Merged lp:maria/maria-10.0-galera up to revision 3879.
Added a new functions to handler API to forcefully abort_transaction,
producing fake_trx_id, get_checkpoint and set_checkpoint for XA. These
were added for future possiblity to add more storage engines that
could use galera replication.
Let TABLE_SHARE::tdc.free_tables, TABLE_SHARE::tdc.all_tables,
TABLE_SHARE::tdc.flushed and corresponding invariants be protected by
per-share TABLE_SHARE::tdc.LOCK_table_share instead of global LOCK_open.
Reduced number of my_hash_sort_bin() calls from 4 to 1 per query.
Reduced number of memory accesses done by my_hash_sort_bin().
Details:
- let MDL subsystem use pre-calculated hash value for hash
inserts and deletes
- let table cache use pre-calculated MDL hash value
- MDL namespace is excluded from hash value calculation, so that
hash value can be used by table cache as is
- hash value for MDL is calculated as resulting hash value + MDL
namespace
- extended hash implementation to accept user defined hash function
includes:
* remove some remnants of "Bug#14521864: MYSQL 5.1 TO 5.5 BUGS PARTITIONING"
* introduce LOCK_share, now LOCK_ha_data is strictly for engines
* rea_create_table() always creates .par file (even in "frm-only" mode)
* fix a 5.6 bug, temp file leak on dummy ALTER TABLE
The patch contributed by Konstantin Osipov applied.
Native comments:
Implement multiple user-level locks per connection.
GET_LOCK() function in MySQL allows a connection to hold at most
one user level lock. Taking a new lock automatically releases the
old lock, if any.
The limit of one lock per session existed since early versions
of MySQL didn't have a deadlock detector for SQL locks.
MDL patches in MySQL 5.5 added a deadlock detector,
so starting from 5.5 it became possible to take multiple locks
in any order -- a deadlock, should it occur, would be detected
and an error returned to the client which closed the wait chain.
This is exactly what is done in this patch: ULLs are moved
to use MDL subsystem.
ROBUST AGAINST BUGS IN CALLERS".
Both MDL subsystems and Table Definition Cache code assume
that callers ensure that names of objects passed to them are
not longer than NAME_LEN bytes. Unfortunately due to bugs in
callers this assumption might be broken in some cases. As
result we get nasty bugs causing buffer overruns when we
construct MDL key or TDC key from object names.
This patch makes MDL and TDC code more robust against such
bugs by ensuring that we always checking size of result
buffer when constructing MDL and TDC keys. This doesn't
free its callers from ensuring that both db and table names
are shorter than NAME_LEN bytes. But at least these steps
prevents buffer overruns in case of bug in caller, replacing
them with less harmful behavior.
This is 5.5-only version of patch.
Changed code of MDL_key::mdl_key_init() to take into account
size of buffer for the key.
Introduced new version of create_table_def_key() helper function
which constructs TDC key without risk of result buffer overrun.
Places in code that construct TDC keys were changed to use this
function.
Also changed rm_temporary_table() and open_new_frm() functions
to avoid use of "unsafe" strmov() and strxmov() functions and
use safer strnxmov() instead.
Code in MDL subsystem assumes that identifiers of objects can't
be longer than NAME_LEN characters. This assumption was broken
when one tried to construct MDL_key based on table alias, which
can have arbitrary length. Since MDL_key's (and MDL locks) are
not really used for table aliases this patch changes code to
not initialize MDL_key object for table list element representing
aliases.
and small collateral changes
mysql-test/lib/My/Test.pm:
somehow with "print" we get truncated writes sometimes
mysql-test/suite/perfschema/r/digest_table_full.result:
md5 hashes of statement digests differ, because yacc token codes are different in mariadb
mysql-test/suite/perfschema/r/dml_handler.result:
host table is not ported over yet
mysql-test/suite/perfschema/r/information_schema.result:
host table is not ported over yet
mysql-test/suite/perfschema/r/nesting.result:
this differs, because we don't rewrite general log queries, and multi-statement
packets are logged as a one entry. this result file is identical to what mysql-5.6.5
produces with the --log-raw option.
mysql-test/suite/perfschema/r/relaylog.result:
MariaDB modifies the binlog index file directly, while MySQL 5.6 has a feature "crash-safe binlog index" and modifies a special "crash-safe" shadow copy of the index file and then moves it over. That's why this test shows "NONE" index file writes in MySQL and "MANY" in MariaDB.
mysql-test/suite/perfschema/r/server_init.result:
MariaDB initializes the "manager" resources from the "manager" thread, and starts this thread only when --flush-time is not 0. MySQL 5.6 initializes "manager" resources unconditionally on server startup.
mysql-test/suite/perfschema/r/stage_mdl_global.result:
this differs, because MariaDB disables query cache when query_cache_size=0. MySQL does not
do that, and this causes useless mutex locks and waits.
mysql-test/suite/perfschema/r/statement_digest.result:
md5 hashes of statement digests differ, because yacc token codes are different in mariadb
mysql-test/suite/perfschema/r/statement_digest_consumers.result:
md5 hashes of statement digests differ, because yacc token codes are different in mariadb
mysql-test/suite/perfschema/r/statement_digest_long_query.result:
md5 hashes of statement digests differ, because yacc token codes are different in mariadb
mysql-test/suite/rpl/r/rpl_mixed_drop_create_temp_table.result:
will be updated to match 5.6 when alfranio.correia@oracle.com-20110512172919-c1b5kmum4h52g0ni and anders.song@greatopensource.com-20110105052107-zoab0bsf5a6xxk2y are merged
mysql-test/suite/rpl/r/rpl_non_direct_mixed_mixing_engines.result:
will be updated to match 5.6 when anders.song@greatopensource.com-20110105052107-zoab0bsf5a6xxk2y is merged
FOREVER MDL LOCK
Analysis:
----------
While granting MDL lock for the lock requests in wait queue,
first the lock is granted to the high priority lock types
and then to the low priority lock types.
