JOIN_CACHE's were initialized in check_join_cache_usage()
from make_join_readinfo(). After that make_join_readinfo() was looking
whether it's possible to use keyread. Later, after make_join_readinfo(),
optimizer decided whether to use filesort. And even later, at the
execution time, from join_read_first(), keyread was actually enabled.
The problem is, that if a query uses a vcol, base columns that it
depends on are automatically added to the read_set - because they're
needed to calculate the vcol. But if we're doing keyread, vcol is taken
from the index, not calculated, and base columns do not need to be
in the read set (even should not be - as they aren't getting values).
The bug was that JOIN_CACHE used read_set with base columns,
they were not read because of keyread, so it was caching garbage.
So read_set is only known after the keyread was decided. And after the
filesort was decided, as filesort doesn't use keyread. But
check_join_cache_usage() needs to be done in make_join_readinfo(),
as the code below depends on these checks,
Fix: keep JOIN_CACHE checks where they were, but move initialization
down to the very end of JOIN::optimize_inner. If keyread was enabled,
update the read_set to include only columns that are part of the index.
Copy the keyread logic from join_read_first() to happen at optimize time.
move TABLE::key_read into handler. Because in index merge and DS-MRR
there can be many handlers per table, and some of them use
key read while others don't. "keyread" is really per handler,
not per TABLE property.
Optionally do table->update_default_fields() even for INSERT
that supposedly provides values for all column. Because these
"values" might be DEFAULT, which would need table->update_default_fields()
at the end.
Also set Item_default_value::used_tables() from the default expression.
Non-zero used_field() means that mysql_insert() will initialize all
fields to their default values (with restore_record()) even if
all columns are later provided with values. Because default expressions
may refer to other columns and they must be initialized.
Oracle introduced a Memcached plugin interface to the InnoDB
storage engine in MySQL 5.6. That interface is essentially a
fork of Memcached development snapshot 1.6.0-beta1 of an old
development branch 'engine-pu'.
To my knowledge, there have not been any updates to the Memcached code
between MySQL 5.6 and 5.7; only bug fixes and extensions related to
the Oracle modifications.
The Memcached plugin is not part of the MariaDB Server. Therefore it
does not make sense to include the InnoDB interfaces for the Memcached
plugin, or to have any related configuration parameters:
innodb_api_bk_commit_interval
innodb_api_disable_rowlock
innodb_api_enable_binlog
innodb_api_enable_mdl
innodb_api_trx_level
Removing this code in one commit makes it possible to easily restore
it, in case it turns out to be needed later.
These are different bugs, but the fixing code is the same:
if window functions are used over implicit grouping then
now the execution should follow the general path calling
the function set in JOIN::first_select.
Due to this bug many queries that contained a window function
with MIN/MAX aggregation returned wrong results.
Calculation of a MIN/MAX aggregate function uses cache objects
and a comparator object that are created and set up in
Item_sum_hybrid::fix_fields () by a call of Item_sum_hybrid::setup_hybrid().
The latter binds the objects to the first argument of the
MIN/MAX function. Meanwhile window function perform aggregation
over fields of a temporary table. So binding must be done rather to
these fields. The earliest moment when setup the objects used in
MIN/max functions can be done is after all calls of the method
split_sum_func().
This patch introduces this late setup, but only for aggregate
functions used in window functions.
Probably it makes sense to use this late setup for all MIN/MAX
objects.
A proper InnoDB shutdown after aborted startup was introduced
in commit 81b7fe9d38.
Also related to this is MDEV-11985, making read-only shutdown more robust.
If startup was aborted, there may exist recovered transactions that were
not rolled back. Relax the assertions accordingly.
Before killing the server, ensure that the redo log for the
incomplete transaction is flushed, so that the AUTO_INCREMENT
sequence will always be updated. Usually the INSERT
transaction would not have persisted the sequence before the
server was killed, but sometimes it could happen, causing
result mismatch.
Note: This test used to be called innodb_fts.innodb_fts_misc_debug.
Do not effectively set DEBUG_DBUG='d' by setting DEBUG_DBUG='-d,...'.
Instead, restore the saved value of DEBUG_DBUG.
Also, split the test innodb_fts.innodb_fts_misc_debug into
innodb_fts.crash_recovery and innodb_fts.misc_debug, and enable
these tests for --valgrind, the latter test for --embedded,
and the former tests for the non-debug server.
Encryption stores used key_version to
FIL_PAGE_FILE_FLUSH_LSN_OR_KEY_VERSION (offset 26)
field. Spatial indexes store RTREE Split Sequence Number
(FIL_RTREE_SPLIT_SEQ_NUM) in the same field. Both values
can't be stored in same field. Thus, current encryption
implementation does not support encrypting spatial indexes.
fil_space_encrypt(): Do not encrypt page if page type is
FIL_PAGE_RTREE (this is required for background
encryption innodb-encrypt-tables=ON).
create_table_info_t::check_table_options() Do not allow creating
table with ENCRYPTED=YES if table contains spatial index.
recv_log_format_0_recover(): Invoke log_decrypt_after_read() after
reading the old-format redo log buffer.
With this change, we will upgrade to an encrypted redo log that
is misleadingly carrying a MySQL 5.7.9 compatible format tag while
the log blocks (other than the header and the checkpoint blocks)
are in an incompatible, encrypted format.
That needs to be fixed by introducing a new redo log format tag that
indicates that the entire redo log is encrypted.
LOG_CHECKPOINT_ARRAY_END, LOG_CHECKPOINT_SIZE: Remove.
Change some error messages to refer to MariaDB 10.2.2 instead of
MySQL 5.7.9.
recv_find_max_checkpoint_0(): Do not abort when decrypting one of the
checkpoint pages fails.
This patch complements the patch for bug 11138.
Without this patch some table-less queries with window functions
could cause crashes due to a memory overwrite.
If InnoDB is started in innodb_read_only mode such that
recovered incomplete transactions exist at startup
(but the redo logs are clean), an assertion will fail at shutdown,
because there would exist some non-prepared transactions.
logs_empty_and_mark_files_at_shutdown(): Do not wait for incomplete
transactions to finish if innodb_read_only or innodb_force_recovery>=3.
Wait for purge to finish in only one place.
trx_sys_close(): Relax the assertion that would fail first.
trx_free_prepared(): Also free recovered TRX_STATE_ACTIVE transactions
if innodb_read_only or innodb_force_recovery>=3.
Also, revert my earlier fix to MySQL 5.7 because this fix is more generic:
Bug#20874411 INNODB SHUTDOWN HANGS IF INNODB_FORCE_RECOVERY>=3
SKIPPED ANY ROLLBACK
trx_undo_fake_prepared(): Remove.
trx_sys_any_active_transactions(): Revert the changes.
Rewrite the test so that the main server is restarted, instead of
--exec $MYSQLD_CMD. In this way, the test can be run with Valgrind
and with any --mysqld=--innodb-page-size.
Also remove the workaround --skip-innodb-use-native-aio. It should
not be needed when we are inheriting the server parameters from
the test environment.
Datafile::validate_for_recovery(): Remove a redundant error message.
An error is already reported by Datafile::open_read_write() if the
file cannot be opened.
Also, do not assign SEARCH_ABORT, so that the full test will be executed
even if one step fails.
The method Item_sum::print did not print opening '(' after the name
of simple window functions (like rank, dense_rank etc).
As a result the view definitions with such window functions
were formed invalid in .frm files.
Test server startup with an empty encrypted redo log from 10.1.21.
FIXME: Pass the encryption parameters. Currently we only test startup
without properly set up encryption.
