v7 parallel TID range scan patch

Author: Cary Huang

src/backend/access/heap/heapam.c

@@ -1478,6 +1478,19 @@ heap_set_tidrange(TableScanDesc sscan, ItemPointer mintid,
 	/* Set the start block and number of blocks to scan */
 	heap_setscanlimits(sscan, startBlk, numBlks);
 
+	/*
+	 * If parallel mode is used, store startBlk and numBlks in parallel
+	 * scan descriptor as well.
+	 */
+	if (scan->rs_base.rs_parallel != NULL)
+	{
+		ParallelBlockTableScanDesc bpscan = NULL;
+
+		bpscan = (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel;
+		bpscan->phs_startblock = startBlk;
+		bpscan->phs_numblock = numBlks;
+	}
+
 	/* Finally, set the TID range in sscan */
 	ItemPointerCopy(&lowestItem, &sscan->st.tidrange.rs_mintid);
 	ItemPointerCopy(&highestItem, &sscan->st.tidrange.rs_maxtid);

src/backend/access/table/tableam.c

@@ -188,6 +188,34 @@ table_beginscan_parallel(Relation relation, ParallelTableScanDesc pscan)
 											pscan, flags);
 }
 
+TableScanDesc
+table_beginscan_parallel_tidrange(Relation relation, ParallelTableScanDesc pscan)
+{
+	Snapshot	snapshot;
+	uint32		flags = SO_TYPE_TIDRANGESCAN | SO_ALLOW_PAGEMODE;
+	TableScanDesc sscan;
+
+	Assert(RelFileLocatorEquals(relation->rd_locator, pscan->phs_locator));
+
+	if (!pscan->phs_snapshot_any)
+	{
+		/* Snapshot was serialized -- restore it */
+		snapshot = RestoreSnapshot((char *) pscan + pscan->phs_snapshot_off);
+		RegisterSnapshot(snapshot);
+		flags |= SO_TEMP_SNAPSHOT;
+	}
+	else
+	{
+		/* SnapshotAny passed by caller (not serialized) */
+		snapshot = SnapshotAny;
+	}
+
+	sscan = relation->rd_tableam->scan_begin(relation, snapshot, 0, NULL,
+											 pscan, flags);
+
+	return sscan;
+}
+
 
 /* ----------------------------------------------------------------------------
  * Index scan related functions.
@@ -398,6 +426,7 @@ table_block_parallelscan_initialize(Relation rel, ParallelTableScanDesc pscan)
 		bpscan->phs_nblocks > NBuffers / 4;
 	SpinLockInit(&bpscan->phs_mutex);
 	bpscan->phs_startblock = InvalidBlockNumber;
+	bpscan->phs_numblock = InvalidBlockNumber;
 	pg_atomic_init_u64(&bpscan->phs_nallocated, 0);
 
 	return sizeof(ParallelBlockTableScanDescData);
@@ -577,8 +606,22 @@ table_block_parallelscan_nextpage(Relation rel,
 		pbscanwork->phsw_chunk_remaining = pbscanwork->phsw_chunk_size - 1;
 	}
 
+	/*
+	 * In a parallel TID range scan, 'pbscan->phs_numblock' is non-zero if an
+	 * upper TID range limit is specified, or InvalidBlockNumber if no limit
+	 * is given. This value may be less than or equal to 'pbscan->phs_nblocks'
+	 * , which is the total number of blocks in the relation.
+	 *
+	 * The scan can terminate early once 'nallocated' reaches
+	 * 'pbscan->phs_numblock', even if the full relation has remaining blocks
+	 * to scan. This ensures that parallel workers only scan the subset of
+	 * blocks that fall within the TID range.
+	 */
 	if (nallocated >= pbscan->phs_nblocks)
-		page = InvalidBlockNumber;	/* all blocks have been allocated */
+	    page = InvalidBlockNumber; /* all blocks have been allocated */
+	else if (pbscan->phs_numblock != InvalidBlockNumber &&
+			 nallocated >= pbscan->phs_numblock)
+		page = InvalidBlockNumber; /* upper scan limit reached */
 	else
 		page = (nallocated + pbscan->phs_startblock) % pbscan->phs_nblocks;
 

