Before this fix, two performance schema unit tests crashed on windows.
The problem was a missing initialization to PFS_atomics,
which caused the crash only for platform not compiled with native atomics.
This fix adds the missing initialization in the unit tests.
No production code was changed, this is a unit test bug only.
MAP 'REPAIR TABLE' TO RECREATE +ANALYZE FOR ENGINES NOT
SUPPORTING NATIVE REPAIR
Executing 'mysqlcheck --check-upgrade --auto-repair ...' will first issue
'CHECK TABLE FOR UPGRADE' for all tables in the database in order to check if the
tables are compatible with the current version of MySQL. Any tables that are
found incompatible are then upgraded using 'REPAIR TABLE'.
The problem was that some engines (e.g. InnoDB) do not support 'REPAIR TABLE'.
This caused any such tables to be left incompatible. As a result such tables were
not properly fixed by the mysql_upgrade tool.
This patch fixes the problem by first changing 'CHECK TABLE FOR UPGRADE' to return
a different error message if the engine does not support REPAIR. Instead of
"Table upgrade required. Please do "REPAIR TABLE ..." it will report
"Table rebuild required. Please do "ALTER TABLE ... FORCE ..."
Second, the patch changes mysqlcheck to do 'ALTER TABLE ... FORCE' instead of
'REPAIR TABLE' in these cases.
This patch also fixes 'ALTER TABLE ... FORCE' to actually rebuild the table.
This change should be reflected in the documentation. Before this patch,
'ALTER TABLE ... FORCE' was unused (See Bug#11746162)
Test case added to mysqlcheck.test
client/mysqlcheck.c:
Changed mysqlcheck to do 'ALTER TABLE ... FORCE' if
'CHECK TABLE FOR UPGRADE' reports ER_TABLE_NEEDS_REBUILD
and not ER_TABLE_NEEDS_UPGRADE.
mysql-test/r/mysqlcheck.result:
Added regression test.
mysql-test/std_data/bug47205.frm:
InnoDB 5.0 FRM which contains a varchar primary key using
utf8_general_ci. This is an incompatible FRM for 5.5.
mysql-test/t/mysqlcheck.test:
Added regression test.
sql/handler.h:
Added new HA_CAN_REPAIR flag.
sql/share/errmsg-utf8.txt:
Added new error message ER_TABLE_NEEDS_REBUILD
sql/sql_admin.cc:
Changed 'CHECK TABLE FOR UPDATE' to give ER_TABLE_NEEDS_REBUILD
instead of ER_TABLE_NEEDS_UPGRADE if the engine does not support
REPAIR (as indicated by the new HA_CAN_REPAIR flag).
sql/sql_lex.h:
Remove unused ALTER_FORCE flag.
sql/sql_yacc.yy:
Make sure ALTER TABLE ... FORCE recreates the table
by setting the ALTER_RECREATE flag as the ALTER_FORCE
flag was unused.
storage/archive/ha_archive.h:
Added new HA_CAN_REPAIR flag to Archive
storage/csv/ha_tina.h:
Added new HA_CAN_REPAIR flag to CSV
storage/federated/ha_federated.h:
Added new HA_CAN_REPAIR flag to Federated
storage/myisam/ha_myisam.cc:
Added new HA_CAN_REPAIR flag to MyISAM
Setting lowercase_table_names to 2 on Windows causing Foreign Key problems
This problem was exposed by the fix for Bug#55222. There was a codepath in dict0load.c,
dict_load_foreigns() that made sure the table name matched case sensitive in order to
load a referenced table into the dictionary as needed. If an engine is rebooted which
accesses a table with foreign keys, and lower_case_table_names=2, then the table with
foreign keys will get an error when it is changed (insert/updated/delete).
Once the referenced tables are loaded into the dictionary cache by a select statement
on those tables, the same change would succeed because the affected code path would
not get followed.
