assertion failure that was accidentally introduced in r4036.
Instead of calling buf_block_get_frame(), which asserts that the
block must be buffer-fixed, access block->frame directly. That
is safe, because changes of block->page.state are protected by
the buffer pool mutex, which we are holding.
This bug was reported by Michael.
within UNIV_DEBUG. The two remaining callers in non-debug builds,
btr_search_guess_on_hash() and btr_search_validate(), were rewritten
to call buf_page_hash_get().
To implement support for a resizeable buffer pool, the function
buf_block_align() had been rewritten to perform a page hash lookup in
the buffer pool. The caller was also made responsible for holding the
buffer pool mutex.
Because the page hash lookup is expensive and it has to be done while
holding the buffer pool mutex, implement buf_block_align() by pointer
arithmetics again, and make btr_search_guess_on_hash() call it. Note
that this will have to be adjusted if the interface to the resizeable
buffer pool is actually implemented.
rb://83 approved by Heikki Tuuri, to address Issue #161.
As a deviation from the approved patch, this patch also makes
btr_search_validate() (invoked by CHECK TABLE) check that
buf_pool->page_hash is consistent with buf_block_align().
This will help trim the dependencies of InnoDB Hot Backup.
recv_recovery_from_checkpoint_start(): Rename to
recv_recovery_from_checkpoint_start_func(), and remove the two first
parameters unless UNIV_LOG_ARCHIVE is defined. Define and use
the auxiliary macros TYPE_CHECKPOINT and LIMIT_LSN in the function.
struct recv_sys_struct: Remove archive_group unless UNIV_LOG_ARCHIVE
is defined.
Do not define LOG_ARCHIVE unless UNIV_LOG_ARCHIVE is defined.
------------------------------------------------------------------------
r4004 | marko | 2009-01-20 16:19:00 +0200 (Tue, 20 Jan 2009) | 12 lines
branches/5.1: Merge r4003 from branches/5.0:
rec_set_nth_field(): When the field already is SQL null,
do nothing when it is being changed to SQL null. (Bug #41571)
Normally, MySQL does not pass "do-nothing" updates to the storage engine.
When it does and a column of an InnoDB table that is in ROW_FORMAT=COMPACT
is being updated from NULL to NULL, the InnoDB buffer pool will be corrupted
without this fix.
rb://81 approved by Heikki Tuuri
------------------------------------------------------------------------
r4005 | marko | 2009-01-20 16:22:36 +0200 (Tue, 20 Jan 2009) | 8 lines
branches/5.1: lock_is_table_exclusive(): Acquire kernel_mutex before
accessing table->locks and release kernel_mutex before returning from
the function. This fixes a portential race condition in the
"commit every 10,000 rows" in ALTER TABLE, CREATE INDEX, DROP INDEX,
and OPTIMIZE TABLE. (Bug #42152)
rb://80 approved by Heikki Tuuri
------------------------------------------------------------------------
------------------------------------------------------------------------
r3911 | sunny | 2009-01-13 14:15:24 +0200 (Tue, 13 Jan 2009) | 13 lines
branches/5.1: Fix Bug#38187 Error 153 when creating savepoints
InnoDB previously treated savepoints as a stack e.g.,
SAVEPOINT a;
SAVEPOINT b;
SAVEPOINT c;
SAVEPOINT b; <- This would delete b and c.
This fix changes the behavior to:
SAVEPOINT a;
SAVEPOINT b;
SAVEPOINT c;
SAVEPOINT b; <- Does not delete savepoint c
------------------------------------------------------------------------
r3930 | marko | 2009-01-14 15:51:30 +0200 (Wed, 14 Jan 2009) | 4 lines
branches/5.1: dict_load_table(): If dict_load_indexes() fails,
invoke dict_table_remove_from_cache() instead of dict_mem_table_free(),
so that the data dictionary will not point to freed data.
(Bug #42075, Issue #153, rb://76 approved by Heikki Tuuri)
------------------------------------------------------------------------
satisfy some conditions when UNIV_DEBUG is defined.
