- after a change in 5.6.20, tests with big blobs require bigger
innodb_log_file_size value;
- innobase relay was running with the default InnoDB (XtraDB)
The problem was that my_hash_sort didn't properly delete end-space characters properly, so strings that should compare
identically was seen as different strings. (Space was handled correctly, but not NBSP)
This caused duplicate key errors when a heap table was converted to Aria as part of overflow in group by.
Fixed by removing all characters that compares as end space when creating a hash.
Other things:
- Fixed that --sorted_results also works for errors in mysqltest.
- Speed up hash by not comparing strings that has different hash.
- Speed up many my_hash_sort functions by using registers to calculate hash instead of pointers.
This was previously done for some functions, but not for all.
- Made a macro of the hash function, to simplify code and to be able to experiment with new hash functions.
client/mysqltest.cc:
Fixed that --sorted_results also works for error messages.
mysql-test/r/ctype_partitions.result:
New test to ensure that partitions on hash works
mysql-test/suite/multi_source/gtid.result:
Updated result
mysql-test/suite/multi_source/gtid.test:
Test that --sorted_result works for error messages
mysql-test/suite/multi_source/gtid_ignore_duplicates.result:
Updated result
mysql-test/suite/multi_source/gtid_ignore_duplicates.test:
Updated result
mysql-test/suite/multi_source/load_data.result:
Updated result
mysql-test/suite/multi_source/load_data.test:
Updated result
mysql-test/t/ctype_partitions.test:
New test to ensure that partitions on hash works
storage/heap/hp_write.c:
Speed up hash by not comparing strings that has different hash.
storage/maria/ma_check.c:
Extra debug
strings/ctype-bin.c:
Use macro for hash function
strings/ctype-latin1.c:
Use macro for hash function
Use registers to calculate hash (speedup)
strings/ctype-mb.c:
Use macro for hash function
Use registers to calculate hash (speedup)
strings/ctype-simple.c:
Use macro for hash function
Use same variable names as in other my_hash_sort functions.
Update my_hash_sort_simple() to properly remove end space (patch by Bar)
strings/ctype-uca.c:
Ignore duplicated space inside strings and end space in my_hash_sort_uca(). This fixed MDEV-6255
Use macro for hash function
Use registers to calculate hash (speedup)
strings/ctype-ucs2.c:
Use macro for hash function
Use registers to calculate hash (speedup)
strings/ctype-utf8.c:
Use macro for hash function
Use registers to calculate hash (speedup)
strings/strings_def.h:
Made a macro of the hash function, to simplify code and to be able to experiment with new hash functions.
file trx0trx.ic line 60
Problem was that trx might have not been started when we enter release
savepoint, this can happen when trx with savepoint is already aborted
or when we try to release non-existing savepoint.
* handler::get_auto_increment() was not expecting any errors from the storage engine.
That was wrong, errors could happen.
* ha_partition::get_auto_increment() was doing index lookups in partition under a mutex.
This was redundant (engine transaction isolation was covering that anyway)
and harmful (causing deadlocks).
that uninitialised characters be written in op file.
modified:
storage/connect/tabdos.cpp
- Typo
modified:
storage/connect/filamap.cpp
storage/connect/filamdbf.cpp
storage/connect/filamdbf.h
storage/connect/valblk.cpp
storage/connect/xindex.cpp
MDEV-6483 - Deadlock around rw_lock_debug_mutex on PPC64
This problem affects only debug builds on PPC64.
There are at least two race conditions around
rw_lock_debug_mutex_enter and rw_lock_debug_mutex_exit:
- rw_lock_debug_waiters was loaded/stored without setting
appropriate locks/memory barriers.
- there is a gap between calls to os_event_reset() and
os_event_wait() and in such case we're supposed to pass
return value of the former to the latter.
Fixed by replacing self-cooked spinlocks with system mutexes.
These days system mutexes offer much better performance. OTOH
performance is not that critical for debug builds.
MDEV-6450 - MariaDB crash on Power8 when built with advance tool
chain
InnoDB mutex_exit() function calls __sync_test_and_set() to release
the lock. According to manual this function is supposed to create
"acquire" memory barrier whereas in fact we need "release" memory
barrier at mutex_exit().
The problem isn't repeatable with gcc because it creates
"acquire-release" memory barrier for __sync_test_and_set().
ATC creates just "acquire" barrier.
