revid:georgi.kodinov@oracle.com-20120309130449-82e3bs5v3et1x0ef
committer: Georgi Kodinov <Georgi.Kodinov@Oracle.com>
timestamp: Fri 2012-03-09 15:04:49 +0200
message:
Bug #12408412: GROUP_CONCAT + ORDER BY + INPUT/OUTPUT SAME
USER VARIABLE = CRASH
Moved the preparation of the variables that receive the output from
SELECT INTO from execution time (JOIN:execute) to compile time
(JOIN::prepare). This ensures that if the same variable is used in the
SELECT part of SELECT INTO it will be properly marked as non-const
for this query.
Test case added.
Used proper fast iterator.
a better fix (much smaller and without regressions) is coming from 5.1
MySQL Bug #12408412: GROUP_CONCAT + ORDER BY + INPUT/OUTPUT SAME USER VARIABLE = CRASH
and
MySQL Bug#14664077 SEVERE PERFORMANCE DEGRADATION IN SOME CASES WHEN USER VARIABLES ARE USED
sql/item_func.cc:
don't use anything from Item_func_set_user_var::fix_fields()
in Item_func_set_user_var::save_item_result()
sql/sql_class.cc:
Call suv->save_item_result(item) *before* doing suv->fix_fields(), because
the former evaluates the item (and caches its value), while the latter marks
the user variable as non-const. The problem is that the item was fix_field'ed
when the user variable was const, and it doesn't expect it to change to non-const
in the middle of the execution.
allow only three failed change_user per connection.
successful change_user do NOT reset the counter
tests/mysql_client_test.c:
make --error to work for --change_user errors
(because it's conceptually wrong. only the user can decide whether the kill is
allowed to leave tables in the inconsistent state, storage engine has no say in that)
KILL now breaks locks inside InnoDB
Fixed possible deadlock when running INNODB STATUS
Added ha_kill_query() and kill_query() to send kill signal to all storage engines
Added reset_killed() to ensure we don't reset killed state while awake() is getting called
include/mysql/plugin.h:
Added thd_mark_as_hard_kill()
include/mysql/plugin_audit.h.pp:
Added thd_mark_as_hard_kill()
include/mysql/plugin_auth.h.pp:
Added thd_mark_as_hard_kill()
include/mysql/plugin_ftparser.h.pp:
Added thd_mark_as_hard_kill()
sql/handler.cc:
Added ha_kill_query() to send kill signal to all storage engines
sql/handler.h:
Added ha_kill_query() and kill_query() to send kill signal to all storage engines
sql/log_event.cc:
Use reset_killed()
sql/mdl.cc:
use thd->killed instead of thd_killed() to abort on soft kill
sql/sp_rcontext.cc:
Use reset_killed()
sql/sql_class.cc:
Fixed possible deadlock in INNODB STATUS by not getting thd->LOCK_thd_data if it's locked.
Use reset_killed()
Tell storge engines that KILL has been sent
sql/sql_class.h:
Added reset_killed() to ensure we don't reset killed state while awake() is getting called.
Added mark_as_hard_kill()
sql/sql_insert.cc:
Use reset_killed()
sql/sql_parse.cc:
Simplify detection of killed queries.
Use reset_killed()
sql/sql_select.cc:
Use reset_killed()
sql/sql_union.cc:
Use reset_killed()
storage/innobase/handler/ha_innodb.cc:
Added innobase_kill_query()
Fixed error reporting for interrupted queries.
storage/xtradb/handler/ha_innodb.cc:
Added innobase_kill_query()
Fixed error reporting for interrupted queries.
The patch decreases the duration of LOCK_thread_count, so it is not hold during THD destructor and freeing memory.
This mutex now only protects the integrity of threads list, when removing THD from it, and thread_count variable.
The add_to_status() function that updates global status during client disconnect, is now correctly protected by the LOCK_status mutex.
Benchmark : in a "non-persistent" sysbench test (oltp_ro with reconnect after each query), ~ 25% more connects/disconnects were measured
XID_STATE->XID.KEY(),
XID_STATE->XID.KEY_LENGTH())==0
This bug is a regression of bug#11759534 - 51855: RACE CONDITION
IN XA START.