MDL Priority Matrix,
+-------------+----+---+---+---+----+-----+
| Locks | | | | | | |
| has Priority| | | | | | |
| over ---> | S | SR| SW| SU| SNW| SNRW|
+-------------+----+---+---+---+----+-----+
| X | + | + | + | + | + | + |
+-------------|----|---|---|---|----|-----|
| SNRW | - | + | + | - | - | - |
+-------------|----|---|---|---|----|-----|
| SNW | - | - | + | - | - | - |
+-------------+----+---+---+---+----+-----+
Here '+' means, Lock priority is higher.
'-' means, Has same priority
In the scenario where,
*. Lock wait queue has requests of type S/SR/SW/SU.
*. And locks of high priority X/SNRW/SNW are requested
continuously.
In this case, while granting lock, always first high priority
lock requests(X/SNRW/SNW) are considered. Low priority
locks(S/SR/SW/SU) will not get chance and they will
wait forever.
In the scenario for which this bug is reported, application
executed many LOCK TABLES ... WRITE statements concurrently.
These statements request SNRW lock. Also there were some
connections trying to execute DML statements requesting SR
lock. Since SNRW lock request has higher priority (and as
they were too many waiting SNRW requests) lock is always
granted to it. So, lock request SR will wait forever, resulting
in DML starvation.
How is this handled in 5.1?
---------------------------
Even in 5.1 we have low priority lock starvation issue.
But, in 5.1 thread locking, system variable
"max_write_lock_count" can be configured to grant
some pending read lock requests. After
"max_write_lock_count" of write lock grants all the low
priority locks are granted.
Why this issue is seen in 5.5/trunk?
---------------------------------
In 5.5/trunk MDL locking, "max_write_lock_count" system
variable exists but not used in MDL, only thread lock uses
it. So no effect of "max_write_lock_count" in MDL locking.
This means that starvation of metadata locks is possible
even if max_write_lock_count is used.
Looks like, customer was using "max_write_lock_count" in
5.1 and when upgraded to 5.5, starvation is seen because
of not having effect of "max_write_lock_count" in MDL.
Fix:
----------
As a fix, support for max_write_lock_count is added to MDL.
To maintain write lock counter per MDL_lock object, new
member "m_hog_lock_count" is added in MDL_lock.
And following logic is added to increment the counter in
function reschedule_waiters,
(reschedule_waiters function is called while thread is
releasing the lock)
- After granting lock request from the wait queue.
- Check if there are any S/SR/SU/SW exists in the wait queue
- If yes then increment the "m_hog_lock_count"
And following logic is added in the same function to
handle pending S/SU/SR/SW locks
- Before granting locks
- Check if max_write_lock_count <= m_hog_lock_count
- If Yes, then try to grant S/SR/SW/SU locks.
(Since all of these has same priority, all locks are
granted together. But some lock grant may fail because
of grant incompatibility)
- Reset m_hog_lock_count if there no low priority lock
requests in wait queue.
- return
Note:
--------------------------
In the lock priority matrix explained above,
though X has priority over the SNW and SNRW. X locks is
taken mostly for RENAME, TRUNCATE, CREATE ... operations.
So lock type X may not be requested in loop continuously
in real world applications, as compared to other lock
request types. So, lock request of type SNW and SNRW are
not starved. So, we can grant all S/SR/SU/SW in one shot,
without considering SNW & SNRW lock request starvation.
ALTER table operations take SU lock first and then
upgrade to SNW if required. All S, SR, SW, SU have same
lock priority. So while granting SU, request of types
SR, SW, S are also granted in one shot. So, lock request
of type SU->SNW in loop will not make other low priority
lock request to starve.
But, when there is request for lock of type SNRW, lock
requests of lower priority types are not granted. And if
SNRW is requested in loop continuously then all
S, SR, SW, SU are starved.
This patch addresses the latter scenario.
When we have S/SR/SW/SU in wait queue and if
there are
- Continuous SNRW lock requests
- OR one or more X and Continuous SNRW lock requests.
- OR one SNW and Continuous SNRW lock requests.
- OR one SNW, one or more X and continuous SNRW lock
requests.
in wait queue then, S/SR/SW/SU lock request are starved.
BY CACHING OR REDUCING CREATEEVENT CALLS".
5.5 versions of MySQL server performed worse than 5.1 versions
under single-connection workload in autocommit mode on Windows XP.
Part of this slowdown can be attributed to overhead associated
with constant creation/destruction of MDL_lock objects in the MDL
subsystem. The problem is that creation/destruction of these
objects causes creation and destruction of associated
synchronization primitives, which are expensive on Windows XP.
This patch tries to alleviate this problem by introducing a cache
of unused MDL_object_lock objects. Instead of destroying such
objects we put them into the cache and then reuse with a new
key when creation of a new object is requested.