Fixed by creating proper barrier at mutex_exit() by using
__sync_lock_release() instead of __sync_test_and_set().
tables. This caused a CR character to be left in the line and in
particular caused the updelx test to fail on Windows.
modified:
storage/connect/filamtxt.cpp
while returning an error on Linux. Now accepted on linux.
modified:
storage/connect/maputil.cpp
- Fix a BUG in the XHUGE class. lseek64 was wrongly regarded as in error
when returning 0 instead of -1. This produced wrong index files.
modified:
storage/connect/filamfix.cpp
storage/connect/maputil.cpp
storage/connect/xindex.cpp
- Fix length mismatch (tab instead of blanks?)
modified:
storage/connect/mysql-test/connect/r/updelx.result
(while doing indexed UPDATE/DELETE)
modified:
storage/connect/array.cpp
storage/connect/filamtxt.cpp
- Force unix like line endings
modified:
storage/connect/tabvct.h
Analysis: When database is migrated from 5.5 or earlier and
database needs crash recovery, there is possibility that
SYS_DATAFILES system table does not exists, but
crash recovery in function dict_check_tablespaces_and_store_max_id()
assumes that SYS_DATAFILES exists.
Fix: If SYS_DATAFILES does not exists, create it before
we end up to function dict_check_tablespaces_and_store_max_id()
on crash recovery.
it belongs to each tables and the catalog being share between several
instances of CONNECT, when a query implied several tables belonging to
different databases, some where pointing on the wrong database. This
fix bugs occuring in queries such as:
INSERT into db1.t1 select * from db2.t2;
Where the t1 data file was made in db2.
modified:
storage/connect/catalog.h
storage/connect/connect.cc
storage/connect/filamdbf.cpp
storage/connect/filamdbf.h
storage/connect/ha_connect.cc
storage/connect/ha_connect.h
storage/connect/mycat.cc
storage/connect/mycat.h
storage/connect/plgdbsem.h
storage/connect/plgdbutl.cpp
storage/connect/reldef.cpp
storage/connect/reldef.h
storage/connect/tabfix.h
storage/connect/tabfmt.cpp
storage/connect/tabfmt.h
storage/connect/tabmul.cpp
a temporary file should be used for UPDATE/DELETE of file tables.
Also use the "sorted" argument of index_init to help decide if
sorting of positions must be done.
modified:
storage/connect/checklvl.h
storage/connect/connect.cc
storage/connect/connect.h
storage/connect/filamdbf.cpp
storage/connect/filamfix.cpp
storage/connect/filamfix.h
storage/connect/filamtxt.cpp
storage/connect/ha_connect.cc
storage/connect/mysql-test/connect/r/part_table.result
storage/connect/plgdbsem.h
storage/connect/plgdbutl.cpp
storage/connect/reldef.cpp
storage/connect/tabdos.cpp
storage/connect/tabdos.h
storage/connect/tabfix.cpp
storage/connect/tabfmt.cpp
storage/connect/tabvct.cpp
storage/connect/tabvct.h
storage/connect/xindex.cpp
- Fix a bug in TDBASE::ColDB that caused some special columns not to
be found in the column list and reallocated without their Value
causing a crash of some queries.
modified:
storage/connect/table.cpp
- Fix a bug causing RestoreNrec to be called before closing a table
causing a wrong value given to Spos
modified:
storage/connect/tabdos.cpp
storage/connect/xindex.cpp
- Add a new CONNECT global variable connect_exact_info. Set to ON, it
tells CONNECT to return exact record numbers on info queries. If OFF
it just gives an estimate. In version 10.0.13 this was unconditionally
ON and caused info queries on remote tables to be extremely long and
was the subject of MDEV-6612.
modified:
storage/connect/ha_connect.cc
storage/connect/tabdos.cpp
storage/connect/tabmysql.cpp
storage/connect/tabodbc.cpp
Part of this work is based on Stewart Smitch's memory barrier and lower priori
patches for power8.
- Added memory syncronization for innodb & xtradb for power8.
- Added HAVE_WINDOWS_MM_FENCE to CMakeList.txt
- Added os_isync to fix a syncronization problem on power
- Added log_get_lsn_nowait which is now used srv_error_monitor_thread to ensur
if log mutex is locked.
All changes done both for InnoDB and Xtradb
them sorted by file position. Firstly a new value is stored in indexes
to know if they are sorted, preventing to do the sorting when it is not
needed. Secondly, almost all in now done in connect instead of being
done by the different file access method classes. This pepares the future
use of temporary files for all table types and also fix the bug that was
occuring when partially using a multi-column index because of false MRR
like call of position followed by unsorted rnd_pos no more using indexing.
modified:
storage/connect/connect.cc
storage/connect/filamap.cpp
storage/connect/filamap.h
storage/connect/filamdbf.cpp
storage/connect/filamdbf.h
storage/connect/filamfix.cpp
storage/connect/filamfix.h
storage/connect/filamtxt.cpp
storage/connect/filamtxt.h
storage/connect/filamvct.cpp
storage/connect/filamvct.h
storage/connect/tabdos.cpp
storage/connect/tabdos.h
storage/connect/tabfix.h
storage/connect/tabfmt.cpp
storage/connect/tabfmt.h
storage/connect/xindex.cpp
storage/connect/xindex.h
storage/connect/xtable.h