The reason for regression is that the changes that fixes the original
bug wasn't merged from mysql-5.1 into mysql-5.5 and mysql-trunk.
Only null-merge was done for the patch changeset.
To incorporate lost changes the manual merge have been done.
Additionally the call of trans_rolback() was added into trans_xa_start()
in case if xid_cache_insert is failed() after transaction has been started.
If we don't call trans_rollback() we would never reset the flag
SERVER_STATUS_IN_TRANS in THD::server_status and therefore all subsequent
attempts to execute XA START in the connection where the error was occurred
will be failed since thd->in_active_multi_stmt_transaction() will return
the true every time when trans_xa_start is called.
The latest changes were absent in patch for mysql-5.1
When master and slave have different schemas, in particular different
AUTO_INCREMENT columns, INSERT_ID events logged for a given table on
master may be applied to a different table on slave on SBR, e.g.:
master has one table (t1) with one auto-inc column and another table
(t2) without auto-inc column, on slave t1 does not have auto-inc
column (despite having the same columns) and t2 has a auto-inc
column. The INSERT_ID that is intended for t1, since t1 on slave
doesn't have auto-inc column is used on t2, causing consistency
problems.
To fix this incorrect behaviour, auto-inc interval allocation via
INSERT_ID is made effectively terminated at the end of top-level
statements on slave and binlog replay.
THREAD POOLING STRESS TEST
PROBLEM:
Connection stress tests which consists of concurrent
kill connections interleaved with mysql ping queries
cause the mysqld server which uses thread pool scheduler
to crash.
FIX:
Killing a connection involves shutdown and close of client
socket and this can cause EPOLLHUP(or EPOLLERR) events to be
to be queued and handled after disarming and cleanup of
of the connection object (THD) is being done.We disarm the
the connection by modifying the epoll mask to zero which
ensure no events come and release the ownership of waiting
thread that collect events and then do the cleanup of THD.
object.As per the linux kernel epoll source code (
http://lxr.linux.no/linux+*/fs/eventpoll.c#L1771), EPOLLHUP
(or EPOLLERR) can't be masked even if we set EPOLL mask
to zero. So we disarm the connection and thus prevent
execution of any query processing handler/queueing to
client ctx. queue by removing the client fd from the epoll
set via EPOLL_CTL_DEL. Also there is a race condition which
involve the following threads:
1) Thread X executing KILL CONNECTION Y and is in THD::awake
and using mysys_var (holding LOCK_thd_data).
2) Thread Y in tp_process_event executing and is being killed.
3) Thread Z receives KILL flag internally and possible call
the tp_thd_cleanup function which set thread session variable
and changing mysys_var.
The fix for the above race is to set thread session variable
under LOCK_thd_data.
We also do not call THD::awake if we found the thread in the
thread list that is to be killed but it's KILL_CONNECTION flag
set thus avoiding any possible concurrent cleanup. This patch
is approved by Mikael Ronstrom via email review.
VARIABLES
Analysis:
-------------
After executing the query, new value of the user defined
variables are set in the function "select_dumpvar::send_data".
"select_dumpvar::send_data" first calls function
"Item_func_set_user_var::save_item_result()". This function
checks the nullness of the Item_field passed as parameter
to it and saves it. The nullness of item is stored with
arg[0]'s null_value flag. Then "select_dumpvar::send_data" calls
"Item_func_set_user_var::update()" which notices null
result that was saved and calls "Item_func_set_user_var::
update_hash". But here null_value is not set and args[0]
is different from that given to function "Item_func_set_user_var::
set_item_result()". This causes "Item_func_set_user_var::
update_hash" function to believe that its getting non-null value.
"user_var_entry::length" set to 0 and hence "user_var_entry::value"
is made to point to extra_area allocated in "user_var_entry".
And "Item_func_set_user_var::update_hash" tries to write
at memory beyond extra_area for result type DECIMAL. Because of
this invalid write issue is reported by Valgrind.
Before this bug was introduced, we avoided this problem by
creating "Item_func_set_user_var" object with the same
Item_field as arg[0] and as parameter to
Item_func_set_user_var::save_item_result(). But now
they are refering to different args[0]. Because of this
null_value flag set in parameter Item_field in function
"Item_func_set_user_var::save_item_result()" is not
reflected in "Item_func_set_user_var" object.