To limit the size of this cache, a new --metadata-locks-cache-size
start-up parameter was introduced.
mysql-test/r/mysqld--help-notwin.result:
Updated test after adding --metadata-locks-cache-size
parameter.
mysql-test/r/mysqld--help-win.result:
Updated test after adding --metadata-locks-cache-size
parameter.
mysql-test/suite/sys_vars/r/metadata_locks_cache_size_basic.result:
Added test coverage for newly introduced --metadata_locks_cache_size
start-up parameter and corresponding global read-only variable.
mysql-test/suite/sys_vars/t/metadata_locks_cache_size_basic-master.opt:
Added test coverage for newly introduced --metadata_locks_cache_size
start-up parameter and corresponding global read-only variable.
mysql-test/suite/sys_vars/t/metadata_locks_cache_size_basic.test:
Added test coverage for newly introduced --metadata_locks_cache_size
start-up parameter and corresponding global read-only variable.
sql/mdl.cc:
Introduced caching of unused MDL_object_lock objects, in order to
avoid costs associated with constant creation and destruction of
such objects in single-connection workloads run in autocommit mode.
Such costs can be pretty high on systems where creation and
destruction of synchronization primitives require a system call
(e.g. Windows XP).
To implement this cache,a list of unused MDL_object_lock instances
was added to MDL_map object. Instead of being destroyed
MDL_object_lock instances are put into this list and re-used later
when creation of a new instance is required. Also added
MDL_lock::m_version counter to allow threads having outstanding
references to an MDL_object_lock instance to notice that it has
been moved to the unused objects list.
Added a global variable for a start-up parameter that limits
the size of the unused objects list.
Note that we don't cache MDL_scoped_lock objects since they
are supposed to be created only during execution of DDL
statements and therefore should not affect performance much.
sql/mdl.h:
Added a global variable for start-up parameter that limits the
size of the unused MDL_object_lock objects list and constant
for its default value.
sql/sql_plist.h:
Added I_P_List<>::pop_front() function.
sql/sys_vars.cc:
Introduced --metadata-locks-cache-size start-up parameter
for specifying size of the cache of unused MDL_object_lock
objects.
GRADUALLY IF A TRIGGER EXISTS".
This bug manifested itself in two ways:
- Firstly execution of any data-changing statement which
required prelocking (i.e. involved stored function or
trigger) as part of transaction slowed down a bit all
subsequent statements in this transaction. So performance
in transaction which periodically involved such statements
gradually degraded over time.
- Secondly execution of any data-changing statement which
required prelocking as part of transaction prevented
concurrent FLUSH TABLES WITH READ LOCK from proceeding
until the end of transaction instead of end of particular
statement.
The problem was caused by incorrect handling of metadata lock
used in FTWRL implementation for statements requiring prelocked
mode.
Each statement which changes data acquires global IX lock
with STATEMENT duration. This lock is supposed to block
concurrent FTWRL from proceeding until the statement ends.
When entering prelocked mode, durations of all metadata locks
acquired so far were changed to EXPLICIT, to prevent
substatements from releasing these locks. When prelocked mode
was left, durations of metadata locks were changed to
TRANSACTIONAL (with a few exceptions) so they can be properly
released at the end of transaction.
Unfortunately, this meant that the global IX lock blocking
FTWRL with STATEMENT duration was moved to TRANSACTIONAL
duration after execution of statement requiring prelocking.
Since each subsequent statement that required prelocking and
tried to acquire global IX lock with STATEMENT duration got
a new instance of MDL_ticket, which was later moved to
TRANSACTIONAL duration, this led to unwarranted growth of
number of tickets with TRANSACITONAL duration in this
connection's MDL_context. As result searching for other
tickets in it became slow and acquisition of other metadata
locks by this transaction started to hog CPU.
Moreover, this also meant that after execution of statement
requiring prelocking concurrent FTWRL was blocked
until the end of transaction instead of end of statement.
This patch solves this problem by not moving locks to EXPLICIT
duration when thread enters prelocked mode (unless it is a real
LOCK TABLES mode). This step turned out to be not really
necessary as substatements don't try to release metadata locks.
Consequently, the global IX lock blocking FTWRL keeps its
STATEMENT duration and is properly released at the end of
statement and the above issue goes away.
mysql-test/r/flush.result:
Added test for bug #12641342 - "61401: UPDATE PERFORMANCE
DEGRADES GRADUALLY IF A TRIGGER EXISTS".
mysql-test/t/flush.test:
Added test for bug #12641342 - "61401: UPDATE PERFORMANCE
DEGRADES GRADUALLY IF A TRIGGER EXISTS".
sql/mdl.h:
Added comment describing various types of metadata lock
duration.
sql/sql_class.cc:
Since we no longer change duration of metadata locks to EXPLICIT
when entering prelocked mode (unless it is a real LOCK TABLES)
there is no need to restore proper duration of the locks when
leaving prelocked mode.
sql/sql_class.h:
Do not change duration of metadata locks to EXPLICIT when
entering prelocking mode (unless it is a real LOCK TABLES).
This allows to avoid problems with restoring correct duration
when leaving this mode. It is possible to do this as
substatements won't release metadata locks in any case.
sql/sql_parse.cc:
Added assert checking that we won't release metadata locks
when in substatement.
test case to give valgrind warnings.
The problem was that when comparing two MDL key buffers using
memcmp(), 1 was added to the buffer length. However, this was
no longer needed since the buffer length already included the
'\0' terminator.
bug #57006 "Deadlock between HANDLER and FLUSH TABLES WITH READ
LOCK" and bug #54673 "It takes too long to get readlock for
'FLUSH TABLES WITH READ LOCK'".
The first bug manifested itself as a deadlock which occurred
when a connection, which had some table open through HANDLER
statement, tried to update some data through DML statement
while another connection tried to execute FLUSH TABLES WITH
READ LOCK concurrently.
What happened was that FTWRL in the second connection managed
to perform first step of GRL acquisition and thus blocked all
upcoming DML. After that it started to wait for table open
through HANDLER statement to be flushed. When the first connection
tried to execute DML it has started to wait for GRL/the second
connection creating deadlock.
