Fix bug with explain and Add more message about join type.

src/backend/commands/explain.c

@@ -2296,8 +2296,9 @@ ExplainNode(PlanState *planstate, List *ancestors,
 			/* fall through to print additional fields the same as SeqScan */
 			/* FALLTHROUGH */
 		case T_SeqScan:
-			show_pushdown_runtime_filter_info("Rows Removed by Pushdown Runtime Filter",
-											  planstate, es);
+			if (gp_enable_runtime_filter_pushdown && IsA(planstate, SeqScanState))
+				show_pushdown_runtime_filter_info("Rows Removed by Pushdown Runtime Filter",
+												  planstate, es);
 			/* FALLTHROUGH */
 		case T_ValuesScan:
 		case T_CteScan:
@@ -4144,25 +4145,21 @@ show_instrumentation_count(const char *qlabel, int which,
 }
 
 /*
- * If it's EXPLAIN ANALYZE, show instrumentation information with pushdown runtime filter
+ * If it's EXPLAIN ANALYZE, show instrumentation information with pushdown
+ * runtime filter.
  */
 static void
-show_pushdown_runtime_filter_info(const char *qlabel, PlanState *planstate, ExplainState *es)
+show_pushdown_runtime_filter_info(const char *qlabel,
+								  PlanState *planstate,
+								  ExplainState *es)
 {
-	double		nfiltered;
-
-	Assert(IsA(planstate, SeqScanState));
+	Assert(gp_enable_runtime_filter_pushdown && IsA(planstate, SeqScanState));
 
 	if (!es->analyze || !planstate->instrument)
 		return;
 
-	nfiltered = planstate->instrument->nfilteredPRF;
-
-	/* In text mode, suppress zero counts; they're not interesting enough */
-	if (nfiltered > 0 || es->format != EXPLAIN_FORMAT_TEXT)
-	{
-		ExplainPropertyFloat(qlabel, NULL, nfiltered, 0, es);
-	}
+	if (planstate->instrument->prf_work)
+		ExplainPropertyFloat(qlabel, NULL, planstate->instrument->nfilteredPRF, 0, es);
 }
 
 /*

src/backend/commands/explain_gp.c

@@ -57,6 +57,7 @@ typedef struct CdbExplain_StatInst
 	double		nloops;			/* # of run cycles for this node */
 	double		nfiltered1;
 	double		nfiltered2;
+	bool		prf_work;
 	double		nfilteredPRF;
 	double		execmemused;	/* executor memory used (bytes) */
 	double		workmemused;	/* work_mem actually used (bytes) */
@@ -887,6 +888,7 @@ cdbexplain_collectStatsFromNode(PlanState *planstate, CdbExplain_SendStatCtx *ct
 	si->nloops = instr->nloops;
 	si->nfiltered1 = instr->nfiltered1;
 	si->nfiltered2 = instr->nfiltered2;
+	si->prf_work     = instr->prf_work;
 	si->nfilteredPRF = instr->nfilteredPRF;
 	si->workmemused = instr->workmemused;
 	si->workmemwanted = instr->workmemwanted;
@@ -1218,6 +1220,7 @@ cdbexplain_depositStatsToNode(PlanState *planstate, CdbExplain_RecvStatCtx *ctx)
 		instr->nloops = nodeAcc->nsimax->nloops;
 		instr->nfiltered1 = nodeAcc->nsimax->nfiltered1;
 		instr->nfiltered2 = nodeAcc->nsimax->nfiltered2;
+		instr->prf_work     = nodeAcc->nsimax->prf_work;
 		instr->nfilteredPRF = nodeAcc->nsimax->nfilteredPRF;
 		instr->execmemused = nodeAcc->nsimax->execmemused;
 		instr->workmemused = nodeAcc->nsimax->workmemused;

src/backend/executor/nodeHashjoin.c

@@ -2195,13 +2195,19 @@ CreateRuntimeFilter(HashJoinState* hjstate)
 	List		*targets;
 