NON-PRIMARY UNIQUE INDEX USING INNODB
This patch adds the HA_INPLACE_ADD_UNIQUE_INDEX_NO_WRITE
capability flag to InnoDB, indicating that concurrent reads
can be allowed while non-primary unique indexes are created.
This is an follow-up to Bug #11751388 which enabled concurrent
reads when creating non-primary non-unique indexes.
Test case added to innodb_mysql_sync.test.
ordered data from archive tables
Archive was using wrong memory address to check if field
is NULL (after filesort, when reading record again).
mysql-test/r/archive.result:
A test case for BUG#11764339.
mysql-test/t/archive.test:
A test case for BUG#11764339.
storage/archive/ha_archive.cc:
Null bytes are restored to "record" buffer, which may
or may not be equal to record buffer for field. Check
null bits in "record" buffer, instead of Field::null_ptr.
and compressed tables
buf_LRU_drop_page_hash_for_tablespace(): after releasing and
reacquiring the buffer pool mutex, do not dereference any block
descriptor pointer that is not known to be a pointer to an
uncompressed page frame (type buf_block_t; state ==
BUF_BLOCK_FILE_PAGE). Also, defer the acquisition of the block_mutex
until it is needed.
buf_page_get_gen(): Add mode == BUF_GET_IF_IN_POOL_PEEK for
buffer-fixing a block without making it young in the LRU list.
buf_page_get_gen(), buf_page_init(), buf_LRU_block_remove_hashed_page():
Set bpage->state = BUF_BLOCK_ZIP_FREE before buf_buddy_free(bpage),
so that similar race conditions might be detected a little easier.
btr_search_drop_page_hash_when_freed(): Use BUF_GET_IF_IN_POOL_PEEK
when dropping the hash indexes.
rb://528 approved by Jimmy Yang
IN OVERFLOW
Do not assign the result of the difference to a signed variable and
checking whether it is negative afterwards because this limits the max diff
to 2G on 32 bit systems. E.g. "signed = 3.5G - 1G" would be negative and the
code would assume that 3.5G < 1G. Instead compare the two variables directly
and assign to unsigned only if we know that the result of the subtraction
will be positive.
Discussed with: Jimmy and Sunny (via IRC)
Bug #11766501: Multiple RBS break the get rseg with mininum trx_t::no code during purge
Bug# 59291 changes:
Main problem is that truncating the UNDO log at the completion of every
trx_purge() call is expensive as the number of rollback segments is increased.
We truncate after a configurable amount of pages. The innodb_purge_batch_size
parameter is used to control when InnoDB does the actual truncate. The truncate
is done once after 128 (or TRX_SYS_N_RSEGS iterations). In other words we
truncate after purge 128 * innodb_purge_batch_size. The smaller the batch
size the quicker we truncate.
Introduce a new parameter that allows how many rollback segments to use for
storing REDO information. This is really step 1 in allowing complete control
to the user over rollback space management.
New parameters:
i) innodb_rollback_segments = number of rollback_segments to use
(default is now 128) dynamic parameter, can be changed anytime.
Currently there is little benefit in changing it from the default.
Optimisations in the patch.
i. Change the O(n) behaviour of trx_rseg_get_on_id() to O(log n)
Backported from 5.6. Refactor some of the binary heap code.
Create a new include/ut0bh.ic file.
ii. Avoid truncating the rollback segments after every purge.
Related changes that were moved to a separate patch:
i. Purge should not do any flushing, only wait for space to be free so that
it only does purging of records unless it is held up by a long running
transaction that is preventing it from progressing.
ii. Give the purge thread preference over transactions when acquiring the
rseg->mutex during commit. This to avoid purge blocking unnecessarily
when getting the next rollback segment to purge.
Bug #11766501 changes:
Add the rseg to the min binary heap under the cover of the kernel mutex and
the binary heap mutex. This ensures the ordering of the min binary heap.
The two changes have to be committed together because they share the same
that fixes both issues.
rb://567 Approved by: Inaam Rana.
attempt to create spatial index on char > 31 bytes".