src/backend/executor/execParallel.c

@@ -41,6 +41,7 @@
 #include "executor/nodeSort.h"
 #include "executor/nodeSubplan.h"
 #include "executor/tqueue.h"
+#include "executor/nodeTidrangescan.h"
 #include "jit/jit.h"
 #include "nodes/nodeFuncs.h"
 #include "pgstat.h"
@@ -266,6 +267,11 @@ ExecParallelEstimate(PlanState *planstate, ExecParallelEstimateContext *e)
 				ExecForeignScanEstimate((ForeignScanState *) planstate,
 										e->pcxt);
 			break;
+		case T_TidRangeScanState:
+			if (planstate->plan->parallel_aware)
+				ExecTidRangeScanEstimate((TidRangeScanState *) planstate,
+										 e->pcxt);
+			break;
 		case T_AppendState:
 			if (planstate->plan->parallel_aware)
 				ExecAppendEstimate((AppendState *) planstate,
@@ -493,6 +499,11 @@ ExecParallelInitializeDSM(PlanState *planstate,
 				ExecForeignScanInitializeDSM((ForeignScanState *) planstate,
 											 d->pcxt);
 			break;
+		case T_TidRangeScanState:
+			if (planstate->plan->parallel_aware)
+				ExecTidRangeScanInitializeDSM((TidRangeScanState *) planstate,
+										 	  d->pcxt);
+			break;
 		case T_AppendState:
 			if (planstate->plan->parallel_aware)
 				ExecAppendInitializeDSM((AppendState *) planstate,
@@ -994,6 +1005,11 @@ ExecParallelReInitializeDSM(PlanState *planstate,
 				ExecForeignScanReInitializeDSM((ForeignScanState *) planstate,
 											   pcxt);
 			break;
+		case T_TidRangeScanState:
+			if (planstate->plan->parallel_aware)
+				ExecTidRangeScanReInitializeDSM((TidRangeScanState *) planstate,
+										   	    pcxt);
+			break;
 		case T_AppendState:
 			if (planstate->plan->parallel_aware)
 				ExecAppendReInitializeDSM((AppendState *) planstate, pcxt);
@@ -1020,7 +1036,6 @@ ExecParallelReInitializeDSM(PlanState *planstate,
 		case T_MemoizeState:
 			/* these nodes have DSM state, but no reinitialization is required */
 			break;
-
 		default:
 			break;
 	}
@@ -1362,6 +1377,11 @@ ExecParallelInitializeWorker(PlanState *planstate, ParallelWorkerContext *pwcxt)
 				ExecForeignScanInitializeWorker((ForeignScanState *) planstate,
 												pwcxt);
 			break;
+		case T_TidRangeScanState:
+			if (planstate->plan->parallel_aware)
+				ExecTidRangeScanInitializeWorker((TidRangeScanState *) planstate,
+												 pwcxt);
+			break;
 		case T_AppendState:
 			if (planstate->plan->parallel_aware)
 				ExecAppendInitializeWorker((AppendState *) planstate, pwcxt);

src/backend/executor/nodeTidrangescan.c

@@ -405,3 +405,84 @@ ExecInitTidRangeScan(TidRangeScan *node, EState *estate, int eflags)
 	 */
 	return tidrangestate;
 }
+/* ----------------------------------------------------------------
+ *						Parallel Scan Support
+ * ----------------------------------------------------------------
+ */
+
+/* ----------------------------------------------------------------
+ *		ExecTidRangeScanEstimate
+ *
+ *		Compute the amount of space we'll need in the parallel
+ *		query DSM, and inform pcxt->estimator about our needs.
+ * ----------------------------------------------------------------
+ */
+void
+ExecTidRangeScanEstimate(TidRangeScanState *node,
+					ParallelContext *pcxt)
+{
+	EState	   *estate = node->ss.ps.state;
+
+	node->trss_pscanlen = table_parallelscan_estimate(node->ss.ss_currentRelation,
+												  estate->es_snapshot);
+	shm_toc_estimate_chunk(&pcxt->estimator, node->trss_pscanlen);
+	shm_toc_estimate_keys(&pcxt->estimator, 1);
+}
+
+/* ----------------------------------------------------------------
+ *		ExecTidRangeScanInitializeDSM
+ *
+ *		Set up a parallel TID scan descriptor.
+ * ----------------------------------------------------------------
+ */
+void
+ExecTidRangeScanInitializeDSM(TidRangeScanState *node,
+						 ParallelContext *pcxt)
+{
+	EState	   *estate = node->ss.ps.state;
+	ParallelTableScanDesc pscan;
+
+	pscan = shm_toc_allocate(pcxt->toc, node->trss_pscanlen);
+	table_parallelscan_initialize(node->ss.ss_currentRelation,
+								  pscan,
+								  estate->es_snapshot);
+	/* disable syncscan in parallel tid range scan. */
+	pscan->phs_syncscan = false;
+	shm_toc_insert(pcxt->toc, node->ss.ps.plan->plan_node_id, pscan);
+	node->ss.ss_currentScanDesc =
+		table_beginscan_parallel_tidrange(node->ss.ss_currentRelation, pscan);
+}
+
+/* ----------------------------------------------------------------
+ *		ExecTidRangeScanReInitializeDSM
+ *
+ *		Reset shared state before beginning a fresh scan.
+ * ----------------------------------------------------------------
+ */
+void
+ExecTidRangeScanReInitializeDSM(TidRangeScanState *node,
+						   ParallelContext *pcxt)
+{
+	ParallelTableScanDesc pscan;
+
+	pscan = node->ss.ss_currentScanDesc->rs_parallel;
+	table_parallelscan_reinitialize(node->ss.ss_currentRelation, pscan);
+
+}
+
+/* ----------------------------------------------------------------
+ *		ExecTidRangeScanInitializeWorker
+ *
+ *		Copy relevant information from TOC into planstate.
+ * ----------------------------------------------------------------
+ */
+void
+ExecTidRangeScanInitializeWorker(TidRangeScanState *node,
+							ParallelWorkerContext *pwcxt)
+{
+	ParallelTableScanDesc pscan;
+
+	pscan = shm_toc_lookup(pwcxt->toc, node->ss.ps.plan->plan_node_id, false);
+	node->ss.ss_currentScanDesc =
+		table_beginscan_parallel_tidrange(node->ss.ss_currentRelation, pscan);
+}

