If a join query uses a derived table (view / CTE) with GROUP BY clause then
the execution plan for such join may employ split optimization. When this
optimization is employed the derived table is not materialized. Rather only
some partitions of the derived table are subject to grouping. Split
optimization can be applied only if:
- there are some indexes over the tables used in the join specifying the
derived table whose prefixes partially cover the field items used in the
GROUP BY list (such indexes are called splitting indexes)
- the WHERE condition of the join query contains conjunctive equalities
between columns of the derived table that comprise major parts of
splitting indexes and columns of the other join tables.
When the optimizer evaluates extending of a partial join by the rows of the
derived table it always considers a possibility of using split optimization.
Different splitting indexes can be used depending on the extended partial
join. At some rare conditions, for example, when there is a non-splitting
covering index for a table joined in the join specifying the derived table
usage of a splitting index to produce rows needed for grouping may be still
less beneficial than usage of such covering index without any splitting
technique. The function JOIN_TAB::choose_best_splitting() must take this
into account.
Approved by Oleksandr Byelkin <sanja@mariadb.com>
The problem was caused by the following scenario:
Subquery's table has two indexes, KEY a(a), KEY a_b(a,b)
- LATERAL DERIVED optimization decides to use index a.
= The subquery uses ref access over key a.
- test_if_skip_sort_order() sees that KEY a_b satisfies the
subquery's GROUP BY clause, and attempts to switch to it.
= It fails to do so, because KEYUSE objects for index a_b
are switched off.
Fixed by disallowing to change the ref access key if it uses KEYUSE
objects injected by LATERAL DERIVED optimization.
- multi_range_read_info_const now uses the new records_in_range interface
- Added handler::avg_io_cost()
- Don't calculate avg_io_cost() in get_sweep_read_cost if avg_io_cost is
not 1.0. In this case we trust the avg_io_cost() from the handler.
- Changed test_quick_select to use TIME_FOR_COMPARE instead of
TIME_FOR_COMPARE_IDX to align this with the rest of the code.
- Fixed bug when using test_if_cheaper_ordering where we didn't use
keyread if index was changed
- Fixed a bug where we didn't use index only read when using order-by-index
- Added keyread_time() to HEAP.
The default keyread_time() was optimized for blocks and not suitable for
HEAP. The effect was the HEAP prefered table scans over ranges for btree
indexes.
- Fixed get_sweep_read_cost() for HEAP tables
- Ensure that range and ref have same cost for simple ranges
Added a small cost (MULTI_RANGE_READ_SETUP_COST) to ranges to ensure
we favior ref for range for simple queries.
- Fixed that matching_candidates_in_table() uses same number of records
as the rest of the optimizer
- Added avg_io_cost() to JT_EQ_REF cost. This helps calculate the cost for
HEAP and temporary tables better. A few tests changed because of this.
- heap::read_time() and heap::keyread_time() adjusted to not add +1.
This was to ensure that handler::keyread_time() doesn't give
higher cost for heap tables than for normal tables. One effect of
this is that heap and derived tables stored in heap will prefer
key access as this is now regarded as cheap.
- Changed cost for index read in sql_select.cc to match
multi_range_read_info_const(). All index cost calculation is now
done trough one function.
- 'ref' will now use quick_cost for keys if it exists. This is done
so that for '=' ranges, 'ref' is prefered over 'range'.
- scan_time() now takes avg_io_costs() into account
- get_delayed_table_estimates() uses block_size and avg_io_cost()
- Removed default argument to test_if_order_by_key(); simplifies code
If a splittable materialized derived table / view T is used in a inner nest
of an outer join with impossible ON condition then T is marked as a
constant table. Yet the execution plan to build T is still searched for
in spite of the fact that is not needed. So it should be set.
The function JOIN_TAB::choose_best_splitting() did not take into account
that for some tables whose fields were used in the GROUP BY list of
the specification of a splittable materialized derived there might exist
no elements in the array ext_keyuses_for_splitting.
The optimizer erroneously allowed to use join cache when joining a
splittable materialized table together with splitting optimization.
As a consequence in some rare cases the server returned wrong result
sets for queries with materialized derived.
This patch allows to use either join cache without usage of splitting
technique for materialization of a splittable derived table or splitting
without usage of join cache when joining such table. The costs the these
alternatives are compared and the best variant is chosen.
