Move AQO from a relid based approach to a relation name based approach.

It allows us to reuse ML data at different instance and learn on temporary
tables.

Author: a.pervushina

aqo--1.2--1.3.sql

@@ -1,4 +1,4 @@
-ALTER TABLE public.aqo_data ADD COLUMN oids OID [] DEFAULT NULL;
+ALTER TABLE public.aqo_data ADD COLUMN oids text [] DEFAULT NULL;
 
 --
 -- Remove data, related to previously dropped tables, from the AQO tables.
@@ -9,7 +9,7 @@ DECLARE
     aqo_queries_row aqo_queries%ROWTYPE;
     aqo_query_texts_row aqo_query_texts%ROWTYPE;
     aqo_query_stat_row aqo_query_stat%ROWTYPE;
-    oid_var oid;
+    oid_var text;
     fspace_hash_var bigint;
     delete_row boolean DEFAULT false;
 BEGIN
@@ -23,7 +23,7 @@ BEGIN
     IF (aqo_data_row.oids IS NOT NULL) THEN
       FOREACH oid_var IN ARRAY aqo_data_row.oids
       LOOP
-        IF NOT EXISTS (SELECT relname FROM pg_class WHERE oid = oid_var) THEN
+        IF NOT EXISTS (SELECT relname FROM pg_class WHERE oid::regclass::text = oid_var) THEN
           delete_row = true;
         END IF;
       END LOOP;

aqo.h

@@ -281,13 +281,12 @@ extern bool find_query(uint64 qhash, QueryContextData *ctx);
 extern bool update_query(uint64 qhash, uint64 fhash,
 						 bool learn_aqo, bool use_aqo, bool auto_tuning);
 extern bool add_query_text(uint64 query_hash, const char *query_string);
-extern bool load_fss_ext(uint64 fs, int fss, OkNNrdata *data,
-						 List **relids, bool isSafe);
-extern bool load_fss(uint64 fhash, int fss_hash, OkNNrdata *data, List **relids);
-extern bool update_fss_ext(uint64 fhash, int fsshash, OkNNrdata *data,
-						   List *relids, bool isTimedOut);
-extern bool update_fss(uint64 fhash, int fss_hash, OkNNrdata *data,
-					   List *relids);
+extern bool load_fss_ext(uint64 fs, int fss, OkNNrdata *data, List **relnames,
+						 bool isSafe);
+extern bool load_fss(uint64 fs, int fss, OkNNrdata *data, List **relnames);
+extern bool update_fss_ext(uint64 fs, int fss, OkNNrdata *data,
+						   List *relnames, bool isTimedOut);
+extern bool update_fss(uint64 fs, int fss, OkNNrdata *data, List *relnames);
 QueryStat *get_aqo_stat(uint64 query_hash);
 void update_aqo_stat(uint64 query_hash, QueryStat * stat);
 extern bool my_index_insert(Relation indexRelation,	Datum *values, bool *isnull,
@@ -308,7 +307,7 @@ extern void print_node_explain(ExplainState *es, PlanState *ps, Plan *plan);
 
 /* Cardinality estimation */
 double predict_for_relation(List *restrict_clauses, List *selectivities,
-					 List *relids, int *fss_hash);
+							List *relnames, int *fss);
 
 /* Query execution statistics collecting hooks */
 void aqo_ExecutorStart(QueryDesc *queryDesc, int eflags);
@@ -320,13 +319,14 @@ void aqo_ExecutorEnd(QueryDesc *queryDesc);
 extern void automatical_query_tuning(uint64 query_hash, QueryStat * stat);
 
 /* Utilities */
-int			int_cmp(const void *a, const void *b);
-int			double_cmp(const void *a, const void *b);
-int *argsort(void *a, int n, size_t es,
-		int (*cmp) (const void *, const void *));
-int		   *inverse_permutation(int *a, int n);
-QueryStat  *palloc_query_stat(void);
-void		pfree_query_stat(QueryStat *stat);
+extern int int64_compare(const void *a, const void *b);
+extern int int_cmp(const void *a, const void *b);
+extern int double_cmp(const void *a, const void *b);
+extern int *argsort(void *a, int n, size_t es,
+					int (*cmp) (const void *, const void *));
+extern int *inverse_permutation(int *a, int n);
+extern QueryStat *palloc_query_stat(void);
+extern void pfree_query_stat(QueryStat *stat);
 
 /* Selectivity cache for parametrized baserels */
 extern void cache_selectivity(int clause_hash, int relid, int global_relid,

cardinality_estimation.c

@@ -26,7 +26,7 @@
 #ifdef AQO_DEBUG_PRINT
 static void
 predict_debug_output(List *clauses, List *selectivities,
-					 List *relids, int fss_hash, double result)
+					 List *relnames, int fss, double result)
 {
 	StringInfoData debug_str;
 	ListCell *lc;
@@ -42,11 +42,11 @@ predict_debug_output(List *clauses, List *selectivities,
 		appendStringInfo(&debug_str, "%lf ", *s);
 	}
 
