The logic of the new code is:
- Log CREATE view to DDL log, with a marker if old view existed
- If old view exists (in case of CREATE or REPLACE view), make a copy
of the old view as view_name.frm-
- Create the new view definition file
- Delete copy of view if it was created.
Crash recovery:
- Delete view_name.frm~ file (Temporary file for view definition)
- If query was logged to binary log
- Delete copy of view if it exists
- else
-rename the copy of the view over the .frm file (restoring the
old definition)
One benefit of the new code is that CREATE OR REPLACE VIEW for an
existing view is no fully atomic: Either the view will be replaced or
the old one will be left unchanged.
There are a few different cases to consider
Logging of CREATE TABLE and CREATE TABLE ... LIKE
- If REPLACE is used and there was an existing table, DDL log the drop of
the table.
- If discovery of table is to be done
- DDL LOG create table
else
- DDL log create table (with engine type)
- create the table
- If table was created
- Log entry to binary log with xid
- Mark DDL log completed
Crash recovery:
- If query was in binary log do nothing and exit
- If discoverted table
- Delete the .frm file
-else
- Drop created table and frm file
- If table was dropped, write a DROP TABLE statement in binary log
CREATE TABLE ... SELECT required a little more work as when one is using
statement logging the query is written to the binary log before commit is
done.
This was fixed by adding a DROP TABLE to the binary log during crash
recovery if the ddl log entry was not closed. In this case the binary log
will contain:
CREATE TABLE xxx ... SELECT ....
DROP TABLE xxx;
Other things:
- Added debug_crash_here() functionality to Aria to be able to test
crash in create table between the creation of the .MAI and the .MAD files.
Description of how DROP DATABASE works after this patch
- Collect list of tables
- DDL log tables as they are dropped
- DDL log drop database
- Delete db.opt
- Delete data directory
- Log either DROP TABLE or DROP DATABASE to binary log
- De active ddl log entry
This is in line of how things where before (minus ddl logging) except that
we delete db.opt file last to not loose it if DROP DATABASE fails.
On recovery we have to ensure that all dropped tables are logged in
binary log and that they are properly dropped (as with atomic drop
table).
No new tables be dropped as part of recovery.
Recovery of active drop database ddl log entry:
- If drop database was logged to ddl log but was not found in the binary
log:
- drop the db.opt file and database directory.
- Log DROP DATABASE to binary log
- If drop database was not logged to ddl log
- Update binary log with DROP TABLE of the dropped tables. If table list
is longer than max_allowed_packet, then the query will be split into
multiple DROP TABLE/VIEW queries.
Other things:
- Added DDL_LOG_STATE and 'current database' as arguments to
mysql_rm_table_no_locks(). This was needed to be able to combine
ddl logging of DROP DATABASE and DROP TABLE and make the generated
DROP TABLE statements shorter.
- To make the DROP TABLE statement created by ddl log shorter, I changed
the binlogged query to use current directory and omit the directory
part for all tables in the current directory.
- Merged some DROP TABLE and DROP VIEW code in ddl logger. This was done
to be able get separate DROP VIEW and DROP TABLE statements in the binary
log.
- Added a 'recovery_state' variable to remember the state of dropped
tables and views.
- Moved out code that drops database objects (stored procedures) from
mysql_rm_db_internal() to drop_database_objects() for better code reuse.
- Made mysql_rm_db_internal() global so that could be used by the ddl
recovery code.
The purpose of this task is to ensure that DROP TRIGGER is atomic.
Description of how atomic drop trigger works:
Logging of DROP TRIGGER
Log the following information:
db
table name
trigger name
xid /* Used to check if query was already logged to binary log */
initial length of the .TRG file
query if there is space for it, if not log a zero length query.
Recovery operations:
- Delete if exists 'database/trigger_name.TRN~'
- If this file existed, it means that we crashed before the trigger
was deleted and there is nothing else to do.
- Get length of .TRG file
- If file length is unchanged, trigger was not dropped. Nothing else to
do.
