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.
always logged properly with binlog_row_image=MINIMAL
There are two issues fixed in this commit.
The first is an observation of a multi-table UPDATE binlogged
in row-format in binlog_row_image=MINIMAL mode. While the UPDATE aims
at a table with an ON-UPDATE attribute its binlog after-image misses
to record also installed default value.
The reason for that turns out missed marking of default-capable fields
in TABLE::write_set.
This is fixed to mark such fields similarly to 10.2's MDEV-10134 patch (db7edfed17)
that introduced it. The marking follows up 93d1e5ce0b841bed's idea
to exploit TABLE:rpl_write_set introduced there though,
and thus does not mess (in 10.1) with the actual MDEV-10134 agenda.
The patch makes formerly arg-less TABLE::mark_default_fields_for_write()
to accept an argument which would be TABLE:rpl_write_set.
The 2nd issue is extra columns in in binlog_row_image=MINIMAL before-image
while merely a packed primary key is enough. The test main.mysqlbinlog_row_minimal
always had a wrong result recorded.
This is fixed to invoke a function that intended for read_set
possible filtering and which is called (supposed to) in all type of MDL, UPDATE
including; the test results have gotten corrected.
At *merging* from 10.1->10.2 the 1st "main" part of the patch is unnecessary
since the bug is not observed in 10.2, so only hunks from
sql/sql_class.cc
are required.
The error message modified.
Then the TABLE_SHARE::error_table_name() implementation taken from 10.3,
to be used as a name of the table in this message.
Part of MDEV-5336 Implement LOCK FOR BACKUP
- Added new locks to MDL_BACKUP for all stages of backup locks and
a new MDL lock needed for backup stages.
- Renamed MDL_BACKUP_STMT to MDL_BACKUP_DDL
- flush_tables() takes a new parameter that decides what should be flushed.
- InnoDB, Aria (transactional tables with checksums), Blackhole, Federated
and Federatedx tables are marked to be safe for online backup. We are
using MDL_BACKUP_TRANS_DML instead of MDL_BACKUP_DML locks for these
which allows any DML's to proceed for these tables during the whole
backup process until BACKUP STAGE COMMIT which will block the final
commit.
- Re-numbered enum_table_category to make some tests easier
- Moved TABLE_CATEGORY_INFORMATION to be first CATEGORY of virtual tables
- Don't take MDL locks for not updateable table category's
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.
Race condition. field->flags were copied from s->field->flags during
field->clone(), early in open_table_from_share(). But s->field->flags
were getting their PART_INDIRECT_KEY_FLAG bit much later in
TABLE::mark_columns_used_by_virtual_fields() and only once per share.
If two threads were executing the code between field->clone()
and mark_columns_used_by_virtual_fields() at the same time, only
one would get PART_INDIRECT_KEY_FLAG bits in field[].
table for purge thread
Problem:
=======
Purge tries to fetch mdl lock for the whole table even though it tries
to open one of the partition. But table name length was wrongly set to indicate
the partition name too.
Solution:
========
- Table name length should identify the table name only not the partition name.
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.
ALTER TABLE locks the table with TL_READ_NO_INSERT, to prevent the
source table modifications while it's being copied. But there's an
indirect way of modifying a table, via cascade FK actions.
After previous commits, an attempt to modify an FK parent table
will cause FK children to be prelocked, so the table-being-altered
cannot be modified by a cascade FK action, because ALTER holds a
lock and prelocking will wait.
But if a new FK is being added by this very ALTER, then the target
table is not locked yet (it's a temporary table). So, we have to
lock FK parents explicitly.
The problem was that join_columns creation was not finished due to error of notfound column in USING, but next execution tried to use join_columns lists.
Solution is cleanup the lists on error. It can eat memory in statement MEM_ROOT but it is an error and error will be fixed or statement/procedure removed/altered.
The previous correction of the patch for mdev-16473 did not work
correctly for the databases whose names started with '*'.
Added a test case with a database named "*".
The bug was that innobase_get_computed_value() trashed record[0] and data
in Field_blob::value
Fixed by using a record on the heap for innobase_get_computed_value()
Reviewer: Marko Mäkelä
This is to mark that a field is indirectly part of a key, which simplifes
checking if we need to have this field up to date to evaluate a key.
For example:
CREATE TABLE t1 (a int, b int as (a) virtual,
c int as (b) virtual, index(c))
would mark a and b with PART_INDIRECT_KEY_FLAG.
c is marked with PART_KEY_FLAG as before.
