------------------------------------------------------------------------
r4994 | marko | 2009-05-14 15:04:55 +0300 (Thu, 14 May 2009) | 18 lines
branches/5.1: Prevent a race condition in innobase_commit() by ensuring
that innodb_commit_concurrency>0 remains constant at run time. (Bug #42101)
srv_commit_concurrency: Make this a static variable in ha_innodb.cc.
innobase_commit_concurrency_validate(): Check that innodb_commit_concurrency
is not changed from or to 0 at run time. This is needed, because
innobase_commit() assumes that innodb_commit_concurrency>0 remains constant.
Without this limitation, the checks for innodb_commit_concurrency>0
in innobase_commit() should be removed and that function would have to
acquire and release commit_cond_m at least twice per invocation.
Normally, innodb_commit_concurrency=0, and introducing the mutex operations
would mean significant overhead.
innodb_bug42101.test, innodb_bug42101-nonzero.test: Test cases.
rb://123 approved by Heikki Tuuri
------------------------------------------------------------------------
This addresses Mantis Issue #23 in InnoDB Hot Backup and some
of MySQL Bug #41609.
In MLOG_FILE_CREATE, we need to distinguish temporary tables, so that
InnoDB Hot Backup can work correctly. It turns out that we can do this
easily, by using a bit of the previously unused parameter for page number.
(The page number parameter of MLOG_FILE_CREATE has been written as 0
ever since MySQL 4.1, which introduced MLOG_FILE_CREATE.)
MLOG_FILE_FLAG_TEMP: A flag for indicating a temporary table in
the page number parameter of MLOG_FILE_ operations.
fil_op_write_log(): Add the parameter log_flags.
fil_op_log_parse_or_replay(): Add the parameter log_flags.
Do not replay MLOG_FILE_CREATE when MLOG_FILE_FLAG_TEMP is set in log_flags.
This only affects ibbackup --apply-log. InnoDB itself never replays file
operations.
rb://117 approved by Heikki Tuuri
and undo_no_t. Each type is still defined as dulint.
This is an initial step towards replacing dulint with a 64-bit data type.
Because modern compilers have no trouble supporting 64-bit arithmetics
even on 32-bit targets, the dulint struct is a relic that should go.
The last remaining major use of dulint is dictionary IDs
(table, index, and row ids).
rb://114 approved by Sunny Bains
access
Mapping the atomic operations to Windows Interlocked functions:
os_compare_and_swap_* to InterlockedCompareExchange(64)
os_atomic_increment_* to InterlockedExchangeAdd(64)
os_atomic_test_and_set_byte to InterlockedExchange
In this patch, the legacy code under UNIV_CAN_USE_X86_ASSEMBLER is
removed all together, and add HAVE_WINDOWS_ATOMICS and
INNODB_RW_LOCKS_USE_ATOMICS to CMakeLists.txt
This is to address mantis issue#194.
rb://113
Approved by: Marko
dict_ind_redundant, dict_ind_compact: Declare these UNIV_INTERN.
innodb_hton_ptr: Declare static. We do not attempt to access the
built-in InnoDB any more.
trx_roll_savepoints_free(): Declare UNIV_INTERN.
Enable atomics on solaris (using the libc functions as defined in
atomic.h) if GCC atomic builtins are not present.
There still remains some work to be done (by Vasil?). This patch
makes changes to plug.in to check pthread_t size and presence of
atomic functions when running on solaris. The same has to become
a part of the generated Makefile.in when we bake our source.
Reviewed by: Heikki rb://106
history. This addresses Mantis issue #116.
dict_index_t: Enable the storage of trx_id.
row_prebuilt_t: Make many fields bit-fields to reduce the memory
footprint. Add index_usable.
ha_innobase::change_active_index(): Check if the index is usable and
set prebuilt->index_usable accordingly. Unfortunately, the return
status of this function is ignored by MySQL, and the actual refusal to
use the index must be made in row_search_for_mysql().
row_search_for_mysql(): Return DB_MISSING_HISTORY if
!prebuilt->index_usable.
convert_error_code_to_mysql(): Map DB_MISSING_HISTORY to
HA_ERR_TABLE_DEF_CHANGED.
innodb-index.test: Add a test case where access to a newly created
secondary index must be blocked for old transactions.
rb://100 approved by Heikki Tuuri
SHOW ENGINE INNODB MUTEX shows all mutexes and rw_locks. This can
be overwhelming particularly when the buffer pool is very large
(note that each block in buffer pool has at least one mutex, one
rw_lock and an additional mutex if rw_lock does not use atomics).
With this patch status of following mutexes and rw-locks is not shown:
1) block->mutex
2) block->lock
3) block->lock->mutex (if applicable)
4) All other mutexes and rw-locks for which number of os-waits are zero
Addresses issue# 179 rb://99
Approved by: Marko
are only protected by buf_pool_mutex, not by the block mutex.
buf_page_release(): Remove the assertion on in_flush_list.
The function is only holding block->mutex, not buf_pool_mutex.
This was reported by Sunny. This was the only piece of code
that accessed in_flush_list while not holding buf_pool_mutex.
before dereferencing it. In this way, ut_list_node_313 will be
pointing to the last non-NULL list item at the time of the assertion
failure. (gcc-4.3.2 -O3 seems to optimize the common subexpressions
and make the variable NULL, though.)
------------------------------------------------------------------------
r4399 | marko | 2009-03-12 09:38:05 +0200 (Thu, 12 Mar 2009) | 2 lines
branches/5.1: row_sel_get_clust_rec_for_mysql(): Store the cursor position
also for unlock_row(). (Bug #39320)
------------------------------------------------------------------------
r4400 | marko | 2009-03-12 10:06:44 +0200 (Thu, 12 Mar 2009) | 5 lines
branches/5.1: Fix a bug in multi-table semi-consistent reads.
