When index_merge_sort_union is turned off only ror scans were considered for range
scans, which is wrong.
To fix the problem ensure both ror scans and non ror scans are considered for range
access
SQL_SELECT::check_quick() returns error status only
test_quick_select() returns -1. Fix error handling when lower frames
throw error, but it is ignored by test_quick_select(). Fix return
status for out-of-memory errors which are obviously must be processed
as error in upper frames.
'index_merge_sort_union=off'
When index_merge_sort_union is set to 'off' and index_merge_union is set
to 'on' then any evaluated index merge scan must consist only of ROR scans.
The cheapest out of such index merges must be chosen. This index merge
might not be the cheapest index merge.
This bug could manifest itself in a very rare cases when the optimizer
chose an execution plan by which a joined table was accessed by a table
scan and the optimizer was checking whether ranges checked for each record
could improve this plan. In such cases the optimizer evaluates range
conditions over a table that depend on other tables. For such conditions
the constructed SEL_ARG trees are marked as MAYBE_KEY. If a SEL_ARG object
constructed for a sargable condition marked as RANGE_KEY had the same
first key part as a MAYBE_KEY SEL_ARG object and the key_and() function
was called for this pair of SEL_ARG objects then an invalid SEL_ARG
object could be constructed that ultimately could lead to a crash before
the execution phase.
'index_merge_sort_union=off'
When index_merge_sort_union is set to 'off' and index_merge_union is set
to 'on' then any evaluated index merge scan must consist only of ROR scans.
The cheapest out of such index merges must be chosen. This index merge
might not be the cheapest index merge.
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 also get_sweep_read_cost() for HEAP tables.
- Added unlikely() to optimize for not having optimizer trace enabled
- Made THD::trace_started() inline
- Added 'if (trace_enabled())' around some potentially expensive code
(not many found)
- Added ASSERT's to ensure we don't call expensive optimizer trace calls
if optimizer trace is not enabled
- Added length to Json_writer functions to speed up buffer writes
when optimizer trace is enabled.
- Changed LEX_CSTRING argument handling to not send full struct to writer
function on_add_str() functions now trusts length arguments
This patch fixes the following defects/bugs.
1. If BKA[H] algorithm was used to join a table for which the optimizer
had decided to employ a rowid filter the filter actually was not built.
2. The patch for the bug MDEV-21356 that added the code canceling pushing
rowid filter into an engine for the table joined with employment of
BKA[H] and MRR was not quite correct for Innodb engine because this
cancellation was done after InnoDB code had already bound the the pushed
filter to internal InnoDB structures.
[Variant 2 of the fix: collect the attached conditions]
Problem:
make_join_select() has a section of code which starts with
"We plan to scan all rows. Check again if we should use an index."
the code in that section will [unnecessarily] re-run the range
optimizer using this condition:
condition_attached_to_current_table AND current_table's_ON_expr
Note that the original invocation of range optimizer in
make_join_statistics was done using the whole select's WHERE condition.
Taking the whole select's WHERE condition and using multiple-equalities
allowed the range optimizer to infer more range restrictions.
The fix:
- Do range optimization using a condition that is an AND of this table's
condition and all of the previous tables' conditions.
- Also, fix the range optimizer to prefer SEL_ARGs with type=KEY_RANGE
over SEL_ARGS with type=MAYBE_KEY, regardless of the key part.
Computing
key_and(
SEL_ARG(type=MAYBE_KEY key_part=1),
SEL_ARG(type=KEY_RANGE, key_part=2)
)
will now produce the SEL_ARG with type=KEY_RANGE.
In the function test_if_cheaper_ordering we make a decision if using an index is better than
using filesort for ordering. If we chose to do range access then in test_quick_select we
should make sure that cost for table scan is set to DBL_MAX so that it is not picked.
selectivity values fails
After having set the assertion that checks validity of selectivity values
returned by the function table_cond_selectivity() a test case from
order_by.tesst failed. The failure occurred because range optimizer could
return as an estimate of the cardinality of the ranges built for an index
a number exceeding the total number of records in the table.
The second bug is more subtle. It may happen when there are several
indexes with same prefix defined on the first joined table t accessed by
a constant ref access. In this case the range optimizer estimates the
number of accessed records of t for each usable index and these
estimates can be different. Only the first of these estimates is taken
into account when the selectivity of the ref access is calculated.
However the optimizer later can choose a different index that provides
a different estimate. The function table_condition_selectivity() could use
this estimate to discount the selectivity of the ref access. This could
lead to an selectivity value returned by this function that was greater
that 1.
