Fix time_bucket_gapfill inside LATERAL subqueries

svenklemm · timescale-automation · commit d11585d6c6d5 · 2026-02-23T14:22:15.000-05:00
The gapfill CustomScan node did not work correctly when used inside a LATERAL subquery because gapfill_rescan had two bugs: 1. It did not call UpdateChangedParamSet on its child plan state before ExecReScan, so the child plan did not know that LATERAL parameters changed and kept returning stale data from the previous outer row. 2. It did not reset internal gapfill state (next_timestamp, next_offset, groups_initialized, column states) on rescan, so gap-filling was skipped on subsequent scans since next_timestamp was already at gapfill_end. Fixes #4693 (cherry picked from commit 7c15939)
diff --git a/.unreleased/pr_9293 b/.unreleased/pr_9293
@@ -0,0 +1,2 @@
+Fixes: #9293 Fix time_bucket_gapfill inside LATERAL subqueries
+Thanks: @CaptainCuddleCube for reporting an issue with time_bucket_gapfill and LATERAL subqueries
diff --git a/tsl/src/nodes/gapfill/gapfill_exec.c b/tsl/src/nodes/gapfill/gapfill_exec.c
@@ -914,11 +914,49 @@ gapfill_end(CustomScanState *node)
 static void
 gapfill_rescan(CustomScanState *node)
 {
+	GapFillState *state = (GapFillState *) node;
+
 	if (node->custom_ps != NIL)
 	{
+		if (node->ss.ps.chgParam != NULL)
+			UpdateChangedParamSet(linitial(node->custom_ps), node->ss.ps.chgParam);
 		ExecReScan(linitial(node->custom_ps));
 	}
-	((GapFillState *) node)->state = FETCHED_NONE;
+
+	state->state = FETCHED_NONE;
+	state->next_timestamp = state->gapfill_start;
+	state->next_offset = state->gapfill_interval;
+
+	if (state->multigroup)
+		state->groups_initialized = false;
+
+	/* Reset column states for locf and interpolate */
+	for (int i = 0; i < state->ncolumns; i++)
+	{
+		GapFillColumnState *column = state->columns[i];
+		switch (column->ctype)
+		{
+			case LOCF_COLUMN:
+				gapfill_locf_group_change((GapFillLocfColumnState *) column);
+				break;
+			case INTERPOLATE_COLUMN:
+			{
+				GapFillInterpolateColumnState *ic = (GapFillInterpolateColumnState *) column;
+				ic->prev.isnull = true;
+				ic->next.isnull = true;
+				break;
+			}
+			case GROUP_COLUMN:
+			case DERIVED_COLUMN:
+			{
+				GapFillGroupColumnState *gc = (GapFillGroupColumnState *) column;
+				gc->isnull = true;
+				break;
+			}
+			default:
+				break;
+		}
+	}
 }
 
 static void
diff --git a/tsl/src/nodes/gapfill/gapfill_plan.c b/tsl/src/nodes/gapfill/gapfill_plan.c
@@ -488,10 +488,11 @@ plan_add_gapfill(PlannerInfo *root, RelOptInfo *group_rel)
 		group_rel->cheapest_startup_path = NULL;
 		group_rel->cheapest_unique_path = NULL;
 
-		/* Parameterized paths pathlist is currently deleted instead of being processed */
-		list_free(group_rel->ppilist);
-		group_rel->ppilist = NULL;
-
+		/*
+		 * cheapest_parameterized_paths will be rebuilt by set_cheapest()
+		 * after this hook returns. We must not delete ppilist as it contains
+		 * ParamPathInfo entries needed for parameterized paths (e.g. LATERAL).
+		 */
 		list_free(group_rel->cheapest_parameterized_paths);
 		group_rel->cheapest_parameterized_paths = NULL;
 
diff --git a/tsl/test/shared/expected/gapfill_bug.out b/tsl/test/shared/expected/gapfill_bug.out
@@ -363,3 +363,63 @@ SELECT
  Fri Sep 30 22:00:00 2022 PDT |      |      |        5
 