Attempt to create spatial index on char field with length
greater than 31 byte led to assertion failure on server
compiled with safemutex support.
The problem occurred in mi_create() function which was called
to create a new version of table being altered. This function
failed since it detected an attempt to create a spatial key
on non-binary column and tried to return an error.
On its error path it tried to unlock THR_LOCK_myisam mutex
which has not been not locked at this point. Indeed such an
incorrect behavior was caught by safemutex wrapper and caused
assertion failure.
This patch fixes the problem by ensuring that mi_create()
doesn't releases THR_LOCK_myisam mutex on error path if it was
not acquired.
mysql-test/r/gis.result:
Added test for bug @59888 "debug assertion when attempt to
create spatial index on char > 31 bytes".
mysql-test/t/gis.test:
Added test for bug @59888 "debug assertion when attempt to
create spatial index on char > 31 bytes".
storage/myisam/mi_create.c:
Changed mi_create() not to release THR_LOCK_myisam mutex on
error path if it was not acquired.
rw_lock_create_func(): Initialize lock->writer_thread, so that Valgrind
will not complain even when Valgrind instrumentation is not enabled.
Flag lock->writer_thread uninitialized, so that Valgrind can complain
when it is used uninitialized.
rw_lock_set_writer_id_and_recursion_flag(): Revert the bogus Valgrind
instrumentation that was pushed in the first attempt to fix this bug.
"set optimizer_switch to e or d causes invalid memory writes/valgrind warnings":
due to prefix support, the argument "e" was overwritten with its full value
"engine_condition_pushdown", which caused a buffer overrun.
This was wrong usage of find_type(); other wrong usages are fixed here too.
Please start reading with the comment of typelib.c.
client/mysqldump.c:
A bug: find_type() expects a bitmap as 3rd argument
(each bit is a flag controlling a behaviour of the function);
here it was instead passed the length of the string to search!
That could give random behaviour of find_type()
depending on the string.
We rather need to pass a correct flag to find_type().
The correct flag is FIND_TYPE_BASIC (0).
Flag 8 is not needed as buff cannot have a comma (see how buff is filled).
Flag 1 looks like a superfluous restriction.
Flag 4 is not user-friendly (why use
--compatible=2 rather than --compatible=mysql40 ?, and
we probably not commit to "2" always meaning "mysql40"
until the end of times).
include/mysql.h.pp:
This isn't a problematic API change as we go from char* to const char*:
existing code will run unchanged.
include/typelib.h:
named constants. Not an enum to not significantly change
the declaration of find_type() which would be an API change
(typelib.h is included in mysql.h).
mysql-test/r/mysqldump.result:
correct result (see the two requested modes in SQL_MODE)
mysql-test/suite/sys_vars/t/optimizer_switch_basic.test:
test for BUG#59894. The second SET used to crash.
mysql-test/t/mysqldump.test:
we had no test for multiple modes in --compatible, which is
supported according to --help
mysys/typelib.c:
Fix for BUG#59894. parse_name() is asked to match "e" with a row
of the TYPELIB (the TYPELIB lists permitted flags of optimizer_switch;
and comes from optimizer_switch_names[] of sys_vars.cc).
find_type() is capable of supporting prefixes, but if it is not
passed flag 2 in third argument, it will overwrite its first
argument (the string to search for) with the complete name,
here overwriting "e" with "engine_condition_pushdown". But
as this "e" was a buffer allocated in an Item, it was not big
enough to host the longer name, thus the crash.
We don't need to know the complete flag's name; the output used
from find_type() is just the flag's number (== function's return
code). So we can pass flag 2 to find_type() in parse_name().
After doing this fix and the other fixes in this patch, all usages
of find_type() were using flag 2; in most usages the string to search for,
is not guaranteed to be long enough to host the complete name
(it is either directly from argv, or from alloc_root/my_malloc
done in an earlier call).
Thus, flag 2 is here made implicit: callers need not pass it anymore,
it is always automatically turned on.