The second bug manifested itself as starvation of FLUSH TABLES
WITH READ LOCK statements in cases when there was a constant
stream of concurrent DML statements (in two or more
connections).
This has happened because requests for protection against GRL
which were acquired by DML statements were ignoring presence of
pending GRL and thus the latter was starved.
This patch solves both these problems by re-implementing GRL
using metadata locks.
Similar to the old implementation acquisition of GRL in new
implementation is two-step. During the first step we block
all concurrent DML and DDL statements by acquiring global S
metadata lock (each DML and DDL statement acquires global IX
lock for its duration). During the second step we block commits
by acquiring global S lock in COMMIT namespace (commit code
acquires global IX lock in this namespace).
Note that unlike in old implementation acquisition of
protection against GRL in DML and DDL is semi-automatic.
We assume that any statement which should be blocked by GRL
will either open and acquires write-lock on tables or acquires
metadata locks on objects it is going to modify. For any such
statement global IX metadata lock is automatically acquired
for its duration.
The first problem is solved because waits for GRL become
visible to deadlock detector in metadata locking subsystem
and thus deadlocks like one in the first bug become impossible.
The second problem is solved because global S locks which
are used for GRL implementation are given preference over
IX locks which are acquired by concurrent DML (and we can
switch to fair scheduling in future if needed).
Important change:
FTWRL/GRL no longer blocks DML and DDL on temporary tables.
Before this patch behavior was not consistent in this respect:
in some cases DML/DDL statements on temporary tables were
blocked while in others they were not. Since the main use cases
for FTWRL are various forms of backups and temporary tables are
not preserved during backups we have opted for consistently
allowing DML/DDL on temporary tables during FTWRL/GRL.
Important change:
This patch changes thread state names which are used when
DML/DDL of FTWRL is waiting for global read lock. It is now
either "Waiting for global read lock" or "Waiting for commit
lock" depending on the stage on which FTWRL is.
Incompatible change:
To solve deadlock in events code which was exposed by this
patch we have to replace LOCK_event_metadata mutex with
metadata locks on events. As result we have to prohibit
DDL on events under LOCK TABLES.
This patch also adds extensive test coverage for interaction
of DML/DDL and FTWRL.
Performance of new and old global read lock implementations
in sysbench tests were compared. There were no significant
difference between new and old implementations.
mysql-test/include/check_ftwrl_compatible.inc:
Added helper script which allows to check that a statement is
compatible with FLUSH TABLES WITH READ LOCK.
mysql-test/include/check_ftwrl_incompatible.inc:
Added helper script which allows to check that a statement is
incompatible with FLUSH TABLES WITH READ LOCK.
mysql-test/include/handler.inc:
Adjusted test case to the fact that now DROP TABLE closes
open HANDLERs for the table to be dropped before checking
if there active FTWRL in this connection.
mysql-test/include/wait_show_condition.inc:
Fixed small error in the timeout message. The correct name
of variable used as parameter for this script is "$condition"
and not "$wait_condition".
mysql-test/r/delayed.result:
Added test coverage for scenario which triggered assert in
metadata locking subsystem.
mysql-test/r/events_2.result:
Updated test results after prohibiting event DDL operations
under LOCK TABLES.
mysql-test/r/flush.result:
Added test coverage for bug #57006 "Deadlock between HANDLER
and FLUSH TABLES WITH READ LOCK".
mysql-test/r/flush_read_lock.result:
Added test coverage for various aspects of FLUSH TABLES WITH
READ LOCK functionality.
mysql-test/r/flush_read_lock_kill.result:
Adjusted test case after replacing custom global read lock
implementation with one based on metadata locks. Use new
debug_sync point. Do not disable concurrent inserts as now
InnoDB we always use InnoDB table.
mysql-test/r/handler_innodb.result:
Adjusted test case to the fact that now DROP TABLE closes
open HANDLERs for the table to be dropped before checking
if there active FTWRL in this connection.
mysql-test/r/handler_myisam.result:
Adjusted test case to the fact that now DROP TABLE closes
open HANDLERs for the table to be dropped before checking
if there active FTWRL in this connection.
mysql-test/r/mdl_sync.result:
Adjusted test case after replacing custom global read lock
implementation with one based on metadata locks. Replaced
usage of GRL-specific debug_sync's with appropriate sync
points in MDL subsystem.
mysql-test/suite/perfschema/r/dml_setup_instruments.result:
Updated test results after removing global
COND_global_read_lock condition variable.
mysql-test/suite/perfschema/r/func_file_io.result:
Ensure that this test doesn't affect subsequent tests.
At the end of its execution enable back P_S instrumentation
which this test disables at some point.
mysql-test/suite/perfschema/r/func_mutex.result:
Ensure that this test doesn't affect subsequent tests.
At the end of its execution enable back P_S instrumentation
which this test disables at some point.
mysql-test/suite/perfschema/r/global_read_lock.result:
Adjusted test case to take into account that new GRL
implementation is based on MDL.
mysql-test/suite/perfschema/r/server_init.result:
Adjusted test case after replacing custom global read
lock implementation with one based on MDL and replacing
LOCK_event_metadata mutex with metadata lock.
mysql-test/suite/perfschema/t/func_file_io.test:
Ensure that this test doesn't affect subsequent tests.
At the end of its execution enable back P_S instrumentation
which this test disables at some point.
mysql-test/suite/perfschema/t/func_mutex.test:
Ensure that this test doesn't affect subsequent tests.