-	appendStringInfoString(&debug_str, "}, relids: { ");
-	foreach(lc, relids)
+	appendStringInfoString(&debug_str, "}, relnames: { ");
+	foreach(lc, relnames)
 	{
-		int relid = lfirst_int(lc);
-		appendStringInfo(&debug_str, "%d ", relid);
+		String *relname = lfirst_node(String, lc);
+		appendStringInfo(&debug_str, "%s ", relname->sval);
 	}
 
 	appendStringInfo(&debug_str, "}, result: %lf", result);
@@ -60,22 +60,22 @@ predict_debug_output(List *clauses, List *selectivities,
  */
 double
 predict_for_relation(List *clauses, List *selectivities,
-					 List *relids, int *fss)
+					 List *relnames, int *fss)
 {
 	double	   *features;
 	double		result;
 	int			i;
 	OkNNrdata	data;
 
-	if (relids == NIL)
+	if (relnames == NIL)
 		/*
 		 * Don't make prediction for query plans without any underlying plane
 		 * tables. Use return value -4 for debug purposes.
 		 */
 		return -4.;
 
-	*fss = get_fss_for_object(relids, clauses,
-								   selectivities, &data.cols, &features);
+	*fss = get_fss_for_object(relnames, clauses, selectivities,
+							  &data.cols, &features);
 
 	if (data.cols > 0)
 		for (i = 0; i < aqo_K; ++i)
@@ -94,7 +94,7 @@ predict_for_relation(List *clauses, List *selectivities,
 		result = -1;
 	}
 #ifdef AQO_DEBUG_PRINT
-	predict_debug_output(clauses, selectivities, relids, *fss_hash, result);
+	predict_debug_output(clauses, selectivities, relnames, *fss, result);
 #endif
 	pfree(features);
 	if (data.cols > 0)

cardinality_hooks.c

@@ -139,8 +139,8 @@ void
 aqo_set_baserel_rows_estimate(PlannerInfo *root, RelOptInfo *rel)
 {
 	double	predicted;
-	Oid		relid;
-	List   *relids = NIL;
+	RangeTblEntry *rte;
+	List   *relnames = NIL;
 	List   *selectivities = NULL;
 	List   *clauses;
 	int		fss = 0;
@@ -161,19 +161,24 @@ aqo_set_baserel_rows_estimate(PlannerInfo *root, RelOptInfo *rel)
 		goto default_estimator;
 	}
 
-	relid = planner_rt_fetch(rel->relid, root)->relid;
-	if (OidIsValid(relid))
-		/* Predict for a plane table only. */
-		relids = list_make1_int(relid);
+	rte = planner_rt_fetch(rel->relid, root);
+	if (rte && OidIsValid(rte->relid))
+	{
+		String *s = makeNode(String);
+
+		/* Predict for a plane table. */
+		Assert(rte->eref && rte->eref->aliasname);
+		s->sval = pstrdup(rte->eref->aliasname);
+		relnames = list_make1(s);
+	}
 
 	clauses = aqo_get_clauses(root, rel->baserestrictinfo);
-	predicted = predict_for_relation(clauses, selectivities,
-									 relids, &fss);
+	predicted = predict_for_relation(clauses, selectivities, relnames, &fss);
 	rel->fss_hash = fss;
 
 	list_free_deep(selectivities);
 	list_free(clauses);
-	list_free(relids);
+	list_free(relnames);
 
 	if (predicted >= 0)
 	{
@@ -209,8 +214,8 @@ aqo_get_parameterized_baserel_size(PlannerInfo *root,
 								   List *param_clauses)
 {
 	double		predicted;
-	Oid			relid = InvalidOid;
-	List	   *relids = NIL;
+	RangeTblEntry *rte = NULL;
+	List	   *relnames = NIL;
 	List	   *allclauses = NULL;
 	List	   *selectivities = NULL;
 	ListCell   *l;
@@ -219,7 +224,7 @@ aqo_get_parameterized_baserel_size(PlannerInfo *root,
 	int		   *args_hash;
 	int		   *eclass_hash;
 	int			current_hash;
-	int fss = 0;
+	int			fss = 0;
 
 	if (IsQueryDisabled())
 		/* Fast path */
@@ -239,7 +244,7 @@ aqo_get_parameterized_baserel_size(PlannerInfo *root,
 		/* Make specific copy of clauses with mutated subplans */
 		allclauses = list_concat(aqo_get_clauses(root, param_clauses),
 								 aqo_get_clauses(root, rel->baserestrictinfo));
-		relid = planner_rt_fetch(rel->relid, root)->relid;
+		rte = planner_rt_fetch(rel->relid, root);
 		get_eclasses(allclauses, &nargs, &args_hash, &eclass_hash);
 
 		old_ctx_m = MemoryContextSwitchTo(AQO_cache_mem_ctx);
@@ -249,7 +254,7 @@ aqo_get_parameterized_baserel_size(PlannerInfo *root,
 			current_hash = get_clause_hash(
 										((RestrictInfo *) lfirst(l))->clause,
 										   nargs, args_hash, eclass_hash);
-			cache_selectivity(current_hash, rel->relid, relid,
+			cache_selectivity(current_hash, rel->relid, rte->relid,
 							  *((double *) lfirst(l2)));
 		}
 