- Log original query to binary, if it was stored in the ddl log. If it was
not stored (long query string), log the following query to binary log:
use `database` ; DROP TRIGGER IF EXISTS `trigger_name`
/* generated by ddl log */;
Other things:
- Added trigger name and DDL_LOG_STATE to drop_trigger()
Trigger name was added to make the interface more consistent and
more general.
Logging logic:
- Log tables, just before they are dropped, to the ddl log
- After the last table for the statement is dropped, log an xid for the
whole ddl log event
In case of crash:
- Remove first any active DROP TABLE events from the ddl log that matches
xids found in binary log (this mean the drop was successful and was
propery logged).
- Loop over all active DROP TABLE events
- Ensure that the table is completely dropped
- Write a DROP TABLE entry to the binary log with the dropped tables.
Other things:
- Added code to ha_drop_table() to be able to tell the difference if
a get_new_handler() failed because of out-of-memory or because the
handler refused/was not able to create a a handler. This was needed
to get sequences to work as sequences needs a share object to be passed
to get_new_handler()
- TC_LOG_BINLOG::recover() was changed to always collect Xid's from the
binary log and always call ddl_log_close_binlogged_events(). This was
needed to be able to collect DROP TABLE events with embedded Xid's
(used by ddl log).
- Added a new variable "$grep_script" to binlog filter to be able to find
only rows that matches a regexp.
- Had to adjust some test that changed because drop statements are a bit
larger in the binary log than before (as we have to store the xid)
Other things:
- MDEV-25588 Atomic DDL: Binlog query event written upon recovery is corrupt
fixed (in the original commit).
- Major rewrite of ddl_log.cc and ddl_log.h
- ddl_log.cc described in the beginning how the recovery works.
- ddl_log.log has unique signature and is dynamic. It's easy to
add more information to the header and other ddl blocks while still
being able to execute old ddl entries.
- IO_SIZE for ddl blocks is now dynamic. Can be changed without affecting
recovery of old logs.
- Code is more modular and is now usable outside of partition handling.
- Renamed log file to dll_recovery.log and added option --log-ddl-recovery
to allow one to specify the path & filename.
- Added ddl_log_entry_phase[], number of phases for each DDL action,
which allowed me to greatly simply set_global_from_ddl_log_entry()
- Changed how strings are stored in log entries, which allows us to
store much more information in a log entry.
- ddl log is now always created at start and deleted on normal shutdown.
This simplices things notable.
- Added probes debug_crash_here() and debug_simulate_error() to simply
crash testing and allow crash after a given number of times a probe
is executed. See comments in debug_sync.cc and rename_table.test for
how this can be used.
- Reverting failed table and view renames is done trough the ddl log.
This ensures that the ddl log is tested also outside of recovery.
- Added helper function 'handler::needs_lower_case_filenames()'
- Extend binary log with Q_XID events. ddl log handling is using this
to check if a ddl log entry was logged to the binary log (if yes,
it will be deleted from the log during ddl_log_close_binlogged_events()
- If a DDL entry fails 3 time, disable it. This is to ensure that if
we have a crash in ddl recovery code the server will not get stuck
in a forever crash-restart-crash loop.
mysqltest.cc changes:
- --die will now replace $variables with their values
- $error will contain the error of the last failed statement
storage engine changes:
- maria_rename() was changed to be more robust against crashes during
rename.
This is required to make Atomic RENAME TABLE work for these engines
The requirement is that if we have a server crash in the middle of a
storage engine rename call, the upcoming ddl log recovery should be able
to finalize it by re-execute the rename.
This happened because in ma_open() we did not take into account that
tran_man (Aria transaction manager) would not be initialized.
Fixed by using the same check for minimum transaction id as we use
during repair.
Other things:
- ariad_read_log now displays a readable timestamp
- Removed printing of datapage for header. This removes
some wrong warnings from the aria_read_log output
This did not server any real purpose and also made it too difficult to add
asserts for string memory overrwrites.