 	/*
-	 * Only applicatable for inner, right and semi join,
+	 * A build-side Bloom filter tells us if a row is definitely not in the build
+	 * side. This allows us to early-eliminate rows or early-accept rows depending
+	 * on the type of join.
+	 * Left Outer Join and Full Outer Join output all rows, so a build-side Bloom
+	 * filter would only allow us to early-output. Left Antijoin outputs only if
+	 * there is no match, so again early output. We don't implement early output
+	 * for now.
+	 * So it's only applicatable for inner, right and semi join.
 	 */
 	jointype = hjstate->js.jointype;
-	if (jointype != JOIN_INNER
-		&& jointype != JOIN_RIGHT
-		&& jointype != JOIN_SEMI
-	   )
+	if (jointype != JOIN_INNER &&
+		jointype != JOIN_RIGHT &&
+		jointype != JOIN_SEMI)
 		return;
 
 	hstate = castNode(HashState, innerPlanState(hjstate));

src/backend/executor/nodeSeqscan.c

@@ -408,6 +408,14 @@ PassByBloomFilter(SeqScanState *node, TupleTableSlot *slot)
 	ListCell *lc;
 	bloom_filter *blm_filter;
 
+	/*
+	 * Mark that the pushdown runtime filter is actually taking effect.
+	 */
+	if (node->ss.ps.instrument &&
+		!node->ss.ps.instrument->prf_work &&
+		list_length(node->filters))
+		node->ss.ps.instrument->prf_work = true;
+
 	foreach (lc, node->filters)
 	{
 		sk = lfirst(lc);

src/include/executor/instrument.h

@@ -78,6 +78,7 @@ typedef struct Instrumentation
 	bool		need_bufusage;	/* true if we need buffer usage data */
 	bool		need_walusage;	/* true if we need WAL usage data */
 	bool		async_mode;		/* true if node is in async mode */
+	bool		prf_work;		/* true if pushdown runtime filters really work */
 	/* Info about current plan cycle: */
 	bool		running;		/* true if we've completed first tuple */
 	instr_time	starttime;		/* Start time of current iteration of node */

src/test/regress/expected/gp_runtime_filter.out

@@ -358,6 +358,59 @@ SELECT * FROM t1, t2 WHERE t1.c2 = t2.c2;
  Optimizer: Postgres query optimizer
 (15 rows)
 
+RESET gp_enable_runtime_filter_pushdown;
+DROP TABLE IF EXISTS t1;
+DROP TABLE IF EXISTS t2;
+-- case 3: bug fix with explain
+DROP TABLE IF EXISTS test_tablesample;
+NOTICE:  table "test_tablesample" does not exist, skipping
+CREATE TABLE test_tablesample (dist int, id int, name text) WITH (fillfactor=10) DISTRIBUTED BY (dist);
+INSERT INTO test_tablesample SELECT 0, i, repeat(i::text, 875) FROM generate_series(0, 9) s(i) ORDER BY i;
+INSERT INTO test_tablesample SELECT 3, i, repeat(i::text, 875) FROM generate_series(10, 19) s(i) ORDER BY i;
+INSERT INTO test_tablesample SELECT 5, i, repeat(i::text, 875) FROM generate_series(20, 29) s(i) ORDER BY i;
+SET gp_enable_runtime_filter_pushdown TO on;
+EXPLAIN (COSTS OFF) SELECT id FROM test_tablesample TABLESAMPLE SYSTEM (50) REPEATABLE (2);
+                                QUERY PLAN                                
+--------------------------------------------------------------------------
+ Gather Motion 3:1  (slice1; segments: 3)
+   ->  Sample Scan on test_tablesample
+         Sampling: system ('50'::real) REPEATABLE ('2'::double precision)
+ Optimizer: Postgres query optimizer
+(4 rows)
+
+RESET gp_enable_runtime_filter_pushdown;
+DROP TABLE IF EXISTS test_tablesample;
+-- case 4: show debug info only when gp_enable_runtime_filter_pushdown is on
+DROP TABLE IF EXISTS t1;
+NOTICE:  table "t1" does not exist, skipping
+DROP TABLE IF EXISTS t2;
+NOTICE:  table "t2" does not exist, skipping
+CREATE TABLE t1(c1 int, c2 int);
+NOTICE:  Table doesn't have 'DISTRIBUTED BY' clause -- Using column named 'c1' as the Cloudberry Database data distribution key for this table.
+HINT:  The 'DISTRIBUTED BY' clause determines the distribution of data. Make sure column(s) chosen are the optimal data distribution key to minimize skew.
+CREATE TABLE t2(c1 int, c2 int);
+NOTICE:  Table doesn't have 'DISTRIBUTED BY' clause -- Using column named 'c1' as the Cloudberry Database data distribution key for this table.
+HINT:  The 'DISTRIBUTED BY' clause determines the distribution of data. Make sure column(s) chosen are the optimal data distribution key to minimize skew.
+INSERT INTO t1 SELECT GENERATE_SERIES(1, 1000), GENERATE_SERIES(1, 1000);
+INSERT INTO t2 SELECT * FROM t1;
+SET gp_enable_runtime_filter_pushdown TO on;
+EXPLAIN (ANALYZE, COSTS OFF, SUMMARY OFF, TIMING OFF) SELECT count(t1.c2) FROM t1, t2 WHERE t1.c1 = t2.c1;
+                                               QUERY PLAN                                                
+---------------------------------------------------------------------------------------------------------
+ Finalize Aggregate (actual rows=1 loops=1)
+   ->  Gather Motion 3:1  (slice1; segments: 3) (actual rows=3 loops=1)
+         ->  Partial Aggregate (actual rows=1 loops=1)
+               ->  Hash Join (actual rows=340 loops=1)
+                     Hash Cond: (t1.c1 = t2.c1)
+                     Extra Text: (seg2)   Hash chain length 1.0 avg, 1 max, using 340 of 524288 buckets.
+                     ->  Seq Scan on t1 (actual rows=340 loops=1)
+                           Rows Removed by Pushdown Runtime Filter: 0
+                     ->  Hash (actual rows=340 loops=1)
+                           Buckets: 524288  Batches: 1  Memory Usage: 4108kB
+                           ->  Seq Scan on t2 (actual rows=340 loops=1)
+ Optimizer: Postgres query optimizer
+(12 rows)
+
 RESET gp_enable_runtime_filter_pushdown;
 DROP TABLE IF EXISTS t1;
 DROP TABLE IF EXISTS t2;