HASH_SEARCH(): New parameter: ASSERTION. All users will pass an appropriate
ut_ad() or nothing.
dict_table_add_to_columns(): Assert that the table being added to the data
dictionary cache is not already being pointed to by the name_hash and
id_hash tables.
HASH_SEARCH_ALL(): New macro, for use in dict_table_add_to_columns().
dict_mem_table_free(): Set ut_d(table->cached = FALSE), so that we can
check ut_ad(table->cached) when traversing the hash tables, as in
HASH_SEARCH(name_hash, dict_sys->table_hash, ...) and
HASH_SEARCH(id_hash, dict_sys->table_id_hash, ...).
dict_table_get_low(), dict_table_get_on_id_low(): Assert
ut_ad(!table || table->cached).
fil_space_get_by_id(): Check ut_ad(space->magic_n == FIL_SPACE_MAGIC_N)
in HASH_SEARCH(hash, fil_system->spaces, ...).
fil_space_get_by_name(): Check ut_ad(space->magic_n == FIL_SPACE_MAGIC_N)
in HASH_SEARCH(name_hash, fil_system->name_hash, ...).
buf_buddy_block_free(): Check that the blocks are in valid state in
HASH_SEARCH(hash, buf_pool->zip_hash, ...).
buf_page_hash_get(): Check that the blocks are in valid state in
HASH_SEARCH(hash, buf_pool->page_hash, ...).
get_share(), free_share(): Check ut_ad(share->use_count > 0) in
HASH_SEARCH(table_name_hash, innobase_open_tables, ...).
This was posted as rb://75 for tracking down errors similar to Issue #153.
buf_page_t: Note that the LRU fields are protected by buf_pool_mutex
only, not block->mutex or buf_pool_zip_mutex.
buf_page_get_freed_page_clock(): Note that this is sometimes invoked
without mutex protection.
buf_pool_get_oldest_modification(): Note that the result may be out of
date.
buf_page_get_LRU_position(), buf_page_is_old(): Assert that the buffer
pool mutex is being held.
buf_page_release(): Assert that dirty blocks are in the flush list.
buf_page_struct: Note that space and offset are also protected by
buf_pool_mutex. They are only assigned to by
buf_block_set_file_page(). Thus, it suffices for buf_flush_batch() to
hold just buf_pool_mutex when checking these fields.
buf_flush_try_page(): Rename "locked" to "is_s_latched", per Heikki's request.
buf_flush_batch(): Move the common statement mutex_exit(block_mutex)
from all if-else if-else branches before the if block. Remove the
redundant test (buf_pool->init_flush[flush_type] == FALSE) that was
apparently copied from buf_flush_write_complete().
buf_flush_write_block_low(): Note why it is safe not to hold buf_pool_mutex
or block_mutex. Enumerate the assumptions in debug assertions.
buf_flush_insert_sorted_into_flush_list(): Remove unused code.
Change the parameter to buf_block_t* block and assert that
block->state == BUF_BLOCK_FILE_PAGE. This is part of Issue #155.
------------------------------------------------------------------------
r3588 | inaam | 2008-12-18 14:26:54 +0200 (Thu, 18 Dec 2008) | 8 lines
branches/5.1
It is a bug in unused code. If we don't calculate the hash value when
calculating the mutex number then two pages which map to same hash
value can get two different mutex numbers.
Approved by: Marko
------------------------------------------------------------------------
r3590 | marko | 2008-12-18 15:33:36 +0200 (Thu, 18 Dec 2008) | 11 lines
branches/5.1: When converting a record to MySQL format, copy the default
column values for columns that are SQL NULL. This addresses failures in
row-based replication (Bug #39648).
row_prebuilt_t: Add default_rec, for the default values of the columns in
MySQL format.
row_sel_store_mysql_rec(): Use prebuilt->default_rec instead of
padding columns.
rb://64 approved by Heikki Tuuri
------------------------------------------------------------------------
r3598 | marko | 2008-12-22 15:28:03 +0200 (Mon, 22 Dec 2008) | 6 lines
branches/5.1: ibuf_delete_rec(): When the record cannot be found and
the tablespace has been dropped, commit the mini-transaction, so that
InnoDB will not hold the insert buffer tree latch in exclusive mode,
causing a potential deadlock. This bug was introduced in the fix of
Bug #27276 in r2924.