 drop table hourly cascade;
+-- github issue #4693: test time_bucket_gapfill inside LATERAL subquery
+CREATE TYPE i4693_type AS (time timestamptz, num bigint);
+CREATE TABLE i4693(id integer, reports i4693_type[]);
+INSERT INTO i4693(id, reports) VALUES
+    (1, ARRAY[ROW('2020-09-01 13:22:02+00', 360), ROW('2020-09-01 13:24:02+00', 3600)]::i4693_type[]),
+    (2, ARRAY[ROW('2020-09-01 13:22:02+00', 90), ROW('2020-09-01 13:25:02+00', 900)]::i4693_type[]);
+-- Gapfill inside LATERAL should produce gap-filled rows for each outer row
+SELECT id, rep.* FROM i4693, LATERAL (
+    SELECT time_bucket_gapfill('1 minute', time,
+        start := '2020-09-01 13:22:00+00'::timestamptz,
+        finish := '2020-09-01 13:28:00+00'::timestamptz) AS mins,
+        sum(num)
+    FROM unnest(reports)
+    GROUP BY mins
+) AS rep
+ORDER BY id, mins;
+ id |             mins             | sum  
+----+------------------------------+------
+  1 | Tue Sep 01 06:22:00 2020 PDT |  360
+  1 | Tue Sep 01 06:23:00 2020 PDT |     
+  1 | Tue Sep 01 06:24:00 2020 PDT | 3600
+  1 | Tue Sep 01 06:25:00 2020 PDT |     
+  1 | Tue Sep 01 06:26:00 2020 PDT |     
+  1 | Tue Sep 01 06:27:00 2020 PDT |     
+  2 | Tue Sep 01 06:22:00 2020 PDT |   90
+  2 | Tue Sep 01 06:23:00 2020 PDT |     
+  2 | Tue Sep 01 06:24:00 2020 PDT |     
+  2 | Tue Sep 01 06:25:00 2020 PDT |  900
+  2 | Tue Sep 01 06:26:00 2020 PDT |     
+  2 | Tue Sep 01 06:27:00 2020 PDT |     
+
+-- test with locf and interpolate
+SELECT id, rep.* FROM i4693, LATERAL (
+    SELECT time_bucket_gapfill('1 minute', time,
+        start := '2020-09-01 13:22:00+00'::timestamptz,
+        finish := '2020-09-01 13:28:00+00'::timestamptz) AS mins,
+        locf(sum(num)),
+        interpolate(sum(num)),
+        sum(num)
+    FROM unnest(reports)
+    GROUP BY mins
+) AS rep
+ORDER BY id, mins;
+ id |             mins             | locf | interpolate | sum  
+----+------------------------------+------+-------------+------
+  1 | Tue Sep 01 06:22:00 2020 PDT |  360 |         360 |  360
+  1 | Tue Sep 01 06:23:00 2020 PDT |  360 |        1980 |     
+  1 | Tue Sep 01 06:24:00 2020 PDT | 3600 |        3600 | 3600
+  1 | Tue Sep 01 06:25:00 2020 PDT | 3600 |             |     
+  1 | Tue Sep 01 06:26:00 2020 PDT | 3600 |             |     
+  1 | Tue Sep 01 06:27:00 2020 PDT | 3600 |             |     
+  2 | Tue Sep 01 06:22:00 2020 PDT |   90 |          90 |   90
+  2 | Tue Sep 01 06:23:00 2020 PDT |   90 |         360 |     
+  2 | Tue Sep 01 06:24:00 2020 PDT |   90 |         630 |     
+  2 | Tue Sep 01 06:25:00 2020 PDT |  900 |         900 |  900
+  2 | Tue Sep 01 06:26:00 2020 PDT |  900 |             |     
+  2 | Tue Sep 01 06:27:00 2020 PDT |  900 |             |     
+
+DROP TABLE i4693;
+DROP TYPE i4693_type;
diff --git a/tsl/test/shared/sql/gapfill_bug.sql b/tsl/test/shared/sql/gapfill_bug.sql
@@ -197,3 +197,38 @@ SELECT
  GROUP BY 1,  agg+signal order by 1,2,3;
 
 drop table hourly cascade;
+
+-- github issue #4693: test time_bucket_gapfill inside LATERAL subquery
+CREATE TYPE i4693_type AS (time timestamptz, num bigint);
+CREATE TABLE i4693(id integer, reports i4693_type[]);
+
+INSERT INTO i4693(id, reports) VALUES
+    (1, ARRAY[ROW('2020-09-01 13:22:02+00', 360), ROW('2020-09-01 13:24:02+00', 3600)]::i4693_type[]),
+    (2, ARRAY[ROW('2020-09-01 13:22:02+00', 90), ROW('2020-09-01 13:25:02+00', 900)]::i4693_type[]);
+
+-- Gapfill inside LATERAL should produce gap-filled rows for each outer row
+SELECT id, rep.* FROM i4693, LATERAL (
+    SELECT time_bucket_gapfill('1 minute', time,
+        start := '2020-09-01 13:22:00+00'::timestamptz,
+        finish := '2020-09-01 13:28:00+00'::timestamptz) AS mins,
+        sum(num)
+    FROM unnest(reports)
+    GROUP BY mins
+) AS rep
+ORDER BY id, mins;
+
+-- test with locf and interpolate
+SELECT id, rep.* FROM i4693, LATERAL (
+    SELECT time_bucket_gapfill('1 minute', time,
+        start := '2020-09-01 13:22:00+00'::timestamptz,
+        finish := '2020-09-01 13:28:00+00'::timestamptz) AS mins,
+        locf(sum(num)),
+        interpolate(sum(num)),
+        sum(num)
+    FROM unnest(reports)
+    GROUP BY mins
+) AS rep
+ORDER BY id, mins;
+
+DROP TABLE i4693;
+DROP TYPE i4693_type;

Original file line number	Diff line number	Diff line change
`@@ -0,0 +1,2 @@`
	`1`	`+Fixes: #9293 Fix time_bucket_gapfill inside LATERAL subqueries`
	`2`	`+Thanks: @CaptainCuddleCube for reporting an issue with time_bucket_gapfill and LATERAL subqueries`