src/backend/optimizer/path/costsize.c

@@ -1367,7 +1367,8 @@ cost_tidrangescan(Path *path, PlannerInfo *root,
 	Selectivity selectivity;
 	double		pages;
 	Cost		startup_cost = 0;
-	Cost		run_cost = 0;
+	Cost		cpu_run_cost = 0;
+	Cost		disk_run_cost = 0;
 	QualCost	qpqual_cost;
 	Cost		cpu_per_tuple;
 	QualCost	tid_qual_cost;
@@ -1396,11 +1397,7 @@ cost_tidrangescan(Path *path, PlannerInfo *root,
 
 	/*
 	 * The first page in a range requires a random seek, but each subsequent
-	 * page is just a normal sequential page read. NOTE: it's desirable for
-	 * TID Range Scans to cost more than the equivalent Sequential Scans,
-	 * because Seq Scans have some performance advantages such as scan
-	 * synchronization and parallelizability, and we'd prefer one of them to
-	 * be picked unless a TID Range Scan really is better.
+	 * page is just a normal sequential page read.
 	 */
 	ntuples = selectivity * baserel->tuples;
 	nseqpages = pages - 1.0;
@@ -1417,32 +1414,47 @@ cost_tidrangescan(Path *path, PlannerInfo *root,
 							  &spc_seq_page_cost);
 
 	/* disk costs; 1 random page and the remainder as seq pages */
-	run_cost += spc_random_page_cost + spc_seq_page_cost * nseqpages;
+	disk_run_cost += spc_random_page_cost + spc_seq_page_cost * nseqpages;
 
 	/* Add scanning CPU costs */
 	get_restriction_qual_cost(root, baserel, param_info, &qpqual_cost);
 
 	/*
 	 * XXX currently we assume TID quals are a subset of qpquals at this
 	 * point; they will be removed (if possible) when we create the plan, so
-	 * we subtract their cost from the total qpqual cost.  (If the TID quals
+	 * we subtract their cost from the total qpqual cost. (If the TID quals
 	 * can't be removed, this is a mistake and we're going to underestimate
 	 * the CPU cost a bit.)
 	 */
 	startup_cost += qpqual_cost.startup + tid_qual_cost.per_tuple;
 	cpu_per_tuple = cpu_tuple_cost + qpqual_cost.per_tuple -
 		tid_qual_cost.per_tuple;
-	run_cost += cpu_per_tuple * ntuples;
+	cpu_run_cost += cpu_per_tuple * ntuples;
 
