Detailed description of changes:
r3601 | marko | 2008-12-22 16:05:19 +0200 (Mon, 22 Dec 2008) | 9 lines
branches/5.1: Make
SET SESSION TRANSACTION ISOLATION LEVEL READ COMMITTED
a true replacement of SET GLOBAL INNODB_LOCKS_UNSAFE_FOR_BINLOG=1.
This fixes an error that was introduced in r370, causing
semi-consistent read not to not unlock rows in READ UNCOMMITTED mode.
(Bug #41671, Issue #146)
rb://67 approved by Heikki Tuuri
+ add workaround for bug 38124
+ messages into the protocol when sessions are switched
+ replace error numbers by error names
+ reset of system variables to initial values per subtest
+ remove a file created by this test
+ minor improvements in structure and formatting
The greedy optimizer tracks the current level of nested joins and the position
inside these by setting and maintaining a state that's global for the whole FROM
clause.
This state was correctly maintained inside the selection of the next partial plan
table (in best_extension_by_limited_search()).
greedy_search() also moves the current position by adding the last partial match
table when there's not enough tables in the partial plan found by
best_extension_by_limited_search().
This may require update of the global state variables that describe the current
position in the plan if the last table placed by greedy_search is not a top-level
join table.
Fixed by updating the state after placing the partial plan table in greedy_search()
in the same way this is done on entering the best_extension_by_limited_search().
Fixed the signature of the function called to update the state :
check_interleaving_with_nj
Added cleanup of status variables to the end of binlog_database.
Re-recorded .result file to account for cleanup statement.
NOTE: binlog.binlog_innodb also has had an FLUSH STATUS; statement added to it as well, but
adding this cleanup as a preventative measure.
Bounds-checks and blocksize corrections were applied to user-input,
but constants in the server were trusted implicitly. If these values
did not actually meet the requirements, the user could not set change
a variable, then set it back to the (wonky) factory default or maximum
by explicitly specifying it (SET <var>=<value> vs SET <var>=DEFAULT).
Now checks also apply to the server's presets. Wonky values and maxima
get corrected at startup. Consequently all non-offsetted values the user
sees are valid, and users can set the variable to that exact value if
they so desire.
Problem 1: The test waits for an error in the slave sql thread,
then resolves the error and issues 'start slave'. However, there
is a gap between when the error is reported and the slave sql
thread stops. If this gap was long, the slave would still be
running when 'start slave' happened, so 'start slave' would fail
and cause a test failure.
Fix 1: Made wait_for_slave_sql_error wait for the slave to stop
instead of wait for error in the IO thread. After stopping, the
error code is verified. If the error code is wrong, debug info
is printed. To print debug info, the debug printing code in
wait_for_slave_param.inc was moved out to a new file,
show_rpl_debug_info.inc.
Problem 2: rpl_stm_mystery22 is a horrible name, the comments in
the file didn't explain anything useful, the test was generally
hard to follow, and the test was essentially duplicated between
rpl_stm_mystery22 and rpl_row_mystery22.
Fix 2: The test is about conflicts in the slave SQL thread,
hence I renamed the tests to rpl_{stm,row}_conflicts. Refactored
the test so that the work is done in
extra/rpl_tests/rpl_conflicts.inc, and
rpl.rpl_{row,stm}_conflicts merely sets some variables and then
sourced extra/rpl_tests/rpl_conflicts.inc.
The tests have been rewritten and comments added.
Problem 3: When calling wait_for_slave_sql_error.inc, you always
want to verify that the sql thread stops because of the expected
error and not because of some other error. Currently,
wait_for_slave_sql_error.inc allows the caller to omit the error
code, in which case all error codes are accepted.
Fix 3: Made wait_for_slave_sql_error.inc fail if no error code
is given. Updated rpl_filter_tables_not_exist accordingly.
Problem 4: rpl_filter_tables_not_exist had a typo, the dollar
sign was missing in a 'let' statement.
Fix 4: Added dollar sign.
Problem 5: When replicating from other servers than the one named
'master', the wait_for_slave_* macros were unable to print debug
info on the master.
Fix 5: Replace parameter $slave_keep_connection by
$master_connection.
2. Avoid bad effects of bug 41925 Warning 1366 Incorrect string value:
... for column processlist.info
3. Add poll routines which ensure that subtests meet stable scenarios.
This does not change the sense of the subtests.
When substituting system constant functions with a constant result
the server was not expecting that the function may return NULL.
Fixed by checking for NULL and returning Item_null (in the relevant
collation) if the result of the system constant function was NULL.
The special TRUNCATE TABLE (DDL) transaction wasn't being properly
rolled back if a error occurred during row by row deletion. The
error can be caused by a foreign key restriction imposed by InnoDB
SE and would cause the server to erroneously issue a implicit
commit.