Remember the acquired record locks per table handle (row_prebuilt_t)
rather than per transaction (trx_t), so that unlock_row should successfully
unlock all non-matching rows in multi-table operations.
This deficiency was found while investigating Bug #39320.
------------------------------------------------------------------------
These were submitted as rb://94 and rb://96 and approved by Heikki Tuuri.
Implement a check whether pthread_t objects can be used by GCC atomic
builtin functions. This check is implemented in plug.in and defines the
macro HAVE_ATOMIC_PTHREAD_T. This macro is checked in univ.i and the
relevant part of the code enabled (the one that uses GCC atomics against
pthread_t objects).
In addition to this, the same program that is compiled as part of the
plug.in check is added in ut/ut0auxconf.c. In the InnoDB Plugin source
archives that are shipped to the users, a generated Makefile.in is added.
That Makefile.in will be modified to compile ut/ut0auxconf.c and define
the macro HAVE_ATOMIC_PTHREAD_T if the compilation succeeds. I.e.
Makefile.in will emulate the work that is done by plug.in. This is done in
order to make the check happen and HAVE_ATOMIC_PTHREAD_T eventually
defined without regenerating MySQL's ./configure from
./storage/innobase/plug.in. The point is not to ask users to install the
autotools and regenerate ./configure.
rb://95
Approved by: Marko
Add FreeBSD to the list of the operating systems that have
sizeof(pthread_t) == sizeof(void*) (i.e. word size).
On FreeBSD pthread_t is defined like:
/usr/include/sys/_pthreadtypes.h:
typedef struct pthread *pthread_t;
I did the following tests (per Inaam's recommendation):
a) appropriate version of GCC is available on that platform (4.1.2 or
higher for atomics to be available)
On FreeBSD 6.x the default compiler is 3.4.6, on FreeBSD 7.x the default
one is 4.2.1. One can always install the version of choice from the ports
collection. If gcc 3.x is used then HAVE_GCC_ATOMIC_BUILTINS will not be
defined and thus the change I am committing will make no difference.
b) find out if sizeof(pthread_t) == sizeof(long)
On 32 bit both are 4 bytes, on 64 bit both are 8 bytes.
c) find out the compiler generated platform define (e.g.: __aix, __sunos__
etc.)
The macro is __FreeBSD__.
d) patch univ.i with the appropriate platform define
e) build the mysql
f) ensure it is using atomic builtins (look at the err.log message at
system startup. It should say we are using atomics for both mutexes and
rw-locks)
g) do sanity testing (keeping in view the smp changes)
I ran the mysql-test suite. All tests pass.
This is a fix for issue#178. Instead of using UNIV_LINUX which is
defined through CFLAGS we use compiler generated define __linux__
that is effective for both .c and .cc files.
again, to finally put Issue #181 to rest.
Revert some parts of r4274. It is best not to call ut_malloc() before
srv_general_init().
mem_init(): Do not call ut_mem_init().
srv_general_init(): Initialize the memory subsystem in two phases:
first ut_mem_init(), then mem_init(). This is because os_sync_init()
and sync_init() depend on ut_mem_init() and mem_init() depends on
os_sync_init() or sync_init().
srv_parse_data_file_paths_and_sizes(),
srv_parse_log_group_home_dirs(): Remove the output parameters. Assign
to the global variables directly. Allocate memory with malloc()
instead of ut_malloc(), because these functions will be called before
srv_general_init().
srv_free_paths_and_sizes(): New function, for cleaning up after
srv_parse_data_file_paths_and_sizes() and
srv_parse_log_group_home_dirs().
rb://92 approved by Sunny Bains
outside Valgrind, with innodb_use_sys_malloc set to 0 and 1.
mem_init(): Invoke ut_mem_init() before mem_pool_create(), because
the latter one will invoke ut_malloc().
srv_general_init(): Do not initialize the memory subsystem (mem_init()).
innobase_init(): Initialize the memory subsystem (mem_init()) before
calling srv_parse_data_file_paths_and_sizes(), which needs ut_malloc().
Call ut_free_all_mem() in error handling to clean up after the mem_init().
ut_list_mutex would be used uninitialized when innodb_use_sys_malloc=1.
This fix addresses Issue #181.
ut_mem_block_list_init(): Rename to ut_mem_init() and make public.
ut_malloc_low(), ut_free_all_mem(): Add ut_a(ut_mem_block_list_inited).
mem_init(): Call ut_mem_init().
Unprotected updates to ut_total_allocated_memory in
os_mem_alloc_large() and os_mem_free_large(), called during
fast index creation, may corrupt the variable and cause assertion failures.
Also, add UNIV_MEM_ALLOC() and UNIV_MEM_FREE() instrumentation around
os_mem_alloc_large() and os_mem_free_large(), so that Valgrind can
detect more errors.
rb://90 approved by Heikki Tuuri. This addresses Issue #177.
sync_array_object_signalled(): Add a (void) cast to eliminate a gcc warning
about the return value of os_atomic_increment() being ignored.
rw_lock_create_func(): Properly indent the preprocessor directives.
rw_lock_x_lock_low(), rw_lock_x_lock_func_nowait(): Split lines correctly.
rw_lock_set_writer_id_and_recursion_flag(): Silence a Valgrind warning.
Do not mix statements and variable declarations.
* Remove old Innobase copyright lines from C source files
* Add a reference to the GPLv2 license as recommended by the lawyers
at Oracle Legal
[Step 22/28]