At the end of its execution enable back P_S instrumentation
which this test disables at some point.
mysql-test/suite/perfschema/t/global_read_lock.test:
Adjusted test case to take into account that new GRL
implementation is based on MDL.
mysql-test/suite/perfschema/t/server_init.test:
Adjusted test case after replacing custom global read
lock implementation with one based on MDL and replacing
LOCK_event_metadata mutex with metadata lock.
mysql-test/suite/rpl/r/rpl_tmp_table_and_DDL.result:
Updated test results after prohibiting event DDL under
LOCK TABLES.
mysql-test/t/delayed.test:
Added test coverage for scenario which triggered assert in
metadata locking subsystem.
mysql-test/t/events_2.test:
Updated test case after prohibiting event DDL operations
under LOCK TABLES.
mysql-test/t/flush.test:
Added test coverage for bug #57006 "Deadlock between HANDLER
and FLUSH TABLES WITH READ LOCK".
mysql-test/t/flush_block_commit.test:
Adjusted test case after changing thread state name which
is used when COMMIT waits for FLUSH TABLES WITH READ LOCK
from "Waiting for release of readlock" to "Waiting for commit
lock".
mysql-test/t/flush_block_commit_notembedded.test:
Adjusted test case after changing thread state name which is
used when DML waits for FLUSH TABLES WITH READ LOCK. Now we
use "Waiting for global read lock" in this case.
mysql-test/t/flush_read_lock.test:
Added test coverage for various aspects of FLUSH TABLES WITH
READ LOCK functionality.
mysql-test/t/flush_read_lock_kill-master.opt:
We no longer need to use make_global_read_lock_block_commit_loop
debug tag in this test. Instead we rely on an appropriate
debug_sync point in MDL code.
mysql-test/t/flush_read_lock_kill.test:
Adjusted test case after replacing custom global read lock
implementation with one based on metadata locks. Use new
debug_sync point. Do not disable concurrent inserts as now
InnoDB we always use InnoDB table.
mysql-test/t/lock_multi.test:
Adjusted test case after changing thread state names which
are used when DML or DDL waits for FLUSH TABLES WITH READ
LOCK to "Waiting for global read lock".
mysql-test/t/mdl_sync.test:
Adjusted test case after replacing custom global read lock
implementation with one based on metadata locks. Replaced
usage of GRL-specific debug_sync's with appropriate sync
points in MDL subsystem. Updated thread state names which
are used when DDL waits for FTWRL.
mysql-test/t/trigger_notembedded.test:
Adjusted test case after changing thread state names which
are used when DML or DDL waits for FLUSH TABLES WITH READ
LOCK to "Waiting for global read lock".
sql/event_data_objects.cc:
Removed Event_queue_element::status/last_executed_changed
members and Event_queue_element::update_timing_fields()
method. We no longer use this class for updating mysql.events
once event is chosen for execution. Accesses to instances of
this class in scheduler thread require protection by
Event_queue::LOCK_event_queue mutex and we try to avoid
updating table while holding this lock.
sql/event_data_objects.h:
Removed Event_queue_element::status/last_executed_changed
members and Event_queue_element::update_timing_fields()
method. We no longer use this class for updating mysql.events
once event is chosen for execution. Accesses to instances of
this class in scheduler thread require protection by
Event_queue::LOCK_event_queue mutex and we try to avoid
updating table while holding this lock.
sql/event_db_repository.cc:
- Changed Event_db_repository methods to not release all
metadata locks once they are done updating mysql.events
table. This allows to keep metadata lock protecting
against GRL and lock protecting particular event around
until corresponding DDL statement is written to the binary
log.
- Removed logic for conditional update of "status" and
"last_executed" fields from update_timing_fields_for_event()
method. In the only case when this method is called now
"last_executed" is always modified and tracking change
of "status" is too much hassle.
sql/event_db_repository.h:
Removed logic for conditional update of "status" and
"last_executed" fields from Event_db_repository::
update_timing_fields_for_event() method.
In the only case when this method is called now "last_executed"
is always modified and tracking change of "status" field is
too much hassle.
sql/event_queue.cc:
Changed event scheduler code not to update mysql.events
table while holding Event_queue::LOCK_event_queue mutex.
Doing so led to a deadlock with a new GRL implementation.
This deadlock didn't occur with old implementation due to
fact that code acquiring protection against GRL ignored
pending GRL requests (which lead to GRL starvation).
One of goals of new implementation is to disallow GRL
starvation and so we have to solve problem with this
deadlock in a different way.
sql/events.cc:
Changed methods of Events class to acquire protection
against GRL while perfoming DDL statement and keep it
until statement is written to the binary log.
Unfortunately this step together with new GRL implementation
exposed deadlock involving Events::LOCK_event_metadata
and GRL. To solve it Events::LOCK_event_metadata mutex was
replaced with a metadata lock on event. As a side-effect
events DDL has to be prohibited under LOCK TABLES even in
cases when mysql.events table was explicitly locked for
write.
sql/events.h:
Replaced Events::LOCK_event_metadata mutex with a metadata
lock on event.
sql/ha_ndbcluster.cc:
Updated code after replacing custom global read lock
implementation with one based on MDL. Since MDL subsystem
should now be able to detect deadlocks involving metadata
locks and GRL there is no need for special handling of
active GRL.
sql/handler.cc:
Replaced custom implementation of global read lock with
one based on metadata locks. Consequently when doing
commit instead of calling method of Global_read_lock
class to acquire protection against GRL we simply acquire
IX in COMMIT namespace.
sql/lock.cc:
Replaced custom implementation of global read lock with
one based on metadata locks. This step allows to expose
wait for GRL to deadlock detector of MDL subsystem and
thus succesfully resolve deadlocks similar to one behind
bug #57006 "Deadlock between HANDLER and FLUSH TABLES
WITH READ LOCK". It also solves problem with GRL starvation
described in bug #54673 "It takes too long to get readlock
for 'FLUSH TABLES WITH READ LOCK'" since metadata locks used
by GRL give preference to FTWRL statement instead of DML
statements (if needed in future this can be changed to
fair scheduling).