src/test/regress/sql/gp_runtime_filter.sql

@@ -136,6 +136,34 @@ RESET gp_enable_runtime_filter_pushdown;
 DROP TABLE IF EXISTS t1;
 DROP TABLE IF EXISTS t2;
 
+-- case 3: bug fix with explain
+DROP TABLE IF EXISTS test_tablesample;
+CREATE TABLE test_tablesample (dist int, id int, name text) WITH (fillfactor=10) DISTRIBUTED BY (dist);
+INSERT INTO test_tablesample SELECT 0, i, repeat(i::text, 875) FROM generate_series(0, 9) s(i) ORDER BY i;
+INSERT INTO test_tablesample SELECT 3, i, repeat(i::text, 875) FROM generate_series(10, 19) s(i) ORDER BY i;
+INSERT INTO test_tablesample SELECT 5, i, repeat(i::text, 875) FROM generate_series(20, 29) s(i) ORDER BY i;
+
+SET gp_enable_runtime_filter_pushdown TO on;
+EXPLAIN (COSTS OFF) SELECT id FROM test_tablesample TABLESAMPLE SYSTEM (50) REPEATABLE (2);
+RESET gp_enable_runtime_filter_pushdown;
+
+DROP TABLE IF EXISTS test_tablesample;
+
+-- case 4: show debug info only when gp_enable_runtime_filter_pushdown is on
+DROP TABLE IF EXISTS t1;
+DROP TABLE IF EXISTS t2;
+CREATE TABLE t1(c1 int, c2 int);
+CREATE TABLE t2(c1 int, c2 int);
+INSERT INTO t1 SELECT GENERATE_SERIES(1, 1000), GENERATE_SERIES(1, 1000);
+INSERT INTO t2 SELECT * FROM t1;
+
+SET gp_enable_runtime_filter_pushdown TO on;
+EXPLAIN (ANALYZE, COSTS OFF, SUMMARY OFF, TIMING OFF) SELECT count(t1.c2) FROM t1, t2 WHERE t1.c1 = t2.c1;
+RESET gp_enable_runtime_filter_pushdown;
+
+DROP TABLE IF EXISTS t1;
+DROP TABLE IF EXISTS t2;
+
 -- Clean up: reset guc
 SET gp_enable_runtime_filter TO off;
 SET optimizer TO default;