This allows to eliminate an oddity: parse_name() took a const char**,
and then removed "const" before calling find_type(), which could
theoretically modify the pointed data, thus lying on constness.
Last, constants for find_type() are now named.
sql-common/client.c:
Two bugs:
1) The enum was not in sync with the array (due to a bad porting of WL 1054;
the extra OPT_ values are about options present in 5.1 and deleted in 5.5);
added a compile_time_assert() to make sure this doesn't happen again
2) find_type() was writing past the end of opt_arg; as opt_arg was allocated
with alloc_root() with no extra space, this was an overrun; it could be seen
when
** building with -DWITH_VALGRIND -DHAVE_purify -DEXTRA_DEBUG
** making execution go through the faulty code; this faulty
code is executed only if the client asks to read a configuration
file like this:
mysql_options(mysql, MYSQL_READ_DEFAULT_FILE, "/tmp/cnf.cnf");
so by adding such line to the start of mysql_client_test.c::client_connect(),
we could see the valgrind warning:
==30548== Invalid write of size 1
==30548== at 0x4C2624C: strcpy (mc_replace_strmem.c:303)
==30548== by 0x48DC29: find_type (typelib.c:120)
==30548== by 0x465686: mysql_read_default_options (client.c:1344)
==30548== by 0x46830F: mysql_real_connect (client.c:2971)
==30548== by 0x409339: client_connect (mysql_client_test.c:331)
==30548== by 0x463A7F: main (mysql_client_test.c:19902)
==30548== Address 0x61875ad is 0 bytes after a block of size 29 alloc'd
==30548== at 0x4C25153: malloc (vg_replace_malloc.c:195)
==30548== by 0x49BFF1: my_malloc (my_malloc.c:38)
==30548== by 0x49C65C: alloc_root (my_alloc.c:166)
==30548== by 0x48EF97: handle_default_option (default.c:381)
==30548== by 0x49068C: search_default_file_with_ext (default.c:992)
==30548== by 0x48F929: search_default_file (default.c:670)
==30548== by 0x48EDC4: my_search_option_files (default.c:312)
==30548== by 0x48F4B1: my_load_defaults (default.c:576)
==30548== by 0x46517A: mysql_read_default_options (client.c:1207)
==30548== by 0x46830F: mysql_real_connect (client.c:2971)
==30548== by 0x409339: client_connect (mysql_client_test.c:331)
==30548== by 0x463A7F: main (mysql_client_test.c:19902)
This is fixed by having find_type() not overwrite anymore.
sql/sql_help.cc:
cast not needed anymore.
sql/table.cc:
cast not needed anymore.
by silencing a bogus Valgrind warning:
==4392== Conditional jump or move depends on uninitialised value(s)
==4392== at 0x5A18416: rw_lock_set_writer_id_and_recursion_flag (sync0rw.ic:283)
==4392== by 0x5A1865C: rw_lock_x_lock_low (sync0rw.c:558)
==4392== by 0x5A18481: rw_lock_x_lock_func (sync0rw.c:617)
==4392== by 0x597EEE6: mtr_x_lock_func (mtr0mtr.ic:271)
==4392== by 0x597EBBD: fsp_header_init (fsp0fsp.c:970)
==4392== by 0x5A15E78: innobase_start_or_create_for_mysql (srv0start.c:1508)
==4392== by 0x598B789: innobase_init(void*) (ha_innodb.cc:2282)
os_compare_and_swap_thread_id() is defined as
__sync_bool_compare_and_swap(). From the GCC doc:
`bool __sync_bool_compare_and_swap (TYPE *ptr, TYPE oldval TYPE newval, ...)'
...
The "bool" version returns true if the comparison is successful and
NEWVAL was written.
So it is not possible that the return value is uninitialized, no matter what
the arguments to os_compare_and_swap_thread_id() are. Probably Valgrind gets
confused by the implementation of the GCC internal function
__sync_bool_compare_and_swap().