Fix:
------------
This issue is reported on versions 5.5.24. Issue does not exists
in 5.5.23, 5.1, 5.6 and trunk.
This issue was introduced by
revid:georgi.kodinov@oracle.com-20120309130449-82e3bs5v3et1x0ef (fix for
bug #12408412), which was pushed into 5.5 and later releases. This patch
has later been reversed in 5.6 and trunk by
revid:norvald.ryeng@oracle.com-20121010135242-xj34gg73h04hrmyh (fix for
bug #14664077). Backported this patch in 5.5 also to fix this issue.
sql/item_func.cc:
here unsigned value is converted to signed value.
sql/item_func.h:
last_insert_id() gives an auto_incremented value which can be
positive only,so defined it as a unsigned longlong sets the
unsigned_flag to 1.
two tests still fail:
main.innodb_icp and main.range_vs_index_merge_innodb
call records_in_range() with both range ends being open
(which triggers an assert)
PROBLEM:
mysql provides a feature where in a session which is
idle for a period specified by the wait_timeout variable
(whose value is in seconds), the session is closed
This feature is not present when we use thread pool.
FIX:
This patch implements the interface functions which is
required to implement the wait_timeout functionality
in the thread pool plugin.
KEY UPDATES WITH A LIMIT OF 1
Problem: The unsafety warning for statements such as
update...limit1 where pk=1 are thrown when binlog-format
= STATEMENT,despite of the fact that such statements are
actually safe. this leads to filling up of the disk space
with false warnings.
Solution: This is not a complete fix for the problem, but
prevents the disks from getting filled up. This should
therefore be regarded as a workaround. In the future this
should be overriden by server general suppress/filtering
framework. It should also be noted that another worklog is
supposed to defeat this case's artificial unsafety.
We use a warning suppression mechanism to detect warning flood,
enable the suppression, and disable this when the average
warnings/second has reduced to acceptable limits.
Activation: The supression for LIMIT unsafe statements are
activated when the last 50 warnings were logged in less
than 50 seconds.
Supression: Once activated this supression will prevent the
individual warnings to be logged in the error log, but print
the warning for every 50 warnings with the note:
"The last warning was repeated N times in last S seconds"
Noteworthy is the fact that this supression works only on the
error logs and the warnings seen by the clients will remain as
it is (i.e. one warning/ unsafe statement)
Deactivation: The supression will be deactivated once the
average # of warnings/sec have gone down to the acceptable limits.
sql/sql_class.cc:
Added code to supress warning while logging them to error-log.
Analysis:
-------------
If server is started with limit of MAX_CONNECTIONS and
MAX_USER_CONNECTIONS then only MAX_USER_CONNECTIONS of any particular
users can be connected to server and total MAX_CONNECTIONS of client can
be connected to server.
Server maintains a counter for total CONNECTIONS and total CONNECTIONS
from particular user.
Here, MAX_CONNECTIONS of connections are created to server. Out of this
MAX_CONNECTIONS, connections from particular user (say USER1) are
also created. The connections from USER1 is lesser than
MAX_USER_CONNECTIONS. After that there was one more connection request from
USER1. Since USER1 can still create connections as he havent reached
MAX_USER_CONNECTIONS, server increments counter of CONNECTIONS per user.
As server already has MAX_CONNECTIONS of connections, next check to total
CONNECTION count fails. In this case control is returned WITHOUT
decrementing the CONNECTIONS per user. So the counter per user CONNECTIONS goes
on incrementing for each attempt until current connections are closed.
And because of this counter per CONNECTIONS reached MAX_USER_CONNECTIONS.
So, next connections form USER1 user always returns with MAX_USER_CONNECTION
limit error, even when total connection to sever are less than MAX_CONNECTIONS.
Fix:
-------------
This issue is occurred because of not handling counters properly in the
server. Changed the code to handle per user connection counters properly.
PROBLEM:
Threads end-up in deadlock due to locks acquired as described
below,
con1: Run Query on a table.
It is important that this SELECT must back-off while
trying to open the t1 and enter into wait_for_condition().
