r FROM generate_series(1, 10) g;
+SELECT count(*) FROM mv_volatile;
+ count 
+-------
+    10
+(1 row)
+
+SELECT a FROM mv_volatile ORDER BY a;
+ a  
+----
+  1
+  2
+  3
+  4
+  5
+  6
+  7
+  8
+  9
+ 10
+(10 rows)
+
+REFRESH MATERIALIZED VIEW mv_volatile;
+SELECT count(*) FROM mv_volatile;
+ count 
+-------
+    10
+(1 row)
+
+SELECT a FROM mv_volatile ORDER BY a;
+ a  
+----
+  1
+  2
+  3
+  4
+  5
+  6
+  7
+  8
+  9
+ 10
+(10 rows)
+
+DROP MATERIALIZED VIEW mv_volatile;
+-- WITH NO DATA: unchanged behavior
+CREATE MATERIALIZED VIEW mv_nodata AS SELECT 1 AS a WITH NO DATA;
+SELECT count(*) FROM mv_nodata;  -- error: not populated
+ERROR:  materialized view "mv_nodata" has not been populated
+HINT:  Use the REFRESH MATERIALIZED VIEW command.
+REFRESH MATERIALIZED VIEW mv_nodata WITH NO DATA;
+DROP MATERIALIZED VIEW mv_nodata;
diff --git a/src/test/regress/sql/matview.sql b/src/test/regress/sql/matview.sql
index 934426b9ae8..43ceee50e90 100644
--- a/src/test/regress/sql/matview.sql
+++ b/src/test/regress/sql/matview.sql
@@ -318,3 +318,62 @@ EXPLAIN (ANALYZE, COSTS OFF, SUMMARY OFF, TIMING OFF, BUFFERS OFF)
   CREATE MATERIALIZED VIEW IF NOT EXISTS matview_ine_tab AS
     SELECT 1 / 0 WITH NO DATA; -- ok
 DROP MATERIALIZED VIEW matview_ine_tab;
+
+--
+-- Tests for CMV/RMV with buffered-insert path.
+--
+-- These tests verify correctness under both the buffered path (no volatile
+-- functions) and the fallback single-row path (volatile functions present).
+-- Path selection is not directly observable in SQL output.
+--
+
+-- CMV: basic correctness
+CREATE MATERIALIZED VIEW mv_buffered_1 AS
+  SELECT g AS a, g * 10 AS b FROM generate_series(1, 5) g;
+SELECT count(*) FROM mv_buffered_1;
+SELECT * FROM mv_buffered_1 ORDER BY a;
+
+-- RMV: refresh repopulates correctly
+REFRESH MATERIALIZED VIEW mv_buffered_1;
+SELECT count(*) FROM mv_buffered_1;
+SELECT * FROM mv_buffered_1 ORDER BY a;
+DROP MATERIALIZED VIEW mv_buffered_1;
+
+-- CMV + RMV: bulk case to exercise auto-flush (>1000 rows)
+CREATE MATERIALIZED VIEW mv_buffered_bulk AS
+  SELECT g AS a FROM generate_series(1, 2500) g;
+SELECT count(*) FROM mv_buffered_bulk;
+SELECT min(a), max(a) FROM mv_buffered_bulk;
+REFRESH MATERIALIZED VIEW mv_buffered_bulk;
+SELECT count(*) FROM mv_buffered_bulk;
+SELECT min(a), max(a) FROM mv_buffered_bulk;
+DROP MATERIALIZED VIEW mv_buffered_bulk;
+
+-- Wide tuples: verify no regression
+CREATE MATERIALIZED VIEW mv_buffered_wide AS
+  SELECT g AS id,
+         repeat('x', 200) AS col1,
+         repeat('y', 200) AS col2,
+         repeat('z', 200) AS col3
+  FROM generate_series(1, 100) g;
+SELECT count(*) FROM mv_buffered_wide;
+SELECT id, length(col1), length(col2), length(col3)
+  FROM mv_buffered_wide WHERE id IN (1, 50, 100) ORDER BY id;
+DROP MATERIALIZED VIEW mv_buffered_wide;
+
+-- Volatile-function fallback: random() triggers conservative guard.
+-- Result must still be correct through the single-row path.
+CREATE MATERIALIZED VIEW mv_volatile AS
+  SELECT g AS a, random() AS r FROM generate_series(1, 10) g;
+SELECT count(*) FROM mv_volatile;
+SELECT a FROM mv_volatile ORDER BY a;
+REFRESH MATERIALIZED VIEW mv_volatile;
+SELECT count(*) FROM mv_volatile;
+SELECT a FROM mv_volatile ORDER BY a;
+DROP MATERIALIZED VIEW mv_volatile;
+
+-- WITH NO DATA: unchanged behavior
+CREATE MATERIALIZED VIEW mv_nodata AS SELECT 1 AS a WITH NO DATA;
+SELECT count(*) FROM mv_nodata;  -- error: not populated
+REFRESH MATERIALIZED VIEW mv_nodata WITH NO DATA;
+DROP MATERIALIZED VIEW mv_nodata;
-- 
2.52.0