 	/* tlist eval costs are paid per output row, not per tuple scanned */
 	startup_cost += path->pathtarget->cost.startup;
-	run_cost += path->pathtarget->cost.per_tuple * path->rows;
+	cpu_run_cost += path->pathtarget->cost.per_tuple * path->rows;
+
+	/* Adjust costing for parallelism, if used. */
+	if (path->parallel_workers > 0)
+	{
+		double		parallel_divisor = get_parallel_divisor(path);
+
+		/* The CPU cost is divided among all the workers. */
+		cpu_run_cost /= parallel_divisor;
+
+		/*
+		 * In the case of a parallel plan, the row count needs to represent
+		 * the number of tuples processed per worker.
+		 */
+		path->rows = clamp_row_est(path->rows / parallel_divisor);
+	}
 
 	/* we should not generate this path type when enable_tidscan=false */
 	Assert(enable_tidscan);
 	path->disabled_nodes = 0;
 	path->startup_cost = startup_cost;
-	path->total_cost = startup_cost + run_cost;
+	path->total_cost = startup_cost + cpu_run_cost + disk_run_cost;
 }
 
 /*

src/backend/optimizer/path/tidpath.c

@@ -47,6 +47,7 @@
 #include "optimizer/pathnode.h"
 #include "optimizer/paths.h"
 #include "optimizer/restrictinfo.h"
+#include "optimizer/cost.h"
 
 
 /*
@@ -553,7 +554,22 @@ create_tidscan_paths(PlannerInfo *root, RelOptInfo *rel)
 
 		add_path(rel, (Path *) create_tidrangescan_path(root, rel,
 														tidrangequals,
-														required_outer));
+														required_outer,
+														0));
+
+		/* If appropriate, consider parallel tid range scan. */
+		if (rel->consider_parallel && required_outer == NULL)
+		{
+			int			parallel_workers;
+
+			parallel_workers = compute_parallel_worker(rel, rel->pages, -1,
+													   max_parallel_workers_per_gather);
+			if (parallel_workers > 0)
+			{
+				add_partial_path(rel, (Path *) create_tidrangescan_path(root, rel, tidrangequals,
+								required_outer, parallel_workers));
+			}
+		}
 	}
 
 	/*

src/backend/optimizer/util/pathnode.c

@@ -1262,7 +1262,8 @@ create_tidscan_path(PlannerInfo *root, RelOptInfo *rel, List *tidquals,
  */
 TidRangePath *
 create_tidrangescan_path(PlannerInfo *root, RelOptInfo *rel,
-						 List *tidrangequals, Relids required_outer)
+						 List *tidrangequals, Relids required_outer,
+						 int parallel_workers)
 {
 	TidRangePath *pathnode = makeNode(TidRangePath);
 
@@ -1271,9 +1272,9 @@ create_tidrangescan_path(PlannerInfo *root, RelOptInfo *rel,
 	pathnode->path.pathtarget = rel->reltarget;
 	pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
 														  required_outer);
-	pathnode->path.parallel_aware = false;
+	pathnode->path.parallel_aware = (parallel_workers > 0);
 	pathnode->path.parallel_safe = rel->consider_parallel;
-	pathnode->path.parallel_workers = 0;
+	pathnode->path.parallel_workers = parallel_workers;
 	pathnode->path.pathkeys = NIL;	/* always unordered */
 
 	pathnode->tidrangequals = tidrangequals;

src/include/access/relscan.h

@@ -96,6 +96,8 @@ typedef struct ParallelBlockTableScanDescData
 	BlockNumber phs_nblocks;	/* # blocks in relation at start of scan */
 	slock_t		phs_mutex;		/* mutual exclusion for setting startblock */
 	BlockNumber phs_startblock; /* starting block number */
+	BlockNumber phs_numblock; 	/* # blocks to scan, or InvalidBlockNumber if
+								 * no limit */
 	pg_atomic_uint64 phs_nallocated;	/* number of blocks allocated to
 										 * workers so far. */
 }			ParallelBlockTableScanDescData;

src/include/access/tableam.h

@@ -1125,6 +1125,16 @@ extern void table_parallelscan_initialize(Relation rel,
 extern TableScanDesc table_beginscan_parallel(Relation relation,
 											  ParallelTableScanDesc pscan);
 
+/*
+ * Begin a parallel tidrange scan. `pscan` needs to have been initialized with
+ * table_parallelscan_initialize(), for the same relation. The initialization
+ * does not need to have happened in this backend.
+ *
+ * Caller must hold a suitable lock on the relation.
+ */
+extern TableScanDesc table_beginscan_parallel_tidrange(Relation relation,
+													   ParallelTableScanDesc pscan);
+
 /*
  * Restart a parallel scan.  Call this in the leader process.  Caller is
  * responsible for making sure that all workers have finished the scan