------------------------------------------------------------------------
interpreter. It was only used for updating the InnoDB internal data
dictionary when renaming or dropping tables. It could have caused
deadlocks after acquiring latches on insert buffer bitmap pages.
This and r3544 should fix Issue #135.
Furthermore, the update-in-place-in-select does not account for
compression failure. That was not a problem yet, since the InnoDB SQL
interpreter has so far assumed ROW_FORMAT=REDUNDANT.
rb://63 approved by Heikki Tuuri
Originally, there were provisions in InnoDB for multiple insert buffer
B-trees, apparently one for each tablespace.
When Heikki implemented innodb_file_per_table (multiple InnoDB
tablespaces) in MySQL 4.1, he made the insert buffer live only in the
system tablespace (space 0) but left the provisions in the code.
When Osku Salerma implemented delete buffering, he also cleaned up the
insert buffer subsystem so that only one insert buffer B-tree exists.
This patch applies the clean-up to the InnoDB Plugin.
Having a separate patch of the insert buffer clean-up should help us
better compare the essential changes of the InnoDB Plugin and InnoDB+
and to track down bugs that are specific to InnoDB+.
IBUF_SPACE_ID: New constant, defined as 0.
ibuf_data_t: Remove.
ibuf_t: Add the applicable fields from ibuf_data_t. There is only one
insert buffer tree from now on.
ibuf_page_low(), ibuf_page(): Merge to a single function ibuf_page().
fil_space_t: Remove ibuf_data.
fil_space_get_ibuf_data(): Remove. There is only one ibuf_data, for
space IBUF_SPACE_ID.
fil_ibuf_init_at_db_start(): Remove.
ibuf_init_at_db_start(): Fuse with ibuf_data_init_for_space().
ibuf_validate_low(): Remove. There is only one ibuf tree.
ibuf_free_excess_pages(), ibuf_header_page_get(),
ibuf_free_excess_pages(): Remove the parameter space, which was always
0.
ibuf_tree_root_get(): Remove the parameters space and data. There is
only one ibuf tree, for space IBUF_SPACE_ID.
ibuf_data_sizes_update(): Rename to ibuf_size_update(), and remove the
parameter data. There is only one ibuf data struct.
ibuf_build_entry_pre_4_1_x(): New function, refactored from
ibuf_build_entry_from_ibuf_rec().
ibuf_data_enough_free_for_insert(), ibuf_data_too_much_free(): Remove
the parameter data. There is only one insert buffer tree.
ibuf_add_free_page(), ibuf_remove_free_page(): Remove the parameters
space and data. There is only one insert buffer tree.
ibuf_get_merge_page_nos(): Add parenthesis, to reduce diffs to
branches/innodb+.
ibuf_contract_ext(): Do not pick an insert buffer tree at random.
There is only one.
ibuf_print(): Print the single insert buffer tree.
rb://19 approved by Heikki on IM
Make the variables rw_latch and buf_mode local in the for loop.
Initialize them at the beginning of each for loop round to reduce
register spilling on register-starved platforms such as the x86. Move
the assignment of rw_latch and buf_mode from the end of the loop to
the beginning of the loop. These parameters will only be needed in
the buf_page_get_gen() call at the start of the loop.
Remove the second (redundant) call to ibuf_should_try().
ibuf_should_try(): Now that the successful calls to this function will
be halved, halve the magic constant that ibuf_flush_count will be
compared to, accordingly.
rb://61 approved by Heikki over IM.
innodb_buffer_pool_size >= 4096M on x64 Windows
All three srv_buf_pool related variables are defined as ulong, which is
32-bit on 64-bit Windows. They are changed to 64-bit ulint. Also
system_info.dwPageSize appears to be 32-bit only. Casting to 64-bit
is required.
Approved by: Marko (on IM)
Revert our temporary fix for Bug#40360 Binlog related errors with binlog off
This bug was fixed in MySQL code.