The SELECT then is blocked trying to lock mysys_var->mutex
which is held by con3. The very significant fact here is
that mysys_var->current_mutex will still point to LOCK_open,
even if LOCK_open is no longer held by con1 at this point.
con2: Try dropping table used in con1 or query some table.
It will hold LOCK_open and be blocked trying to lock
kernel_mutex held by con4.
con3: Try killing the query run by con1.
It will hold THD::LOCK_thd_data belonging to con1 while
trying to lock mysys_var->current_mutex belonging to con1.
But current_mutex will point to LOCK_open which is held
by con2.
con4: Get innodb engine status
It will hold kernel_mutex, trying to lock THD::LOCK_thd_data
belonging to con1 which is held by con3.
So while technically only con2, con3 and con4 participate in the
deadlock, con1's mysys_var->current_mutex pointing to LOCK_open
is a vital component of the deadlock.
CYCLE = (THD::LOCK_thd_data -> LOCK_open ->
kernel_mutex -> THD::LOCK_thd_data)
FIX:
LOCK_thd_data has responsibility of protecting,
1) thd->query, thd->query_length
2) VIO
3) thd->mysys_var (used by KILL statement and shutdown)
4) THD during thread delete.
Among above responsibilities, 1), 2)and (3,4) seems to be three
independent group of responsibility. If there is different LOCK
owning responsibility of (3,4), the above mentioned deadlock cycle
can be avoid. This fix introduces LOCK_thd_kill to handle
responsibility (3,4), which eliminates the deadlock issue.
Note: The problem is not found in 5.5. Introduction MDL subsystem
caused metadata locking responsibility to be moved from TDC/TC to
MDL subsystem. Due to this, responsibility of LOCK_open is reduced.
As the use of LOCK_open is removed in open_table() and
mysql_rm_table() the above mentioned CYCLE does not form.
Revision ID for changes,
open_table() = dlenev@mysql.com-20100727133458-m3ua9oslnx8fbbvz
mysql_rm_table() = jon.hauglid@oracle.com-20101116100012-kxep9txz2fxy3nmw
BUG#11761686 insert_id event is not filtered.
Two issues are covered.
INSERT into autoincrement field which is not the first part in the composed primary key
is unsafe by autoincrement logging design. The case is specific to MyISAM engine
because Innodb does not allow such table definition.
However no warnings and row-format logging in the MIXED mode was done, and
that is fixed.
Int-, Rand-, User-var log-events were not filtered along with their parent
query that made possible them to screw up execution context of the following
query.
Fixed with deferring their execution until the parent query.
******
Bug#11754117
Post review fixes.
mysql-test/suite/rpl/r/rpl_auto_increment_bug45679.result:
a new result file is added.
mysql-test/suite/rpl/r/rpl_filter_tables_not_exist.result:
results updated.
mysql-test/suite/rpl/t/rpl_auto_increment_bug45679.test:
regression test for BUG#11754117-45670 is added.
mysql-test/suite/rpl/t/rpl_filter_tables_not_exist.test:
regression test for filtering issue of BUG#11754117 - 45670 is added.
sql/log_event.cc:
Logics are added for deferring and executing events associated
with the Query event.
sql/log_event.h:
Interface to deferred events batch execution is added.
sql/rpl_rli.cc:
initialization for new RLI members is added.
sql/rpl_rli.h:
New members to RLI are added to facilitate deferred events gathering
and execution control;
two general character RLI cleanup methods are constructed.
sql/rpl_utility.cc:
Deferred_log_events methods are difined.
sql/rpl_utility.h:
A new class Deferred_log_events is defined to implement
IRU events gathering, execution and cleanup.
sql/slave.cc:
Necessary changes to initialize `rli->deferred_events' and prevent
deferred event deletion in the main read-exec branch.
sql/sql_base.cc:
A new safe-check function for multi-part pk with auto-increment is defined
and deployed in lock_tables().
sql/sql_class.cc:
Initialization for a new member and replication cleanups are added
to THD class.
sql/sql_class.h:
THD class receives a new member to hold a specific execution
context for slave applier.
sql/sql_parse.cc:
Execution of the deferred event in started prior to its parent query.