Similar to old implementation of acquisition of GRL is
two-step. During the first step we block all concurrent
DML and DDL statements by acquiring global S metadata lock
(each DML and DDL statement acquires global IX lock for
its duration). During the second step we block commits by
acquiring global S lock in COMMIT namespace (commit code
acquires global IX lock in this namespace).
Note that unlike in old implementation acquisition of
protection against GRL in DML and DDL is semi-automatic.
We assume that any statement which should be blocked by GRL
will either open and acquires write-lock on tables or acquires
metadata locks on objects it is going to modify. For any such
statement global IX metadata lock is automatically acquired
for its duration.
To support this change:
- Global_read_lock::lock/unlock_global_read_lock and
make_global_read_lock_block_commit methods were changed
accordingly.
- Global_read_lock::wait_if_global_read_lock() and
start_waiting_global_read_lock() methods were dropped.
It is now responsibility of code acquiring metadata locks
opening tables to acquire protection against GRL by
explicitly taking global IX lock with statement duration.
- Global variables, mutex and condition variable used by
old implementation was removed.
- lock_routine_name() was changed to use statement duration for
its global IX lock. It was also renamed to lock_object_name()
as it now also used to take metadata locks on events.
- Global_read_lock::set_explicit_lock_duration() was added which
allows not to release locks used for GRL when leaving prelocked
mode.
sql/lock.h:
- Renamed lock_routine_name() to lock_object_name() and changed
its signature to allow its usage for events.
- Removed broadcast_refresh() function. It is no longer needed
with new GRL implementation.
sql/log_event.cc:
Release metadata locks with statement duration at the end
of processing legacy event for LOAD DATA. This ensures that
replication thread processing such event properly releases
its protection against global read lock.
sql/mdl.cc:
Changed MDL subsystem to support new MDL-based implementation
of global read lock.
Added COMMIT and EVENTS namespaces for metadata locks. Changed
thread state name for GLOBAL namespace to "Waiting for global
read lock".
Optimized MDL_map::find_or_insert() method to avoid taking
m_mutex mutex when looking up MDL_lock objects for GLOBAL
or COMMIT namespaces. We keep pre-created MDL_lock objects
for these namespaces around and simply return pointers to
these global objects when needed.
Changed MDL_lock/MDL_scoped_lock to properly handle
notification of insert delayed handler threads when FTWRL
takes global S lock.
Introduced concept of lock duration. In addition to locks with
transaction duration which work in the way which is similar to
how locks worked before (i.e. they are released at the end of
transaction), locks with statement and explicit duration were
introduced.
Locks with statement duration are automatically released at the
end of statement. Locks with explicit duration require explicit
release and obsolete concept of transactional sentinel.
* Changed MDL_request and MDL_ticket classes to support notion
of duration.
* Changed MDL_context to keep locks with different duration in
different lists. Changed code handling ticket list to take
this into account.
* Changed methods responsible for releasing locks to take into
account duration of tickets. Particularly public
MDL_context::release_lock() method now only can release
tickets with explicit duration (there is still internal
method which allows to specify duration). To release locks
with statement or transaction duration one have to use
release_statement/transactional_locks() methods.
* Concept of savepoint for MDL subsystem now has to take into
account locks with statement duration. Consequently
MDL_savepoint class was introduced and methods working with
savepoints were updated accordingly.
* Added methods which allow to set duration for one or all
locks in the context.
sql/mdl.h:
Changed MDL subsystem to support new MDL-based implementation
of global read lock.
Added COMMIT and EVENTS namespaces for metadata locks.
Introduced concept of lock duration. In addition to locks with
transaction duration which work in the way which is similar to
how locks worked before (i.e. they are released at the end of
transaction), locks with statement and explicit duration were
introduced.
Locks with statement duration are automatically released at the
end of statement. Locks with explicit duration require explicit
release and obsolete concept of transactional sentinel.
* Changed MDL_request and MDL_ticket classes to support notion
of duration.
* Changed MDL_context to keep locks with different duration in
different lists. Changed code handling ticket list to take
this into account.
* Changed methods responsible for releasing locks to take into
account duration of tickets. Particularly public
MDL_context::release_lock() method now only can release
tickets with explicit duration (there is still internal
method which allows to specify duration). To release locks
with statement or transaction duration one have to use
release_statement/transactional_locks() methods.
* Concept of savepoint for MDL subsystem now has to take into
account locks with statement duration. Consequently
MDL_savepoint class was introduced and methods working with
savepoints were updated accordingly.
* Added methods which allow to set duration for one or all
locks in the context.
sql/mysqld.cc:
Removed global mutex and condition variables which were used
by old implementation of GRL.
Also we no longer need to initialize Events::LOCK_event_metadata
mutex as it was replaced with metadata locks on events.
sql/mysqld.h:
Removed global variable, mutex and condition variables which
were used by old implementation of GRL.
sql/rpl_rli.cc:
When slave thread closes tables which were open for handling
of RBR events ensure that it releases global IX lock which
was acquired as protection against GRL.
sql/sp.cc:
Adjusted code to the new signature of lock_object/routine_name(),
to the fact that one now needs specify duration of lock when
initializing MDL_request and to the fact that savepoints for MDL
subsystem are now represented by MDL_savepoint class.
sql/sp_head.cc:
Ensure that statements in stored procedures release statement
metadata locks and thus release their protectiong against GRL
in proper moment in time.