- Fix the LooseScan code to support storage engines that return
HA_ERR_END_OF_FILE if the index scan goes out of provided range
bounds
- Add a DBUG_EXECUTE_IF("force_group_by",...) to allow a test to
force a LooseScan
- Adjust rocksdb.group_min_max test not to use features not present
in MariaDB 10.2 (e.g. optimizer_trace. In MariaDB 10.4 it's present
but it doesn't meet the assumptions that the test makes about it
- Adjust the test result file:
= MariaDB doesn't support "Enhanced Loose Scan" that FB/MySQL has
= MariaDB has different cost calculations.
Introduced a print_key_value function to makes sure that the trace prints data in readable format
for readable characters and the rest of the characters are printed as hexadecimal.
Changed the function append_range_all_keyparts to use sel_arg_range_seq_init / sel_arg_range_seq_next to produce ranges.
Also adjusted to print format for the ranges, now the ranges are printed as:
(keypart1_min, keypart2_min,..) OP (keypart1_name,keypart2_name, ..) OP (keypart1_max,keypart2_max, ..)
Also added more tests for range and index merge access for optimizer trace
we had the statistics tables in the FROM list of the select.
The statistics for tables are not read in such cases, so we need
to check this case separately.
With MAX_INDEXIES=64(default), key_map=Bitmap<64> is just a wrapper around
ulonglong and thus "trivial" (can be bzero-ed, or memcpy-ed, and stays
valid still)
With MAX_INDEXES=128, key_map = Bitmap<128> is not a "trivial" type
anymore. The implementation uses MY_BITMAP, and MY_BITMAP contains pointers
which make Bitmap invalid, when it is memcpy-ed/bzero-ed.
The problem in 10.4 is that there are many new key_map members, inside TABLE
or KEY, and those are often memcopied and bzeroed
The fix makes Bitmap "trivial", by inlining most of MY_BITMAP functionality.
pointers/heap allocations are not used anymore.
Do not attempt to set param->table->with_impossible_ranges if the
range optimizer is using pseudo-indexes (which is true when we are
computing EITS selectivity estimates or doing partition pruning).
The MDEV-17262 commit 26432e49d3
was skipped. In Galera 4, the implementation would seem to require
changes to the streaming replication.
In the tests archive.rnd_pos main.profiling, disable_ps_protocol
for SHOW STATUS and SHOW PROFILE commands until MDEV-18974
has been fixed.
There were two newly enabled warnings:
1. cast for a function pointers. Affected sql_analyse.h, mi_write.c
and ma_write.cc, mf_iocache-t.cc, mysqlbinlog.cc, encryption.cc, etc
2. memcpy/memset of nontrivial structures. Fixed as:
* the warning disabled for InnoDB
* TABLE, TABLE_SHARE, and TABLE_LIST got a new method reset() which
does the bzero(), which is safe for these classes, but any other
bzero() will still cause a warning
* Table_scope_and_contents_source_st uses `TABLE_LIST *` (trivial)
instead of `SQL_I_List<TABLE_LIST>` (not trivial) so it's safe to
bzero now.
* added casts in debug_sync.cc and sql_select.cc (for JOIN)
* move assignment method for MDL_request instead of memcpy()
* PARTIAL_INDEX_INTERSECT_INFO::init() instead of bzero()
* remove constructor from READ_RECORD() to make it trivial
* replace some memcpy() with c++ copy assignments
When the chosen execution plan accesses a join table employing a range
rowid filter a quick select to scan this range has to be built. This
quick select is built by a call of SQL_SELECT::test_quick_select().
At this call the function should allow to evaluate only single index
range scans. In order to be able to do this a new parameter was added
to this function.
This task involves the implementation for the optimizer trace.
This feature produces a trace for any SELECT/UPDATE/DELETE/,
which contains information about decisions taken by the optimizer during
the optimization phase (choice of table access method, various costs,
transformations, etc). This feature would help to tell why some decisions were
taken by the optimizer and why some were rejected.
Trace is session-local, controlled by the @@optimizer_trace variable.
To enable optimizer trace we need to write:
set @@optimizer_trace variable= 'enabled=on';
To display the trace one can run:
SELECT trace FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE;
This task also involves:
MDEV-18489: Limit the memory used by the optimizer trace
introduces a switch optimizer_trace_max_mem_size which limits
the memory used by the optimizer trace. This was implemented by
Sergei Petrunia.