The logic and the implementation scheme are similar with the
MDEV-9197 Pushdown conditions into non-mergeable views/derived tables
How the push down is made on the example:
select * from t1
where a>3 and b>10 and
(a,b) in (select x,max(y) from t2 group by x);
-->
select * from t1
where a>3 and b>10 and
(a,b) in (select x,max(y)
from t2
where x>3
group by x
having max(y)>10);
The implementation scheme:
1. Search for the condition cond that depends only on the fields
from the left part of the IN subquery (left_part)
2. Find fields F_group in the select of the right part of the
IN subquery (right_part) that are used in the GROUP BY
3. Extract from the cond condition cond_where that depends only on the
fields from the left_part that stay at the same places in the left_part
(have the same indexes) as the F_group fields in the projection of the
right_part
4. Transform cond_where so it can be pushed into the WHERE clause of the
right_part and delete cond_where from the cond
5. Transform cond so it can be pushed into the HAVING clause of the right_part
The optimization is made in the
Item_in_subselect::pushdown_cond_for_in_subquery() and is controlled by the
variable condition_pushdown_for_subquery.
New test file in_subq_cond_pushdown.test is created.
There are also some changes made for setup_jtbm_semi_joins().
Now it is decomposed into the 2 procedures: setup_degenerate_jtbm_semi_joins()
that is called before optimize_cond() for cond and setup_jtbm_semi_joins()
that is called after optimize_cond().
New setup_jtbm_semi_joins() is made in the way so that the result of its work is
the same as if it was called before optimize_cond().
The code that is common for pushdown into materialized derived and into materialized
IN subqueries is factored out into pushdown_cond_for_derived(),
Item_in_subselect::pushdown_cond_for_in_subquery() and
st_select_lex::pushdown_cond_into_where_clause().
MDEV-16100 FOR SYSTEM_TIME erroneously resolves string user variables as transaction IDs
Problem:
Vers_history_point::resolve_unit() tested item->result_type() before
item->fix_fields() was called.
- Item_func_get_user_var::result_type() returned REAL_RESULT by default.
This caused MDEV-16100.
- Item_func_sp::result_type() crashed on assert.
This caused MDEV-16094
Changes:
1. Adding item->fix_fields() into Vers_history_point::resolve_unit()
before using data type specific properties of the history point
expression.
2. Adding a new virtual method Type_handler::Vers_history_point_resolve_unit()
3. Implementing type-specific
Type_handler_xxx::Type_handler::Vers_history_point_resolve_unit()
in the way to:
a. resolve temporal and general purpose string types to TIMESTAMP
b. resolve BIT and general purpose INT types to TRANSACTION
c. disallow use of non-relevant data type expressions in FOR SYSTEM_TIME
Note, DOUBLE and DECIMAL data types are disallowed intentionally.
- DOUBLE does not have enough precision to hold huge BIGINT UNSIGNED values
- DECIMAL rounds on conversion to INT
Both lack of precision and rounding might potentionally lead to
very unpredictable results when a wrong transaction ID would be chosen.
If one really wants dangerous use of DOUBLE and DECIMAL, explicit CAST
can be used:
FOR SYSTEM_TIME AS OF CAST(double_or_decimal AS UNSIGNED)
QQ: perhaps DECIMAL(N,0) could still be allowed.
4. Adding a new virtual method Item::type_handler_for_system_time(),
to make HEX hybrids and bit literals work as TRANSACTION rather
than TIMESTAMP.
5. sql_yacc.yy: replacing the rule temporal_literal to "TIMESTAMP TEXT_STRING".
Other temporal literals now resolve to TIMESTAMP through the new
Type_handler methods. No special grammar needed. This removed
a few shift/resolve conflicts.
(TIMESTAMP related conflicts in "history_point:" will be removed separately)
6. Removing the "timestamp_only" parameter from
vers_select_conds_t::resolve_units() and Vers_history_point::resolve_unit().
It was a hint telling that a table did not have any TRANSACTION-aware
system time columns, so it's OK to resolve to TIMESTAMP in case of uncertainty.
In the new reduction it works as follows:
- the decision between TIMESTAMP and TRANSACTION is first made
based only on the expression data type only
- then, in case if the expression resolved to TRANSACTION, the table
is checked if TRANSACTION-aware columns really exist.
This way is safer against possible ALTER TABLE statements changing
ROW START and ROW END columns from "BIGINT UNSIGNED" to "TIMESTAMP(x)"
or the other way around.
Make sure that SELECT_LEX_UNIT::derived, behaves as documented
(points to the "TABLE_LIST representing this union in the
embedding select"). For recursive CTE this was not necessarily
the case, it could've pointed to the TABLE_LIST inside the CTE,
not in the embedding select.
To fix:
* don't update unit->derived in mysql_derived_prepare(), pass derived
as an argument to st_select_lex_unit::prepare()
* prefer to set unit->derived in TABLE_LIST::init_derived()
to the TABLE_LIST in the embedding select, not to the recursive
reference. Fail if there are many TABLE_LISTs in the embedding
select with conflicting FOR SYSTEM_TIME clauses.
cleanup:
* remove redundant THD* argument from st_select_lex_unit::prepare()
