This is a backport of the applicable part of
commit 93475aff8de80a0ef53cbee924bcb70de6e86f2c and
commit 2c39f69d34e64a5cf94720e82e78c0ee91bd4649
from 10.4.
Before 10.4 and Galera 4, WSREP_ON is a macro that points to
a global Boolean variable, so it is not that expensive to
evaluate, but we will add an unlikely() hint around it.
WSREP_ON_NEW: Remove. This macro was introduced in
commit c863159c320008676aff978a7cdde5732678f975
when reverting WSREP_ON to its previous definition.
We replace some use of WSREP_ON with WSREP(thd), like it was done
in 93475aff8de80a0ef53cbee924bcb70de6e86f2c. Note: the macro
WSREP() in 10.1 is equivalent to WSREP_NNULL() in 10.4.
Item_func_rand::seed_random(): Avoid invoking current_thd
when WSREP is not enabled.
Unfortunate DROP TEMPORARY..IF EXISTS on a regular table may allow
subsequent CREATE TABLE statements to steal away the PFS_table_share
instance from the dropped table.
Problem:
=======
Failed CREATE OR REPLACE TEMPORARY TABLE statement which dropped the table but
failed at a later stage of creation of temporary table is not written to
binarylog in row based replication. This causes the slave to diverge.
Analysis:
========
CREATE OR REPLACE statements work as shown below.
CREATE OR REPLACE TABLE table_name (a int);
is basically the same as:
DROP TABLE IF EXISTS table_name;
CREATE TABLE table_name (a int);
Hence every CREATE OR REPLACE TABLE command which dropped the table should be
written to binary log, even when following CREATE TABLE part fails. In order
to achieve this, during the execution of CREATE OR REPLACE command, when a
table is dropped 'thd->log_current_statement' flag is set. When table creation
results in an error within 'mysql_create_table' code, the error handling part
looks for this flag. If it is set the failed CREATE OR REPLACE statement is
written into the binary log inspite of error. This ensure that slave doesn't
diverge from the master. In case of row based replication the error handling
code returns very early, if the table is of type temporary. This is done based
on the assumption that temporary tables are not replicated in row based
replication.
It fails to handle the cases where a temporary table was created as part of
statement based replication at an earlier stage and the binary log format was
changed to row because of an unsafe statement. In this case when a CREATE OR
REPLACE statement is executed on this temporary table it will dropped but the
query will not be written to binary log. Hence slave diverges.
Fix:
===
In error handling code check the return status of create table operation. If
it is successful and replication mode is row based and table is of type
temporary then return. Other wise proceed further to the code which checks for
thd->log_current_statement flag and does appropriate logging.
MDEV-5589 commit set up a policy to skip DROP TEMPORARY TABLE binary logging
in case the target table has not been "CREATEed" in binlog (no CREATE
Query-log-event was logged into the binary log).
It turns out that
1. the rule did not cover non-existing table DROPped with IF-EXISTS clause.
The logged-create knowledge for the non-existing one does not even need
MDEV-5589 patch, and
2. connection close disobeys it to trigger automatic DROP-IF-EXISTS
binlogging.
Either 1 or 2 or even both is/are also responsible for unexpected binlog
records observed in MDEV-17863, actually rendering a referred
@@global.read_only irrelevant as far as the described stored procedure
definition *and* the ROW binlog-format are concerned.
Analysis
========
Point in time recovery using mysqlbinlog containing queries
operating on temporary tables results in an error.
While writing the query log event in the binary log, the
thread id used for execution of DROP TABLE and DELETE commands
were incorrect. The thread variable 'thread_specific_used'
is used to determine whether a specific thread id is to used
while executing the statements i.e using 'SET
@@session.pseudo_thread_id'. This variable was not set
correctly for DROP TABLE query and was never set for DELETE
query. The thread id is important for temporary tables
since the tables are session specific. DROP TABLE and DELETE
queries executed using a wrong thread id resulted in errors
while applying the queries generated by mysqlbinlog utility.
Fix
===
Set the 'thread_specific_used' THD variable for DROP TABLE and
DELETE queries.
ReviewBoard: 21833
A sequel to 9180e86 and 149b754.
ALTER TABLE ... ADD FOREIGN KEY may crash if parent table is updated
concurrently.