The solution is to rollback the transaction if a truncation via row
by row deletion fails, otherwise commit. All effects of a TRUNCATE
ABLE operation are rolled back if a row by row deletion fails.
Problem: when mtr tries to create a directory, and the target
exists but is a file instead of directory, it tries several times
to create the directory again before it fails.
Fix: make it check if the target exists and is a non-directory.
The problem is that a mysql connection instance is not thread-safe
and reentrant, meaning that it can't be used concurrently and can't
be re-entered while it's already running. This applies for any form
of the server (embedded or not), but this rule can be violated in a
test case if the test sends a new command without waiting for the
result of previous command that was sent asynchronously and this can
lead to hangs when over a network or to crashes under embedded server
as the server query execution path will be re-entered concurrently
with the same connection structure.
The solution is to rework the test case so that the aforementioned
rule is obeyed.
Passing dubious "year zero" in non-zero date (not "0000-00-00") could
lead to negative value for year internally, while variable was unsigned.
This led to Really Bad Things further down the line.
Now doing calculations with signed type for year internally.
The binlog_innodb test was sensitive to what tests ran before it. Now run
FLUSH STATUS before performing operations that need to be checked.
sys_var_thd_ulong::update() was improperly casting an option value from
ulonglong to ulong before comparing it to the max allowed value. On systems
where ulong and ulonglong are of different size, this caused values greater
than ULONG_MAX to wrap around (not be truncated to ULONG_MAX, which appears to
have been the intention of the original coder), and caused some checks to work
incorrectly. This wasn't generally visible to the user, because later checks
would prevent the wrapped-around value from being used. But it caused warning
messages to differ between 32- and 64-bit platforms. Fix is to just remove the
cast. Also added a DBUG_ASSERT to ensure that the value really is capped
properly before finally stuffing it into the ulong.
locking type of temp table
The problem is that INSERT INTO .. SELECT FROM .. and CREATE
TABLE .. SELECT FROM a temporary table could inadvertently
overwrite the locking type of the temporary table. The lock
type of temporary tables should be a write lock by default.
The solution is to reset the lock type of temporary tables
back to its default value after they are used in a statement.
Execution of queries containing the CASE function of
aggregate function like in "SELECT ... CASE ARGV(...) WHEN ..."
crashed the server.
The CASE function caches pointers to concrete comparison
functions for an each pair of types of CASE-WHERE clause
parameters, i.e. for the "CASE INT_RESULT WHERE REAL_RESULT
THEN ... WHERE DECIMAL_RESULT ... END" function call it
caches comparisons for INT_RESULT with REAL_RESULT and
for INT_RESULT with DECIMAL_RESULT. Usually a result
type is known after a call to the fix_fields function,
however, the setup_copy_fields function call may
wrap aggregate items with Item_copy_string that has
STRING_RESULT result type, so setup_copy_fields may
change argument result types of the CASE function after
call to Item_func_case::fix_fields/fix_length_and_dec.
Then the Item_func_case::find_item function tries to
use comparison function for unexpected pair of the
STRING_RESULT and some other type - that caused
an assertion failure of server crash.
The Item_func_case::fix_length_and_dec function has
been modified to take into account possible STRING_RESULT
result type in the presence of aggregate arguments of
the CASE function.
Problem: When an Incident_log_event contains a bad incident number on disk,
the server crashes with an assertion.
Fix: Don't validate input with assertions. Use errors.
The problem: data file can not be deleted on win because
there is another opened instance of this file.
Data file might be opened twice, on table opening stage and
during write_row execution. We need to close both instances
to satisfy Win.
Added global status variable 'Queries' which represents
total amount of queries executed by server including
statements executed by SPs.
note: It's old behaviour of 'Questions' variable.
Problem was an errornous date that lead to end partition
was before the start, leading to a crash.
Solution was to check greater or equal instead of only
equal between start and end partition.
NOTE: partitioning pruning handles incorrect dates
differently than index lookup, which can give different
results in a partitioned table versus a non partitioned
table for queries having 'bad' dates in the where clause.
If server has not been initialized as a slave (by CHANGE MASTER), then
SHOW SLAVE STATUS will return an empty set, and caused the waiting for
Slave_IO_running or Slave_SQL_running to 'No' fail.
This patch fixed the problem by return immediately if slave is not
initialized in include/wait_for_slave_*_to_stop.inc.
mysqltest command 'shutdown_server' is supposed to shutdown the server
and wait for it to be gone, and kill it when timeout. But because the
arguments passed to my_kill were in the wrong order, 'shutdown_server'
does not wait nor kill the server at all. So after 'shutdown_server',
the server is still running, and the server may still accepting
connections.