Our fix went into r2944 and r2947, but this patch does not entirely revert
those revisions because we want to leave the test case that was introduced
and also r2944 itself reverted r2935 and r2936. So if we completely revert
r2944 and r2947 then we would loose the test and will restore r2935 and
r2936.
This resolves Issue#108 We should remove ib_bin_log_is_engaged() once mysql
add an equivallent, see Bug#40360
an index or table. We have to skip this check when loading table definitions
from the data dictionary, because we could otherwise refuse to load old
tables (even uncompressed ones). This addresses Issue #119.
The first "row too large" check was implemented in MySQL 5.0.3
to address MySQL Bug #5682. In the InnoDB Plugin 1.0.2, a more
accurate check was implemented in innodb_strict_mode. We now
make the check unconditional.
dict_create_index_step(): Pass strict=TRUE to dict_index_add_to_cache().
trx_is_strict(), thd_is_strict(): Remove.
innodb-zip.test: Test in innodb_strict_mode=OFF.
innodb_bug36169.test: Ensure that none of the tables can be created.
rb://56 approved by Sunny Bains.
Revert our temporary fix for "Bug#40360 Binlog related errors with binlog off"
(r2935, r2936) and deploy MySQL's one, but put the function
mysql_bin_log_is_engaged() inside mysql_addons.cc instead of in mysql's log.cc
and use a different name for it so there is no collision when MySQL adds this
function in log.cc.
[note from the future: the windows part of this patch went into r2947]
Approved by: Marko (https://svn.innodb.com/rb/r/41/)
1. We add a vector of locks to trx_t. This array contains the autoinc
locks granted to a transaction. There is one per table.
2. We enforce releasing of these locks in the reverse order from the
one in which they are acquired. The assumption is that since the
AUTOINC locks are statement level locks. Nested statements introduced
by triggers are stacked it should hold.
There was some cleanup done to the vector code too by adding const and
some new functions. Rename dict_table_t::auto_inc_lock to autoinc_lock.
Fix Bug#26316 Triggers create duplicate entries on auto-increment columns
rb://22
except r2924, which was merged from branches/zip r2866 to branches/5.1
and except r2879 which was merged separately by Calvin:
------------------------------------------------------------------------
r2902 | vasil | 2008-10-28 12:10:25 +0200 (Tue, 28 Oct 2008) | 10 lines
branches/5.1:
Fix Bug#38189 innodb_stats_on_metadata missing
Make the variable innodb_stats_on_metadata visible to the users and
also settable at runtime. Previously it was only "visible" as a command
line startup option to mysqld.
Approved by: Marko (https://svn.innodb.com/rb/r/36)
------------------------------------------------------------------------
r2929 | marko | 2008-10-29 21:26:14 +0200 (Wed, 29 Oct 2008) | 13 lines
branches/5.1: dtype_get_sql_null_size(): return the correct storage
size of a SQL NULL column. (Bug #40369)
When MySQL Bug #20877 was fixed in r834, this function was
accidentally modified to return 0 or 1. Apparently, the only impact of
this bug is that fixed-length columns cannot be updated in-place from
or to SQL NULL, even in ROW_FORMAT=REDUNDANT. After this fix,
fixed-length columns in ROW_FORMAT=REDUNDANT will have a constant
storage size as they should, no matter if NULL or non-NULL. The bug
caused fixed-length NULL columns to occupy 1 byte.
rb://37 approved by Heikki over IM.
------------------------------------------------------------------------
trx_sys_file_format_max_upgrade(): Rename from
trx_sys_file_format_max_update(). Improve the documentation. Add a
const qualifier to the parameter "name". Replace the parameter
"flags" with "format_id", because this function should deal with file
format identifiers, not with table flags.
trx_sys_file_format_max_write(), trx_sys_file_format_max_set(): Add a
const qualifier to the parameter "name".
ha_innodb.cc: Correct the spelling in some comments: "side effect".
Remove redundant prototypes for some static callback functions.
innodb_file_format_name_update(), innodb_file_format_check_update():
Correct the function signature. Use appropriate pointer type conversions.
MYSQL_SYSVAR_STR(file_format), MYSQL_SYSVAR_STR(file_format_check):
Remove the type conversions from the callback function pointers. When
the function signatures match, no type conversion is needed. The type
conversions would only prevent compilation warnings for any mismatch.
