Support index condition pushdown within partitioned tables.
- ha_partition will pass the pushed index condition into all of the used
partitions.
- We require that all of the partitions to handle the pushed index
condition in the same way.
- When using ICP, one may read rows (e.g. call h->index_read_map(buf, ...)
only to buf= table->record[0], for two reasons:
* Pushed index condition's Item_field objects point into record[0]
* InnoDB requires this: it calls offset() which assumes record[0].
So, when using ICP, ha_partition will read partition records to
table->record[0] and then will copy record away if it needs it to be
elsewhere.
This includes all test changes from
"Changing all cost calculation to be given in milliseconds"
and forwards.
Some of the things that caused changes in the result files:
- As part of fixing tests, I added 'echo' to some comments to be able to
easier find out where things where wrong.
- MATERIALIZED has now a higher cost compared to X than before. Because
of this some MATERIALIZED types have changed to DEPENDEND SUBQUERY.
- Some test cases that required MATERIALIZED to repeat a bug was
changed by adding more rows to force MATERIALIZED to happen.
- 'Filtered' in SHOW EXPLAIN has in many case changed from 100.00 to
something smaller. This is because now filtered also takes into
account the smallest possible ref access and filters, even if they
where not used. Another reason for 'Filtered' being smaller is that
we now also take into account implicit filtering done for subqueries
using FIRSTMATCH.
(main.subselect_no_exists_to_in)
This is caluculated in best_access_path() and stored in records_out.
- Table orders has changed because more accurate costs.
- 'index' and 'ALL' for small tables has changed to use 'range' or
'ref' because of optimizer_scan_setup_cost.
- index can be changed to 'range' as 'range' optimizer assumes we don't
have to read the blocks from disk that range optimizer has already read.
This can be confusing in the case where there is no obvious where clause
but instead there is a hidden 'key_column > NULL' added by the optimizer.
(main.subselect_no_exists_to_in)
- Scan on primary clustered key does not report 'Using Index' anymore
(It's a table scan, not an index scan).
- For derived tables, the number of rows is now 100 instead of 2,
which can be seen in EXPLAIN.
- More tests have "Using index for group by" as the cost of this
optimization is now more correct (lower).
- A primary key could be preferred for a normal key, even if it would
access more rows, as it's faster to do 1 lokoup and 3 'index_next' on a
clustered primary key than one lookup trough a secondary.
(main.stat_tables_innodb)
Notes:
- There was a 4.7% more calls to best_extension_by_limited_search() in
the main.greedy_optimizer test. However examining the test results
it looked that the plans where slightly better (eq_ref where more
chained together) so I assume this is ok.
- I have verified a few test cases where there was notable/unexpected
changes in the plan and in all cases the new optimizer plans where
faster. (main.greedy_optimizer and some others)
- 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
This was to remove a performance regression between 10.3 and 10.4
In 10.5 we will have a better implementation of records_in_range
that will enable us to get more statistics.
This change was not done in 10.4 because the 10.5 will be part of
a larger change that is not suitable for the GA 10.4 version
Other things:
- Changed default handler block_size to 8192 to fix things statistics
for engines that doesn't set the block size.
- Fixed a bug in spider when using multiple part const ranges
(Patch from Kentoku)
join_cache_level=6+
The patch fixes two similar bugs in the commit 8eeb689e9f
that added multi_range_read support to partitions. The commit opened
a possibility to join a partition table using BKA+MRR. However in some
cases it could lead to wrong results or even crashes.
This could happened when
- index condition pushdown was used to join the table or
- the joined table was an inner table of an outer join and 'not exist'
optimization was applied or
- the join table was the inner table of a semi-join and the first match
optimization was applied
The bugs were in the code of the call-back functions
- partition_multi_range_key_skip_record() and
- partition_multi_range_key_skip_index_tuple().
Each of this function consist only of an invocation of another function.
Yet a wrong parameter was passed at this invocation.
The fix was suggested by Sergey Petrunia and it is apparently in line
with original design.
The corresponding comprehensive test cases demonstrating the problems
caused by the bugs were constructed by me.
Remove usage of deprecated variable storage_engine. It was deprecated in 5.5 but
it never issued a deprecation warning. Make it issue a warning in 10.5.1.
Replaced with default_storage_engine.
Fix partitioning and DS-MRR to work together
- In ha_partition::index_end(): take into account that ha_innobase (and
other engines using DS-MRR) will have inited=RND when initialized for
DS-MRR scan.
- In ha_partition::multi_range_read_next(): if the MRR scan is using
HA_MRR_NO_ASSOCIATION mode, it is not guaranteed that the partition's
handler will store anything into *range_info.
- In DsMrr_impl::choose_mrr_impl(): ha_partition will inquire partitions
about how much memory their MRR implementation needs by passing
*buffer_size=0. DS-MRR code didn't know about this (actually it used
uint for buffer size calculation and would have an under-flow).
Returning *buffer_size=0 made ha_partition assume that partitions do
not need MRR memory and pass the same buffer to each of them.
Now, this is fixed. If DS-MRR gets *buffer_size=0, it will return
the amount of buffer space needed, but not more than about
@@mrr_buffer_size.
* Fix ha_{innobase,maria,myisam}::clone. If ha_partition uses MRR on its
partitions, and partition use DS-MRR, the code will call handler->clone
with TABLE (*NOT partition*) name as an argument.
DS-MRR has no way of knowing the partition name, so the solution was
to have the ::clone() function for the affected storage engine to ignore
the name argument and get it elsewhere.
