The MLOG_FILE_WRITE_CRYPT_DATA record was completely redundant.
It can be replaced with a single MLOG_WRITE_STRING record.
To facilitate upgrade from older versions, we will retain
fil_parse_write_crypt_data().
fil_crypt_parse(): Recover fil_space_crypt_t::write_page0().
fil_space_crypt_t::write_page0(): Write everything in a single
MLOG_WRITE_STRING for easy parsing.
fil_space_crypt_t::page0_offset: Remove.
Don't do skip_setup_conds() unless all errors are checked.
Fixes following errors:
ER_PERIOD_NOT_FOUND
ER_VERS_QUERY_IN_PARTITION
ER_VERS_ENGINE_UNSUPPORTED
ER_VERS_NOT_VERSIONED
Don't do skip_setup_conds() unless all errors are checked.
Fixes following errors:
ER_PERIOD_NOT_FOUND
ER_VERS_QUERY_IN_PARTITION
ER_VERS_ENGINE_UNSUPPORTED
ER_VERS_NOT_VERSIONED
LIMIT history partitions cannot be checked by existing algorithm of
check_misplaced_rows() because working history partition is
incremented each time another one is filled. The existing algorithm
gets record and tries to decide partition id for it by
get_partition_id(). For LIMIT history it will just get first
non-filled partition.
To fix such partitions it is required to do REBUILD instead of REPAIR.
When view is merged by DT_MERGE_FOR_INSERT it is then skipped from
processing and doesn't update WHERE clause with
vers_setup_conds(). Note that view itself cannot work in
vers_setup_conds() because it doesn't have row_start, row_end
fields. Thus it is required to descend down to material TABLE_LIST
through calls of mysql_derived_prepare() and run vers_setup_conds()
from there. Luckily, all views (views of views, views of views of
views, etc.) are linked in one list through next_global pointer, so we
can skip all views of views and get straight to non-view TABLE_LIST by
checking its merge_underlying_list property for zero value (it is
assigned by DT_MERGE_FOR_INSERT for merged derived tables).
We have to do that only for UPDATE and DELETE. Other DML commands
don't use WHERE clause.
MDEV-21146 Assertion `m_lock_type == 2' in handler::ha_drop_table upon LOAD DATA
LOAD DATA does not use WHERE and the above call of vers_setup_conds()
is not needed. unit->prepare() led to wrongly locked temporary table.
"write set" for replication finally got its correct place
(mark_columns_per_binlog_row_image()). When done generally in
mark_columns_needed_for_update() it affects optimization
algorithm. used_key_is_modified, query_plan.using_io_buffer are
wrongly set and that leads to wrong prepare_for_keyread() which limits
read_set.
Turn read cache off for update and multi-update for versioned
table. no_cache is reinited on each TABLE open because it is
applicable for specific algorithms.
As a side fix vers_insert_history_row() honors vers_write setting.
Aria with row_format=fixed uses IO_CACHE of type READ_CACHE for
sequential read in update loop. When history row is inserted inside
this loop the cache misses it and fails with error.
TODO:
Currently maria_extra() does not support SEQ_READ_APPEND. Probably it
might be possible to use this type of cache.
page_zip_compress(), page_zip_decompress_low(), page_zip_reorganize():
Make use of memcpy_aligned() and memset_aligned(), so that some
operations can be translated more efficiently.
No difference was observed for AMD64 on GCC 9.2.1. But, such hints could
enable more efficient code on less common instruction set architectures,
such as 32-bit ARM, POWER or MIPS.
Patch `36a2a185fe18` introduced `wsrep_server_incoming_address()` in `10.4`.
Since AIX `/usr/include/netinet/ip.h` header defines `ip_len` as `ip_ff.ip_flen`
and `size_t const ip_len` is preprocessed as `size_t const ip_ff.ip_vhltl.ip_x.ip_xlen`,
to prevent the define from overwriting code in MariaDB,
rename the variable name to `ip_len_mdb`.
This patch is done by Tony Reix <tony.reix@atos.net>.
This patch was submitted under MCA.