Moved all functionallity from Static_binary_string to Binary_string.
Other things:
- Added asserts to q_xxx and qs_xxx functions to check for memory overruns
- Fixed wrong test in String_buffer::set_buffer_if_not_allocated().
The idea is to reuse allocated buffers (to avoid extra allocs), which
the code did not do.
This change is to get rid of randomly failing tests, especially those
that reads random position of the binary log. From looking at the logs
it's clear that some failures is because of a read char (with value >= 128)
is converted to a big long value. Using uchar everywhere makes this much
less likely to happen.
Another benefit is that a lot of cast of char to uchar could be removed.
Other things:
- Removed some extra space before '=' and '+=' in assignments
- Fixed indentations and lines > 80 characters
- Replace '16' with 'element_size' (from class definition) in
Gtid_list_log_event()
The reason for the removal are:
- Generates more code
- Storing and retreving THD
- Causes extra code and daata to be generated to handle possible throw
exceptions (which never happens in MariaDB code)
- Uses more stack space
Other things:
- Changed convert_const_to_int() to use item->save_in_field_no_warnings(),
which made the code shorter and simpler.
- Removed not needed code in Sp_handler::sp_create_routine()
- Added thd as argument to store_key.copy() to make function simpler
- Added thd as argument to some subselect* constructor that inherites
from Item_subselect.
TO_CHAR(expr, fmt)
- expr: required parameter, data/time/timestamp type expression
- fmt: optional parameter, format string, supports
YYYY/YYY/YY/RRRR/RR/MM/MON/MONTH/MI/DD/DY/HH/HH12/HH24/SS and special
characters. The default value is "YYYY-MM-DD HH24:MI:SS"
In Oracle, TO_CHAR() can also be used to convert numbers to strings, but
this is not supported. This will gave an error in this patch.
Other things:
- If format strings is a constant, it's evaluated only once and if there
is any errors in it, they are given at once and the statement will abort.
Original author: woqutech
Lots of optimizations and cleanups done as part of review
The ROWNUM() function is for SELECT mapped to JOIN->accepted_rows, which is
incremented for each accepted rows.
For Filesort, update, insert, delete and load data, we map ROWNUM() to
internal variables incremented when the table is changed.
The connection between the row counter and Item_func_rownum is done
in sql_select.cc::fix_items_after_optimize() and
sql_insert.cc::fix_rownum_pointers()
When ROWNUM() is used anywhere in query, the optimization to ignore ORDER
BY in sub queries are disabled. This was done to get the following common
Oracle query to work:
select * from (select * from t1 order by a desc) as t where rownum() <= 2;
MDEV-3926 "Wrong result with GROUP BY ... WITH ROLLUP" contains a discussion
about this topic.
LIMIT optimization is enabled when in a top level WHERE clause comparing
ROWNUM() with a numerical constant using any of the following expressions:
- ROWNUM() < #
- ROWNUM() <= #
- ROWNUM() = 1
ROWNUM() can be also be the right argument to the comparison function.
LIMIT optimization is done in two cases:
- For the current sub query when the ROWNUM comparison is done on the top
level:
SELECT * from t1 WHERE rownum() <= 2 AND t1.a > 0
- For an inner sub query, when the upper level has only a ROWNUM comparison
in the WHERE clause:
SELECT * from (select * from t1) as t WHERE rownum() <= 2
In Oracle mode, one can also use ROWNUM without parentheses.
Other things:
- Fixed bug where the optimizer tries to optimize away sub queries
with RAND_TABLE_BIT set (non-deterministic queries). Now these
sub queries will not be converted to joins. This bug fix was also
needed to get rownum() working inside subqueries.
