diff options
Diffstat (limited to 'sql')
| -rw-r--r-- | sql/rowid_filter.h | 11 | ||||
| -rw-r--r-- | sql/sql_analyze_stmt.h | 7 | ||||
| -rw-r--r-- | sql/sql_explain.cc | 1 | ||||
| -rw-r--r-- | sql/sql_select.cc | 50 |
4 files changed, 62 insertions, 7 deletions
diff --git a/sql/rowid_filter.h b/sql/rowid_filter.h index 467b6884ca6..b76b8b1e635 100644 --- a/sql/rowid_filter.h +++ b/sql/rowid_filter.h @@ -192,6 +192,9 @@ public: */ virtual bool check(void *ctxt, char *elem) = 0; + /* True if the container does not contain any element */ + virtual bool is_empty() = 0; + virtual ~Rowid_filter_container() {} }; @@ -231,6 +234,8 @@ public: virtual ~Rowid_filter() {} + bool is_empty() { return container->is_empty(); } + Rowid_filter_container *get_container() { return container; } void set_tracker(Rowid_filter_tracker *track_arg) { tracker= track_arg; } @@ -268,6 +273,8 @@ public: bool check(char *elem) { + if (container->is_empty()) + return false; bool was_checked= container->check(table, elem); tracker->increment_checked_elements_count(was_checked); return was_checked; @@ -339,6 +346,8 @@ public: my_qsort2(array->front(), array->elements()/elem_size, elem_size, (qsort2_cmp) cmp, cmp_arg); } + + bool is_empty() { return elements() == 0; } }; @@ -368,6 +377,8 @@ public: bool add(void *ctxt, char *elem) { return refpos_container.add(elem); } bool check(void *ctxt, char *elem); + + bool is_empty() { return refpos_container.is_empty(); } }; /** diff --git a/sql/sql_analyze_stmt.h b/sql/sql_analyze_stmt.h index 990c79fb9ad..968b8d38295 100644 --- a/sql/sql_analyze_stmt.h +++ b/sql/sql_analyze_stmt.h @@ -416,12 +416,13 @@ public: uint get_container_elements() const { return container_elements; } + uint get_container_lookups() { return n_checks; } + double get_r_selectivity_pct() const { - return static_cast<double>(n_positive_checks) / - static_cast<double>(n_checks); + return n_checks ? static_cast<double>(n_positive_checks) / + static_cast<double>(n_checks) : 0; } size_t get_container_buff_size() const { return container_buff_size; } }; - diff --git a/sql/sql_explain.cc b/sql/sql_explain.cc index 539e427ee2f..1b59dce10b9 100644 --- a/sql/sql_explain.cc +++ b/sql/sql_explain.cc @@ -1676,6 +1676,7 @@ void Explain_rowid_filter::print_explain_json(Explain_query *query, if (is_analyze) { writer->add_member("r_rows").add_double(tracker->get_container_elements()); + writer->add_member("r_lookups").add_ll(tracker->get_container_lookups()); writer->add_member("r_selectivity_pct"). add_double(tracker->get_r_selectivity_pct() * 100.0); writer->add_member("r_buffer_size"). diff --git a/sql/sql_select.cc b/sql/sql_select.cc index 81c4991e801..3d880365d42 100644 --- a/sql/sql_select.cc +++ b/sql/sql_select.cc @@ -7496,9 +7496,11 @@ best_access_path(JOIN *join, double best= DBL_MAX; double best_time= DBL_MAX; double records= DBL_MAX; + ha_rows records_for_key= 0; table_map best_ref_depends_map= 0; Range_rowid_filter_cost_info *best_filter= 0; double tmp; + double keyread_tmp= 0; ha_rows rec; bool best_uses_jbuf= FALSE; MY_BITMAP *eq_join_set= &s->table->eq_join_set; @@ -7772,8 +7774,12 @@ best_access_path(JOIN *join, /* Limit the number of matched rows */ tmp= cost_for_index_read(thd, table, key, (ha_rows) records, (ha_rows) s->worst_seeks); + records_for_key= (ha_rows) records; + set_if_smaller(records_for_key, thd->variables.max_seeks_for_key); + keyread_tmp= table->file->keyread_time(key, 1, records_for_key); got_cost: tmp= COST_MULT(tmp, record_count); + keyread_tmp= COST_MULT(keyread_tmp, record_count); } } else @@ -7950,12 +7956,14 @@ best_access_path(JOIN *join, } /* Limit the number of matched rows */ - tmp= records; - set_if_smaller(tmp, (double) thd->variables.max_seeks_for_key); - tmp= cost_for_index_read(thd, table, key, (ha_rows) tmp, + tmp= cost_for_index_read(thd, table, key, (ha_rows) records, (ha_rows) s->worst_seeks); + records_for_key= (ha_rows) records; + set_if_smaller(records_for_key, thd->variables.max_seeks_for_key); + keyread_tmp= table->file->keyread_time(key, 1, records_for_key); got_cost2: tmp= COST_MULT(tmp, record_count); + keyread_tmp= COST_MULT(keyread_tmp, record_count); } else { @@ -7974,7 +7982,35 @@ best_access_path(JOIN *join, (found_part & 1)) // start_key->key can be used for index access { double rows= record_count * records; - double access_cost_factor= MY_MIN(tmp / rows, 1.0); + + /* + If we use filter F with selectivity s the the cost of fetching data + by key using this filter will be + cost_of_fetching_1_row * rows * s + + cost_of_fetching_1_key_tuple * rows * (1 - s) + + cost_of_1_lookup_into_filter * rows + Without using any filter the cost would be just + cost_of_fetching_1_row * rows + + So the gain in access cost per row will be + cost_of_fetching_1_row * (1 - s) - + cost_of_fetching_1_key_tuple * (1 - s) - + cost_of_1_lookup_into_filter + = + (cost_of_fetching_1_row - cost_of_fetching_1_key_tuple) * (1 - s) + - cost_of_1_lookup_into_filter + + Here we have: + cost_of_fetching_1_row = tmp/rows + cost_of_fetching_1_key_tuple = keyread_tmp/rows + + Note that access_cost_factor may be greater than 1.0. In this case + we still can expect a gain of using rowid filter due to smaller number + of checks for conditions pushed to the joined table. + */ + double rows_access_cost= MY_MIN(rows, s->worst_seeks); + double access_cost_factor= MY_MIN((rows_access_cost - keyread_tmp) / + rows, 1.0); filter= table->best_range_rowid_filter_for_partial_join(start_key->key, rows, access_cost_factor); @@ -8135,6 +8171,10 @@ best_access_path(JOIN *join, double rows= record_count * s->found_records; double access_cost_factor= MY_MIN(tmp / rows, 1.0); uint key_no= s->quick->index; + + /* See the comment concerning using rowid filter for with ref access */ + keyread_tmp= s->table->opt_range[key_no].index_only_cost; + access_cost_factor= MY_MIN((rows - keyread_tmp) / rows, 1.0); filter= s->table->best_range_rowid_filter_for_partial_join(key_no, rows, access_cost_factor); @@ -20929,6 +20969,8 @@ sub_select(JOIN *join,JOIN_TAB *join_tab,bool end_of_records) DBUG_RETURN(NESTED_LOOP_ERROR); join_tab->build_range_rowid_filter_if_needed(); + if (join_tab->rowid_filter && join_tab->rowid_filter->is_empty()) + rc= NESTED_LOOP_NO_MORE_ROWS; join->return_tab= join_tab; |