mysql-test/suite/innodb/t/group_commit_crash.test:
remove autoincrement to avoid rbr being used for insert ... select
mysql-test/suite/innodb/t/group_commit_crash_no_optimize_thread.test:
remove autoincrement to avoid rbr being used for insert ... select
mysys/my_addr_resolve.c:
a pointer to a buffer is returned to the caller -> the buffer cannot be on the stack
mysys/stacktrace.c:
my_vsnprintf() is ok here, in 5.5
bug in semisync plugin. It didn't check thd->killed before waiting on mysys->current_cond,
and thus an attepmt to kill the thread (on shutdown) was lost
plugin/semisync/semisync_master.cc:
mdev:201 lp:962540
https://mariadb.atlassian.net/browse/MDEV-28
This task implements a new clause LIMIT ROWS EXAMINED <num>
as an extention to the ANSI LIMIT clause. This extension
allows to limit the number of rows and/or keys a query
would access (read and/or write) during query execution.
USER VARIABLE = CRASH
Moved the preparation of the variables that receive the output from
SELECT INTO from execution time (JOIN:execute) to compile time
(JOIN::prepare). This ensures that if the same variable is used in the
SELECT part of SELECT INTO it will be properly marked as non-const
for this query.
Test case added.
Used proper fast iterator.
The result of materialization of the right part of an IN subquery predicate
is placed into a temporary table. Each row of the materialized table is
distinct. A unique key over all fields of the temporary table is defined and
created. It allows to perform key look-ups into the table.
The table created for a materialized subquery can be accessed by key as
any other table. The function best_access-path search for the best access
to join a table to a given partial join. With some where conditions this
function considers a possibility of a ref_or_null access. If such access
employs the unique key on the temporary table then when estimating
the cost this access the function tries to use the array rec_per_key. Yet,
such array is not built for this unique key. This causes a crash of the server.
Rows returned by the subquery that contain nulls don't have to be placed
into temporary table, as they cannot be match any row produced by the
left part of the subquery predicate. So all fields of the temporary table
can be defined as non-nullable. In this case any ref_or_null access
to the temporary table does not make any sense and it does not make sense
to estimate such an access.
The fix makes sure that the temporary table for a materialized IN subquery
is defined with columns that are all non-nullable. The also ensures that
any row with nulls returned by the subquery is not placed into the
temporary table.
KEY HANDLING ON SUBSEQUENT CREATE TABLE IF NOT EXISTS
PROBLEM:
--------
Consider a SP routine which does CREATE TABLE
with REFERENCES clause. The first call to this routine
invokes parser and the parsed items are cached, so as
to avoid parsing for the second execution of the routine.
It is obsevered that valgrind reports a warning
upon read of thd->lex->alter_info->key_list->Foreign_key object,
which seem to be pointing to a invalid memory address
during second time execution of the routine. Accessing this object
theoretically could cause a crash.
ANALYSIS:
---------
The problem stems from the fact that for some reason
elements of ref_columns list in thd->lex->alter_info->
key_list->Foreign_key object are changed to point to
objects allocated on runtime memory root.
During the first execution of routine we create
a copy of thd->lex->alter_info object.
As part of this process we create a clones of objects in
Alter_info::key_list and of Foreign_key object in particular.
Then Foreign_key object is cloned for some reason we
perform shallow copies of both Foreign_key::ref_columns
and Foreign_key::columns list. So new instance of
Foreign_key object starts to SHARE contents of ref_columns
and columns list with the original instance.
After that as part of cloning process we call
list_copy_and_replace_each_value() for elements of
ref_columns list. As result ref_columns lists in both
original and cloned Foreign_key object start to contain
pointers to Key_part_spec objects allocated on runtime
memory root because of shallow copy.
So when we start copying of thd->lex->alter_info object
during the second execution of stored routine we indeed
encounter pointer to the Key_part_spec object allocated
on runtime mem-root which was cleared during at the end
of previous execution. This is done in sp_head::execute(),
by a call to free_root(&execute_mem_root,MYF(0));
As result we get valgrind warnings about accessing
unreferenced memory.
FIX:
----
The safest solution to this problem is to
fix Foreign_key(Foreign_key, MEM_ROOT) constructor to do
a deep copy of columns lists, similar to Key(Key, MEM_ROOT)
constructor.