- In remove_const() remove setting simple_order to FALSE if ROLLUP is
USED. This code was disable a long time ago because of wrong assignment
in the following code. Instead we set simple_order to false if
RAND_TABLE_BIT was used in the SELECT list. This ensures that
we don't delete ORDER BY if the result set is not deterministic, like
in 'SELECT RAND() AS 'r' FROM t1 ORDER BY r';
- Updated parameters for Sort_param::init_for_filesort() to be able
to provide filesort with information where the number of accepted
rows should be stored
- Reordered fields in class Filesort to optimize storage layout
- Added new error messsage to tell that a function can't be used in HAVING
- Added field 'with_rownum' to THD to mark that ROWNUM() is used in the
query.
Co-author: Oleksandr Byelkin <sanja@mariadb.com>
LIMIT optimization for sub query
DROP TABLE opens all temporary tables at start, but then
uses find_temporary_table() to check if a table is temporary
instead of is_temporary_table() which is much faster.
This patch fixes this issue.
- Moved out creating StringBuffers in loops and instead create them
outside and just reset the buffer if it was not allocated (to avoid
a possible malloc/free for every entry)
Other things related to set_buffer_if_not_allocated()
- Changed Valuebuffer to not call set_buffer_if_not_allocated() when
it is created.
- Fixed geometry functions to reset string length before calling
String::reserve(). This is because one should not access length()
of an undefined.
- Added Item_func_conv_charset::save_in_field() as the item is using
str_value to store cached values, which conflicts with
Item::save_str_in_field().
- Changed Item_proc_string to not store the string value in sql_string
as this clashes with Item::save_str_in_field().
- Locally store value of full_name_cstring() in analyse::end_of_records()
as Item::save_str_in_field() may overwrite it.
- Marked some strings as set_thread_specific()
- Added String::free_buffer() to be used internally in String functions
to just free the buffer but not reset other String values.
- Fixed uses_buffer_owned_by() to check for allocated length instead of
strlength, which could be marked MEM_UNDEFINED().
This change removed 68 explict strlen() calls from the code.
The following renames was done to ensure we don't use the old names
when merging code from earlier releases, as using the new variables
for print function could result in crashes:
- charset->csname renamed to charset->cs_name
- charset->name renamed to charset->coll_name
Almost everything where mechanical changes except:
- Changed to use the new Protocol::store(LEX_CSTRING..) when possible
- Changed to use field->store(LEX_CSTRING*, CHARSET_INFO*) when possible
- Changed to use String->append(LEX_CSTRING&) when possible
Other things:
- There where compiler issues with ensuring that all character set names
points to the same string: gcc doesn't allow one to use integer constants
when defining global structures (constant char * pointers works fine).
To get around this, I declared defines for each character set name
length.
Changes:
- To detect automatic strlen() I removed the methods in String that
uses 'const char *' without a length:
- String::append(const char*)
- Binary_string(const char *str)
- String(const char *str, CHARSET_INFO *cs)
- append_for_single_quote(const char *)
All usage of append(const char*) is changed to either use
String::append(char), String::append(const char*, size_t length) or
String::append(LEX_CSTRING)
- Added STRING_WITH_LEN() around constant string arguments to
String::append()
- Added overflow argument to escape_string_for_mysql() and
escape_quotes_for_mysql() instead of returning (size_t) -1 on overflow.
This was needed as most usage of the above functions never tested the
result for -1 and would have given wrong results or crashes in case
of overflows.
- Added Item_func_or_sum::func_name_cstring(), which returns LEX_CSTRING.
Changed all Item_func::func_name()'s to func_name_cstring()'s.
The old Item_func_or_sum::func_name() is now an inline function that
returns func_name_cstring().str.
- Changed Item::mode_name() and Item::func_name_ext() to return
LEX_CSTRING.
- Changed for some functions the name argument from const char * to
to const LEX_CSTRING &:
- Item::Item_func_fix_attributes()
- Item::check_type_...()
- Type_std_attributes::agg_item_collations()
- Type_std_attributes::agg_item_set_converter()
- Type_std_attributes::agg_arg_charsets...()
- Type_handler_hybrid_field_type::aggregate_for_result()
- Type_handler_geometry::check_type_geom_or_binary()
- Type_handler::Item_func_or_sum_illegal_param()
- Predicant_to_list_comparator::add_value_skip_null()
- Predicant_to_list_comparator::add_value()
- cmp_item_row::prepare_comparators()
- cmp_item_row::aggregate_row_elements_for_comparison()
- Cursor_ref::print_func()
- Removes String_space() as it was only used in one cases and that
could be simplified to not use String_space(), thanks to the fixed
my_vsnprintf().