This patch contains a full implementation of the optimization
that allows to use in-memory rowid / primary filters built for range
conditions over indexes. In many cases usage of such filters reduce
the number of disk seeks spent for fetching table rows.
In this implementation the choice of what possible filter to be applied
(if any) is made purely on cost-based considerations.
This implementation re-achitectured the partial implementation of
the feature pushed by Galina Shalygina in the commit
8d5a11122c.
Besides this patch contains a better implementation of the generic
handler function handler::multi_range_read_info_const() that
takes into account gaps between ranges when calculating the cost of
range index scans. It also contains some corrections of the
implementation of the handler function records_in_range() for MyISAM.
This patch supports the feature for InnoDB and MyISAM.
In the function QUICK_RANGE_SELECT::init_ror_merged_scan we create a seperate handler if the handler in
head->file cannot be reused. The flag free_file tells us if we have a seperate handler or not.
There are cases where you might create a handler and then there might be a failure(running ALTER)
and then we have to revert the handler back to the original one. The code does that
but it does not reset the flag 'free_file' in this case.
Also backported f2c418079d.
The problem here is EITS statistics does not calculate statistics for the partitions of the table.
So a temporary solution would be to not read EITS statistics for partitioned tables.
Also disabling reading of EITS for columns that participate in the partition list of a table.
In this case we were trying to access memory for key_parts which we did not
assign for a fields because it did not any EITS statistics.
The check if EITS statistics for a column is avaialable or not was missing.
main.derived_cond_pushdown: Move all 10.3 tests to the end,
trim trailing white space, and add an "End of 10.3 tests" marker.
Add --sorted_result to tests where the ordering is not deterministic.
main.win_percentile: Add --sorted_result to tests where the
ordering is no longer deterministic.
For index merge union[or sort union], the estimates are not taken into account while calculating the selectivity of
a condition. So instead of showing the estimates of the index merge union[or sort union], it shows estimates equal to
all the records of the table.
The fix for the issue is to include the selectivity of index merge
union[or sort union] while calculating the selectivity of a condition.
optimizer_use_condition_selectivity>=3
Selectivity analysis should be disabled for Geometrical columns
for the case like geometric_field= string_constant.
Currently for selectivity calculation we perform range analysis for a column even when we don't have any statistics(EITS).
This makes less sense but is used to catch contradiction for WHERE condition.
So the solution is to not perform range analysis for selectivity calculation for columns that do not have statistics.
and use_stat_tables= PREFERABLY
Currently the code that calculates selectivity for a table does not take into account the case when
we can have GROUP BY optimization (looses index scan).
This patch introduces support for the system variable eq_range_index_dive_limit
that existed in MySQL starting from 5.6. The variable sets a limit for
index dives into equality ranges. Index dives are performed by optimizer
to estimate the number of rows in range scans. Index dives usually provide
good estimate but they are pretty expensive. To estimate the number of rows
in equality ranges statistical data on indexes can be employed. Its usage gives
not so good estimates but it's cheap. So if the number of equality dives
required by an index scan exceeds the set limit no dives for equality
ranges are performed by the optimizer for this index.
As the new system variable is introduced in a stable version the default
value for it is set to a special value meaning there is no limit for the number
of index dives performed by the optimizer.
The patch partially uses the MySQL code for WL 5957
'Statistics-based Range optimization for many ranges'.
Now the boolean data type is preserved in hybrid functions and MIN/MAX,
so COALESCE(bool_expr,bool_expr) and MAX(bool_expr) are correctly
detected by JSON_OBJECT() as being boolean rather than numeric expressions.
- Removing tests of item->type() against INT_ITEM and replacing
them to calls of new method item->is_bool_literal().
- Changing constant conditions to use Item_bool() instead of Item_int().
Lots of changes:
* calculate the current history partition in ::external_lock(),
not in ::write_row() or ::update_row()
* remove dynamically collected per-partition row_end stats
* no full table scan in open_table_from_share to calculate these
stats, no manual MDL/thr_locks in open_table_from_share
* no shared stats in TABLE_SHARE = no mutexes or condition waits when
calculating current history partition
* always compare timestamps, don't convert them to MYSQL_TIME
(avoid DST ambiguity, and it's faster too)
* correct interval handling, 1 month = 1 month, not 30 * 24 * 3600 seconds
* save/restore first partition start time, and count intervals from there
* only allow to drop first partitions if INTERVAL
* when adding new history partitions, split the data in the last history
parition, if it was overflowed
* show partition boundaries in INFORMATION_SCHEMA.PARTITIONS
