Changeset: 654ad5dc8103 for MonetDB
URL: https://dev.monetdb.org/hg/MonetDB/rev/654ad5dc8103
Modified Files:
sql/server/rel_optimizer.c
Branch: properties
Log Message:
Moved rel_select_order and rel_groupby_order for later, so they can be used
with statistics
diffs (105 lines):
diff --git a/sql/server/rel_optimizer.c b/sql/server/rel_optimizer.c
--- a/sql/server/rel_optimizer.c
+++ b/sql/server/rel_optimizer.c
@@ -4569,7 +4569,7 @@ point_select_on_unique_column(sql_rel *r
* A point select on an unique column reduces the number of rows to 1. If the
same select is under a
* join, the opposite side's select can be pushed above the join.
*/
-static sql_rel *
+static inline sql_rel *
rel_push_select_up(visitor *v, sql_rel *rel)
{
if ((is_join(rel->op) || is_semi(rel->op)) && !rel_is_ref(rel) &&
!is_single(rel)) {
@@ -9475,7 +9475,6 @@ rel_optimize_projections(visitor *v, sql
rel = rel_push_aggr_down(v, rel);
rel = rel_push_groupby_down(v, rel);
- rel = rel_groupby_order(v, rel);
rel = rel_reduce_groupby_exps(v, rel);
rel = rel_distinct_aggregate_on_unique_values(v, rel);
rel = rel_groupby_distinct(v, rel);
@@ -9527,8 +9526,6 @@ rel_optimize_select_and_joins_topdown(vi
rel = rel_push_select_down(v, rel);
if (rel && rel->l && (is_select(rel->op) || is_join(rel->op)))
rel = rel_use_index(v, rel);
-
- rel = rel_select_order(v, rel);
return rel;
}
@@ -9608,7 +9605,7 @@ optimize_rel(visitor *v, sql_rel *rel, g
if (gp->cnt[op_join] || gp->cnt[op_left] || gp->cnt[op_right] ||
gp->cnt[op_full] || gp->cnt[op_semi] || gp->cnt[op_anti]) {
rel = rel_visitor_topdown(v, rel, &rel_optimize_joins);
- if (!gp->cnt[op_update])
+ if (level <= 1 &&!gp->cnt[op_update])
rel = rel_join_order(v, rel);
/* Important -> Re-write semijoins after rel_join_order */
rel = rel_visitor_bottomup(v, rel, &rel_optimize_semi_and_anti);
@@ -9626,12 +9623,6 @@ optimize_rel(visitor *v, sql_rel *rel, g
if (level <= 0)
rel = rel_dce(v->sql, rel);
- if (v->storage_based_opt && level <= 0) { /* storage statistics related
optimizations */
- /* Don't prune updates as pruning will possibly result in
removing the joins which therefor cannot be used for constraint checking */
- if (!(is_modify(rel->op) && rel->flag&UPD_COMP))
- rel = rel_visitor_bottomup(v, rel, &rel_get_statistics);
- }
-
if (gp->cnt[op_join] || gp->cnt[op_left] || gp->cnt[op_right] ||
gp->cnt[op_full] || gp->cnt[op_semi] || gp->cnt[op_anti] || gp->cnt[op_select])
rel = rel_visitor_topdown(v, rel,
&rel_push_func_and_select_down);
@@ -9817,6 +9808,22 @@ rel_setjoins_2_joingroupby(visitor *v, s
return rel;
}
+/* This optimization loop contains optimizations that can potentially use
statistics */
+static sql_rel *
+rel_final_optimization_loop(visitor *v, sql_rel *rel)
+{
+ /* Run rel_push_select_up only once at the end to avoid an infinite
optimization loop */
+ rel = rel_push_select_up(v, rel);
+ rel = rel_select_order(v, rel);
+
+ /* TODO? Maybe later add rel_simplify_count, rel_join2semijoin,
rel_simplify_fk_joins,
+ rel_distinct_project2groupby, rel_simplify_predicates,
rel_simplify_math,
+ rel_distinct_aggregate_on_unique_values */
+
+ rel = rel_groupby_order(v, rel);
+ return rel;
+}
+
/* 'instantiate' means to rewrite logical tables: (merge, remote, replica
tables) */
sql_rel *
rel_optimizer(mvc *sql, sql_rel *rel, int instantiate, int value_based_opt,
int storage_based_opt)
@@ -9844,15 +9851,25 @@ rel_optimizer(mvc *sql, sql_rel *rel, in
#ifndef NDEBUG
assert(level < 20);
#endif
- /* Run rel_push_select_up only once at the end to avoid an infinite
optimization loop */
- if (opt == 2 && gp.cnt[op_select] && (gp.cnt[op_join] ||
gp.cnt[op_left] || gp.cnt[op_right] || gp.cnt[op_full] || gp.cnt[op_semi] ||
gp.cnt[op_anti]))
- rel = rel_visitor_bottomup(&v, rel, &rel_push_select_up);
- if (gp.needs_setjoin_rewrite)
+ if (rel && gp.needs_setjoin_rewrite)
rel = rel_visitor_bottomup(&v, rel,
&rel_setjoins_2_joingroupby);
+ if (rel && opt == 2) {
+ if (storage_based_opt && !(is_modify(rel->op) &&
rel->flag&UPD_COMP)) { /* storage statistics related optimizations */
+ /* Don't prune updates as pruning will possibly result
in removing the joins which therefor cannot be used for constraint checking */
+ rel = rel_visitor_bottomup(&v, rel,
&rel_get_statistics);
+ }
+ /* Run join order one final time with statistics */
+ if (!gp.cnt[op_update] && (gp.cnt[op_join] || gp.cnt[op_left]
|| gp.cnt[op_right] || gp.cnt[op_full]))
+ rel = rel_join_order(&v, rel);
+
+ if (gp.cnt[op_groupby] || gp.cnt[op_select] || gp.cnt[op_join]
|| gp.cnt[op_left] ||
+ gp.cnt[op_right] || gp.cnt[op_full] || gp.cnt[op_semi]
|| gp.cnt[op_anti])
+ rel = rel_visitor_bottomup(&v, rel,
&rel_final_optimization_loop);
+ }
/* merge table rewrites may introduce remote or replica tables */
/* at the moment, make sure the remote table rewriters always run last*/
- if (gp.needs_mergetable_rewrite || gp.needs_remote_replica_rewrite) {
+ if (rel && (gp.needs_mergetable_rewrite ||
gp.needs_remote_replica_rewrite)) {
v.data = NULL;
rel = rel_visitor_bottomup(&v, rel, &rel_rewrite_remote);
v.data = NULL;
_______________________________________________
checkin-list mailing list
[email protected]
https://www.monetdb.org/mailman/listinfo/checkin-list