Adjusted code to the fact that one now needs specify duration
of lock when initializing MDL_request.
sql/sql_admin.cc:
Adjusted code to the fact that one now needs specify duration
of lock when initializing MDL_request.
sql/sql_base.cc:
- Implemented support for new approach to acquiring protection
against global read lock. We no longer acquire such protection
explicitly on the basis of statement flags. Instead we always
rely on code which is responsible for acquiring metadata locks
on object to be changed acquiring this protection. This is
achieved by acquiring global IX metadata lock with statement
duration. Code doing this also responsible for checking that
current connection has no active GRL by calling an
Global_read_lock::can_acquire_protection() method.
Changed code in open_table() and lock_table_names()
accordingly.
Note that as result of this change DDL and DML on temporary
tables is always compatible with GRL (before it was
incompatible in some cases and compatible in other cases).
- To speed-up code acquiring protection against GRL introduced
m_has_protection_against_grl member in Open_table_context
class. It indicates that protection was already acquired
sometime during open_tables() execution and new attempts
can be skipped.
- Thanks to new GRL implementation calls to broadcast_refresh()
became unnecessary and were removed.
- Adjusted code to the fact that one now needs specify duration
of lock when initializing MDL_request and to the fact that
savepoints for MDL subsystem are now represented by
MDL_savepoint class.
sql/sql_base.h:
Adjusted code to the fact that savepoints for MDL subsystem are
now represented by MDL_savepoint class.
Also introduced Open_table_context::m_has_protection_against_grl
member which allows to avoid acquiring protection against GRL
while opening tables if such protection was already acquired.
sql/sql_class.cc:
Changed THD::leave_locked_tables_mode() after transactional
sentinel for metadata locks was obsoleted by introduction of
locks with explicit duration.
sql/sql_class.h:
- Adjusted code to the fact that savepoints for MDL subsystem
are now represented by MDL_savepoint class.
- Changed Global_read_lock class according to changes in
global read lock implementation:
* wait_if_global_read_lock and start_waiting_global_read_lock
are now gone. Instead code needing protection against GRL
has to acquire global IX metadata lock with statement
duration itself. To help it new can_acquire_protection()
was introduced. Also as result of the above change
m_protection_count member is gone too.
* Added m_mdl_blocks_commits_lock member to store metadata
lock blocking commits.
* Adjusted code to the fact that concept of transactional
sentinel was obsoleted by concept of lock duration.
- Removed CF_PROTECT_AGAINST_GRL flag as it is no longer
necessary. New GRL implementation acquires protection
against global read lock automagically when statement
acquires metadata locks on tables or other objects it
is going to change.
sql/sql_db.cc:
Adjusted code to the fact that one now needs specify duration
of lock when initializing MDL_request.
sql/sql_handler.cc:
Removed call to broadcast_refresh() function. It is no longer
needed with new GRL implementation.
Adjusted code after introducing duration concept for metadata
locks. Particularly to the fact transactional sentinel was
replaced with explicit duration.
sql/sql_handler.h:
Renamed mysql_ha_move_tickets_after_trans_sentinel() to
mysql_ha_set_explicit_lock_duration() after transactional
sentinel was obsoleted by locks with explicit duration.
sql/sql_insert.cc:
Adjusted code handling delaying inserts after switching to
new GRL implementation. Now connection thread initiating
delayed insert has to acquire global IX lock in addition
to metadata lock on table being inserted into. This IX lock
protects against GRL and similarly to SW lock on table being
inserted into has to be passed to handler thread in order to
avoid deadlocks.
sql/sql_lex.cc:
LEX::protect_against_global_read_lock member is no longer
necessary since protection against GRL is automatically
taken by code acquiring metadata locks/opening tables.
sql/sql_lex.h:
LEX::protect_against_global_read_lock member is no longer
necessary since protection against GRL is automatically
taken by code acquiring metadata locks/opening tables.
sql/sql_parse.cc:
- Implemented support for new approach to acquiring protection
against global read lock. We no longer acquire such protection
explicitly on the basis of statement flags. Instead we always
rely on code which is responsible for acquiring metadata locks
on object to be changed acquiring this protection. This is
achieved by acquiring global IX metadata lock with statement
duration. This lock is automatically released at the end of
statement execution.
- Changed implementation of CREATE/DROP PROCEDURE/FUNCTION not
to release metadata locks and thus protection against of GRL
in the middle of statement execution.
- Adjusted code to the fact that one now needs specify duration
of lock when initializing MDL_request and to the fact that
savepoints for MDL subsystem are now represented by
MDL_savepoint class.
sql/sql_prepare.cc:
Adjusted code to the to the fact that savepoints for MDL
subsystem are now represented by MDL_savepoint class.
sql/sql_rename.cc:
With new GRL implementation there is no need to explicitly
acquire protection against GRL before renaming tables.
This happens automatically in code which acquires metadata
locks on tables being renamed.
sql/sql_show.cc:
Adjusted code to the fact that one now needs specify duration
of lock when initializing MDL_request and to the fact that
savepoints for MDL subsystem are now represented by
MDL_savepoint class.
sql/sql_table.cc:
- With new GRL implementation there is no need to explicitly
acquire protection against GRL before dropping tables.
This happens automatically in code which acquires metadata
locks on tables being dropped.
- Changed mysql_alter_table() not to release lock on new table
name explicitly and to rely on automatic release of locks
at the end of statement instead. This was necessary since
now MDL_context::release_lock() is supported only for locks
for explicit duration.
sql/sql_trigger.cc:
With new GRL implementation there is no need to explicitly
acquire protection against GRL before changing table triggers.