- Added some const LEX_CSTRING's for common strings:
- NULL_clex_str, DATA_clex_str, INDEX_clex_str.
- Changed primary_key_name to a LEX_CSTRING
- Renamed String::set_quick() to String::set_buffer_if_not_allocated() to
clarify what the function really does.
- Rename of protocol function:
bool store(const char *from, CHARSET_INFO *cs) to
bool store_string_or_null(const char *from, CHARSET_INFO *cs).
This was done to both clarify the difference between this 'store' function
and also to make it easier to find unoptimal usage of store() calls.
- Added Protocol::store(const LEX_CSTRING*, CHARSET_INFO*)
- Changed some 'const char*' arrays to instead be of type LEX_CSTRING.
- class Item_func_units now used LEX_CSTRING for name.
Other things:
- Fixed a bug in mysql.cc:construct_prompt() where a wrong escape character
in the prompt would cause some part of the prompt to be duplicated.
- Fixed a lot of instances where the length of the argument to
append is known or easily obtain but was not used.
- Removed some not needed 'virtual' definition for functions that was
inherited from the parent. I added override to these.
- Fixed Ordered_key::print() to preallocate needed buffer. Old code could
case memory overruns.
- Simplified some loops when adding char * to a String with delimiters.
This patch adds support for right aligned strings and numbers.
Left alignment is left as an exercise for anyone needing it.
MDEV-25612 "Assertion `to <= end' failed in process_args" fixed.
(Was caused by the original version of this patch)
The name change was to make the intention of the flag more clear and
also because most usage of the old flag was to test for
NOT IS_AUTOGENERATED_NAME.
Note that the new flag is the inverse of the old one!
This was done to simplify copying of with_* flags
Other things:
- Changed Flags to C++ enums, which enables gdb to print
out bit values for the flags. This also enables compiler
errors if one tries to manipulate a non existing bit in
a variable.
- Added set_maybe_null() as a shortcut as setting the
MAYBE_NULL flags was used in a LOT of places.
- Renamed PARAM flag to SP_VAR to ensure it's not confused with persistent
statement parameters.
One should instead use Item::fixed() and Item::with_subquery()
Removed Item::is_fixed() and has_subquery() and did the following replace:
replace is_fixed() fixed() -- *.*
replace 'has_subquery()' 'with_subquery()' -- *.*
- Added THD argument to functions that calls current_thd() or
new without a mem_root argument:
make_same(), set_comparator_func(), set_cmp_func(), set_cmp_func*(),
set_aggregator() and prepare_sum_aggregators()
- Changed "new Class" to "new (thd->mem_root) Class"
Almost all changes mechanical, no logic changes.
The reason for the change is that neither clang or gcc can do efficient
code when several bit fields are change at the same time or when copying
one or more bits between identical bit fields.
Updated bits explicitely with & and | is MUCH more efficient than what
current compilers can do.
Added back variable 'with_subquery' to Item class as a bit field.
This made the code shorter, faster (removed some virtual methods,
less code to create an initialized item etc) and made many Item's 7 bytes
smaller.
This is the last set of my patches the decreases the size of Item.
Some examples from gdb:
sizeof(Item): 144 -> 120
sizeof(Item_func) 208 -> 184
sizeof(Item_sum_max) 368 -> 344
Added back variable 'with_sum_func' to Item class as a bit field.
This made the code shorter, faster (removed some virtual methods,
less code to create an initialized item etc) and made many Item's 7 bytes
smaller.
The code is also easier to understand as 'with_sum_func' is threated as any
other Item variable when creating or copying items.