Closes#1307
Before commit c0f47a4a58 (MDEV-12026)
introduced the innodb_checksum_algorithm=full_crc32 format,
it was impossible to tell if InnoDB data files contained garbage in
the FIL_PAGE_TYPE header field (and possibly other fields).
This is because before commit 3926673ce7
in MySQL 5.1.48, InnoDB would write uninitialized data to some fields,
and because there was no way to tell with which InnoDB version a data
file was created.
If fil_space_t::full_crc32() holds, the data file cannot contain
uninitialized garbage or invalid FIL_PAGE_TYPE, and thus
fil_block_check_type() should not be invoked to correct anything.
Use my_thread_var::stack_ends_here inside lf_pinbox_real_free() for address
where thread stack ends.
Remove LF_PINS::stack_ends_here.
It is not safe to assume that mysys_var that was used during pin allocation,
remains correct during free. E.g with binlog group commit in Innodb,
that frees pins for multiple Innodb transactions, it does not work
correctly.
Prior to this fix, when matching addresses using mask,
extra bits could be used for comparison, e.g to
match with "a.b.c.d/24" , 27 bits were compared rather than 24.
The patch fixes the calculation.
- It is not enough to have and DeleteSymlinks actions
because within the same installation some exes executables can be both
added and removed, which needs both adding and removing some symlinks.
Instead of CreateSymlinks, there is now FixSymlinks function, which
goes through the list, and adds or removes the symlinks as needed.
- Implemented rollback for symlink custom action.
- Generate list of symlinks in C++ file, using CMake, rather than storing
lists as MSI properties.
As part of commit 3c09f148f3
trx_undo_commit_cleanup() was always invoked with noredo=true.
The impact of this should be that some undo log pages may not be
correctly freed if the server is killed and crash recovery will be
performed. Similarly, if mariabackup --backup is being executed
concurrently with user transaction commits, it could happen that some
undo log pages in the backup will never be marked as free for reuse.
It seems that this bug should not have any user-visible
impact other than some undo pages being wasted.
If a joined table is accessed by a full text index then now no range rowid
filter can be used when accessing the records of this table.
This bug was fixed by the patch for mdev-20056.
This commit adds only a test case with a query using a full text index
and a range condition for another index.
Revert part of commit 6cedb671e9
because it turns out to be theoretically impossible to parse a
ROW_FORMAT=COMPACT or ROW_FORMAT=DYNAMIC metadata record where
the variable-length fields in the PRIMARY KEY have been written
as nonempty strings.
- Replace "class Repeat_count" with function call.
There is no reason for this class to exists. More complex and more
code than doing a function! In addition the code didn't match what
the val() function was doing.
Other things:
- Fixed compiler failure in print_cached_tables_callback()
in row_search_idx_cond_check
When usage of rowid filter is evaluated by the optimizer to join a table
to the current partial join employing a certain index it should be checked
that a key for at least the major component of this index can be constructed
using values from the columns of the partial join.
As noted in commit abd45cdc38
a search with PAGE_CUR_GE may land on the supremum record on
a leaf page that is not the rightmost leaf page. This could occur
when all keys on the current page are smaller than the search key,
and the smallest key on the successor page is larger than the search key.
Hence, after a failed PAGE_CUR_GE search, assertions
btr_pcur_is_after_last_in_tree() are bogus
and should be replaced with btr_pcur_is_after_last_on_page().
Introduce memcpy_aligned<N>(), memcmp_aligned<N>(), memset_aligned<N>()
and use them for accessing InnoDB page header fields that are known
to be aligned.
MY_ASSUME_ALIGNED(): Wrapper for the GCC/clang __builtin_assume_aligned().
Nothing similar seems to exist in Microsoft Visual Studio, and the
C++20 std::assume_aligned is not available to us yet.
Explicitly specified alignment guarantees allow compilers to generate
faster code on platforms with strict alignment rules, instead of
emitting calls to potentially unaligned memcpy(), memcmp(), or memset().
This particular assertion happened when shutting down Innodb IO.IO shutdown properly waits for all IOs to finish
However there is a race condition -
right after releasing last IO slot and before decrementing task count
in group, pending_io_count will be 0, but tasks_running will be 1,
leading to assertion.
The fix is to make task_group destructor to wait for last running task
to finish.