Approved by Sunny in rb://25.
Merge revisions 2852:2854 from branches/5.1:
------------------------------------------------------------------------
r2854 | sunny | 2008-10-23 08:30:32 +0300 (Thu, 23 Oct 2008) | 13 lines
Changed paths:
M /branches/5.1/dict/dict0dict.c
M /branches/5.1/dict/dict0mem.c
M /branches/5.1/handler/ha_innodb.cc
M /branches/5.1/handler/ha_innodb.h
M /branches/5.1/include/dict0dict.h
M /branches/5.1/include/dict0mem.h
M /branches/5.1/row/row0mysql.c
branches/5.1: Backport changes from branches/zip r2725
Simplify the autoinc initialization code. This removes the
non-determinism related to reading the table's autoinc value for the first
time. This change has also reduced the sizeof dict_table_t by sizeof(ibool)
bytes because we don't need the dict_table_t::autoinc_inited field anymore.
Bug#39830 Table autoinc value not updated on first insert.
Bug#35498 Cannot get table test/table1 auto-inccounter value in ::info
Bug#36411 Failed to read auto-increment value from storage engine" in 5.1.24 auto-inc
rb://16
------------------------------------------------------------------------
Merge revisions 2837:2852 from branches/5.1:
------------------------------------------------------------------------
r2849 | sunny | 2008-10-22 12:01:18 +0300 (Wed, 22 Oct 2008) | 8 lines
Changed paths:
M /branches/5.1/handler/ha_innodb.cc
M /branches/5.1/include/row0mysql.h
M /branches/5.1/row/row0mysql.c
branches/5.1: Return the actual error code encountered when allocating
a new autoinc value. The change in behavior (bug) was introduced in 5.1.22
when we introduced the new AUTOINC locking model.
rb://31
Bug#40224 New AUTOINC changes mask reporting of deadlock/timeout errors
------------------------------------------------------------------------
r2852 | sunny | 2008-10-23 01:42:24 +0300 (Thu, 23 Oct 2008) | 9 lines
Changed paths:
M /branches/5.1/handler/ha_innodb.cc
M /branches/5.1/handler/ha_innodb.h
branches/5.1: Backport r2724 from branches/zip
Check column value against the col max value before updating the table's
global autoinc counter value. This is part of simplifying the AUTOINC
sub-system. We extract the type info from MySQL data structures at runtime.
This fixes Bug#37788 InnoDB Plugin: AUTO_INCREMENT wrong for compressed tables
------------------------------------------------------------------------
in the fix of Mantis issue #73. With key_block_size=16, we will also
have to check the available space on the uncompressed page.
Otherwise, the clustered index record can be almost 16 kilobytes in
size, and the undo log record will not fit.
column of a compressed table, the BTR_EXTERN_LEN field in the BLOB pointer
will be written as 0. Tolerate this in the functions that deal with
externally stored columns. This fixes Issue #80 and was posted at rb://26.
Note that the clustered index record is always deleted or purged last,
after any secondary index records referring to it have been deleted.
btr_free_externally_stored_field(): On an uncompressed table, zero out
the BTR_EXTERN_LEN, so that half-deleted BLOBs can be detected after
crash recovery.
btr_copy_externally_stored_field_prefix(): Return 0 if the BLOB has been
half-deleted.
row_upd_ext_fetch(): Assert that the externally stored column exists.
row_ext_cache_fill(): Allow btr_copy_externally_stored_field_prefix()
to return 0.
row_sel_sec_rec_is_for_blob(): Return FALSE if the BLOB has been half-deleted.
This is correct, because the clustered index record would have been deleted
or purged last, after any secondary index records referring to it had been
deleted.
and the adaptive hash index. This should fix Issue #95 and Issue #87.
page_zip_copy_recs(): Copy PAGE_MAX_TRX_ID as well, to have similar behavior
to page_copy_rec_list_start() and page_copy_rec_list_end().
btr_root_raise_and_insert(), btr_page_split_and_insert(), btr_lift_page_up():
Update the lock table and the adaptive hash index.