Block FK parent table updates even earlier, before intermediate child
table is created.
Use proper charset info for my_casedn_str() and don't update original
identifiers so that lower_cast_table_names == 2 is honoured.
ALTER TABLE ... ADD FOREIGN KEY may trigger assertion failure when
it has LOCK=EXCLUSIVE clause or concurrent FLUSH TABLES is being
executed.
In both cases being altered table is marked as flushed, which forces
subsequent attempt to open parent table to re-open. Which in turn is
not allowed while transaction is running.
Rather than opening parent table, just take appropriate MDL lock.
Also removed table_already_fk_prelocked() check: MDL itself has much
better methods to handle duplicate locks. E.g. the former won't acquire
MDL_SHARED_NO_WRITE if it already has MDL_SHARED_READ.
Problem:
========
Server fails to notify the engine by not setting the ADD_PK_INDEX and
DROP_PK_INDEX When there is a
i) Change in candidate for primary key.
ii) New candidate for primary key.
Fix:
====
Server sets the ADD_PK_INDEX and DROP_PK_INDEX while doing alter for the
above problematic case.
Locked_tables_list::unlock_locked_tables
Similarly to regular DROP TABLE, don't leave locked tables mode if CREATE OR
REPLACE dropped temporary table but failed to cerate new one.
The problem is that there's no track of which temporary table was "locked" by
LOCK TABLES.
ALTER TABLE locks the table with TL_READ_NO_INSERT, to prevent the
source table modifications while it's being copied. But there's an
indirect way of modifying a table, via cascade FK actions.
After previous commits, an attempt to modify an FK parent table
will cause FK children to be prelocked, so the table-being-altered
cannot be modified by a cascade FK action, because ALTER holds a
lock and prelocking will wait.
But if a new FK is being added by this very ALTER, then the target
table is not locked yet (it's a temporary table). So, we have to
lock FK parents explicitly.
Problem was that the number of NULL bit's was record wrong in the
.frm file because there could be more fields marked NOT_NULL after the
number of not_null fields where recorded.
Fixed by copying test for virtual fields from prepare_create_field()
The code change, only the test, doesn't have to be merged to 10.3
as this is fixed there.
MEMORY table could be renamed into a non-extistent database.
rename() is documented to return ENOENT when the source file does not
exist OR when the target directory not exist. Nonexistent source .frm
file is ok (table can still exist in the engine), nonexistent target
directory is not.
Make my_rename to use ENOTDIR for the latter case. Make RENAME TABLE
issue an appropriate error ("unknown database" instead of "unknown table")
Crash happened because in discover, table->work_part_info was not properly
reset before execution.
Fixed by resetting before calling execute alter table, create table or
mysql_create_frm_image.
Problem was that if copy_data_between_tables() didn't do proper
clean up in case of failures:
- copy object was not properly freed
- end_bulk_insert() was not called
- mysql_trans_prepare_alter_copy_data() set THD->transaction.on to
false which was not properly restored
The last part caused a crash in Aria as Aria depends on that THD
is correct.
Other things:
- Reset info->switched_transactional after usage (safety)
- Reset bulk_insert_single_undo (safety)
ALTER TABLE ... ADD PARTITION modifies the open TABLE structure,
and sets table->need_reopen=1 to reset these modifications
in case of an error.
But under LOCK TABLES the table isn't get reopened, despite need_reopen.
Fixed by reopening need_reopen tables under LOCK TABLE.
PREBUILT->TABLE->N_MYSQL_HANDLES_OPENED == 1
ANALYSIS:
=========
Adding unique index to a InnoDB table which is locked as
mutliple instances may trigger an InnoDB assert.
When we add a primary key or an unique index, we need to
drop the original table and rebuild all indexes. InnoDB
expects that only the instance of the table that is being
rebuilt, is open during the process. In the current
scenario we have opened multiple instances of the table.
This triggers an assert during table rebuild.
'Locked_tables_list' encapsulates a list of all
instances of tables locked by LOCK TABLES statement.
FIX:
===
We are now temporarily closing all the instances of the
table except the one which is being altered and later
reopen them via Locked_tables_list::reopen_tables().
The thd->lex->part_info should be kept intact during PS
execution. Or the second execution gets that modified part_info.
Let's modify ths->work_part_info instead.