@@ -269,11 +274,17 @@ aqo_get_parameterized_baserel_size(PlannerInfo *root,
 		goto default_estimator;
 	}
 
-	if (OidIsValid(relid))
-		/* Predict for a plane table only. */
-		relids = list_make1_int(relid);
+	if (rte && OidIsValid(rte->relid))
+	{
+		String *s = makeNode(String);
+
+		/* Predict for a plane table. */
+		Assert(rte->eref && rte->eref->aliasname);
+		s->sval = pstrdup(rte->eref->aliasname);
+		relnames = list_make1(s);
+	}
 
-	predicted = predict_for_relation(allclauses, selectivities, relids, &fss);
+	predicted = predict_for_relation(allclauses, selectivities, relnames, &fss);
 
 	predicted_ppi_rows = predicted;
 	fss_ppi_hash = fss;
@@ -298,7 +309,7 @@ aqo_set_joinrel_size_estimates(PlannerInfo *root, RelOptInfo *rel,
 							   List *restrictlist)
 {
 	double		predicted;
-	List	   *relids;
+	List	   *relnames;
 	List	   *outer_clauses;
 	List	   *inner_clauses;
 	List	   *allclauses;
@@ -324,7 +335,7 @@ aqo_set_joinrel_size_estimates(PlannerInfo *root, RelOptInfo *rel,
 		goto default_estimator;
 	}
 
-	relids = get_list_of_relids(root, rel->relids);
+	relnames = get_relnames(root, rel->relids);
 	outer_clauses = get_path_clauses(outer_rel->cheapest_total_path, root,
 									 &outer_selectivities);
 	inner_clauses = get_path_clauses(inner_rel->cheapest_total_path, root,
@@ -335,7 +346,7 @@ aqo_set_joinrel_size_estimates(PlannerInfo *root, RelOptInfo *rel,
 								list_concat(outer_selectivities,
 											inner_selectivities));
 
-	predicted = predict_for_relation(allclauses, selectivities, relids, &fss);
+	predicted = predict_for_relation(allclauses, selectivities, relnames, &fss);
 	rel->fss_hash = fss;
 
 	if (predicted >= 0)
@@ -366,7 +377,7 @@ aqo_get_parameterized_joinrel_size(PlannerInfo *root,
 								   List *clauses)
 {
 	double		predicted;
-	List	   *relids;
+	List	   *relnames;
 	List	   *outer_clauses;
 	List	   *inner_clauses;
 	List	   *allclauses;
@@ -392,7 +403,7 @@ aqo_get_parameterized_joinrel_size(PlannerInfo *root,
 		goto default_estimator;
 	}
 
-	relids = get_list_of_relids(root, rel->relids);
+	relnames = get_relnames(root, rel->relids);
 	outer_clauses = get_path_clauses(outer_path, root, &outer_selectivities);
 	inner_clauses = get_path_clauses(inner_path, root, &inner_selectivities);
 	allclauses = list_concat(aqo_get_clauses(root, clauses),
@@ -401,7 +412,7 @@ aqo_get_parameterized_joinrel_size(PlannerInfo *root,
 								list_concat(outer_selectivities,
 											inner_selectivities));
 
-	predicted = predict_for_relation(allclauses, selectivities, relids, &fss);
+	predicted = predict_for_relation(allclauses, selectivities, relnames, &fss);
 
 	predicted_ppi_rows = predicted;
 	fss_ppi_hash = fss;
@@ -428,13 +439,13 @@ predict_num_groups(PlannerInfo *root, Path *subpath, List *group_exprs,
 		child_fss = subpath->parent->fss_hash;
 	else
 	{
-		List *relids;
+		List *relnames;
 		List *clauses;
 		List *selectivities = NIL;
 
-		relids = get_list_of_relids(root, subpath->parent->relids);
+		relnames = get_relnames(root, subpath->parent->relids);
 		clauses = get_path_clauses(subpath, root, &selectivities);
-		(void) predict_for_relation(clauses, selectivities, relids, &child_fss);
+		(void) predict_for_relation(clauses, selectivities, relnames, &child_fss);
 	}
 
 	*fss = get_grouped_exprs_hash(child_fss, group_exprs);