This happens automatically in code which acquires metadata
locks on tables which triggers are to be changed.
sql/sql_update.cc:
Fix bug exposed by GRL testing. During prepare phase acquire
only S metadata locks instead of SW locks to keep prepare of
multi-UPDATE compatible with concurrent LOCK TABLES WRITE
and global read lock.
sql/sql_view.cc:
With new GRL implementation there is no need to explicitly
acquire protection against GRL before creating view.
This happens automatically in code which acquires metadata
lock on view to be created.
sql/sql_yacc.yy:
LEX::protect_against_global_read_lock member is no longer
necessary since protection against GRL is automatically
taken by code acquiring metadata locks/opening tables.
sql/table.cc:
Adjusted code to the fact that one now needs specify duration
of lock when initializing MDL_request.
sql/table.h:
Adjusted code to the fact that one now needs specify duration
of lock when initializing MDL_request.
sql/transaction.cc:
Replaced custom implementation of global read lock with
one based on metadata locks. Consequently when doing
commit instead of calling method of Global_read_lock
class to acquire protection against GRL we simply acquire
IX in COMMIT namespace.
Also adjusted code to the fact that MDL savepoint is now
represented by MDL_savepoint class.
MDL deadlock detector".
Deadlock could have occurred when workload containing mix
of DML, DDL and FLUSH TABLES statements affecting same
set of tables was executed in heavily concurrent environment.
This deadlock occurred when several connections tried to
perform deadlock detection in metadata locking subsystem.
The first connection started traversing wait-for graph,
encountered sub-graph representing wait for flush, acquired
LOCK_open and dived into sub-graph inspection. When it has
encounterd sub-graph corresponding to wait for metadata lock
and blocked while trying to acquire rd-lock on
MDL_lock::m_rwlock (*) protecting this subgraph, since some
other thread had wr-lock on it. When this wr-lock was released
it could have happened (if there was other pending wr-lock
against this rwlock) that rd-lock from the first connection
was left unsatisfied but at the same time new rd-lock request
from the second connection sneaked in and was satisfied (for
this to be possible second rd- request should come exactly
after wr-lock is released but before pending wr-lock manages
to grab rwlock, which is possible both on Linux and in our
own rwlock implementation). If this second connection
continued traversing wait-for graph and encountered sub-graph
representing wait for flush it tried to acquire LOCK_open
and thus deadlock was created.
This patch tries to workaround this problem but not allowing
deadlock detector to lock LOCK_open mutex if some other thread
doing deadlock detection already owns it and current search
depth is greater than 0. Instead deadlock is reported.
Other possible solutions are either known to have negative
effects on performance or require much more time for proper
implementation and testing.
No test case is provided as this bug is very hard to repeat
in MTR environment but is repeatable with the help of RQG
tests.
sql/mdl.cc:
Moved Deadlock_detection_visitor::m_current_search_depth to
parent class to make it available in
TABLE_SHARE::visit_subgraph().
Added MDL_wait_for_graph_visitor::abort_traversal() method
which allows to abort traversal of a wait-for graph and
report a deadlock.
sql/mdl.h:
Moved Deadlock_detection_visitor::m_current_search_depth to
parent class to make it available in
TABLE_SHARE::visit_subgraph().
Added MDL_wait_for_graph_visitor::abort_traversal() method
which allows to abort traversal of a wait-for graph and
report a deadlock.
sql/sql_base.cc:
Added dd_owns_lock_open counter and mutex protecting it to
track number of connections which do deadlock detection and
own or try to acquire LOCK_open.
sql/sql_base.h:
Added dd_owns_lock_open counter and mutex protecting it to
track number of connections which do deadlock detection and
own or try to acquire LOCK_open.
sql/table.cc:
Workaround bug #56405 but not allowing MDL deadlock detector
to lock LOCK_open mutex if some other thread doing deadlock
detection already owns it and current search depth is greater
than 0. Instead report deadlock.
FLUSH TABLES <list> WITH READ LOCK are incompatible" to
be pushed as separate patch.
Replaced thread state name "Waiting for table", which was
used by threads waiting for a metadata lock or table flush,
with a set of names which better reflect types of resources
being waited for.
Also replaced "Table lock" thread state name, which was used
by threads waiting on thr_lock.c table level lock, with more
elaborate "Waiting for table level lock", to make it
more consistent with other thread state names.
Updated test cases and their results according to these
changes.
Fixed sys_vars.query_cache_wlock_invalidate_func test to not
to wait for timeout of wait_condition.inc script.
mysql-test/r/query_cache.result:
Added test coverage for query_cache_wlock_invalidate
behavior for implicitly locked tables.
mysql-test/suite/sys_vars/r/query_cache_wlock_invalidate_func.result:
Fixed sys_vars.query_cache_wlock_invalidate_func test to not
to wait for timeout of wait_condition.inc script. Reverted
changes to test which introduced timeout and replaced waiting
condition with a more appropriate one.
Test coverage for query_cache_wlock_invalidate behavior for
implicitly locked tables was added to query_cache.test.
mysql-test/suite/sys_vars/t/query_cache_wlock_invalidate_func.test:
Fixed sys_vars.query_cache_wlock_invalidate_func test to not
to wait for timeout of wait_condition.inc script. Reverted
changes to test which introduced timeout and replaced waiting
condition with a more appropriate one.
Test coverage for query_cache_wlock_invalidate behavior for
implicitly locked tables was added to query_cache.test.
mysql-test/t/query_cache.test:
Added test coverage for query_cache_wlock_invalidate
behavior for implicitly locked tables.
mysys/thr_lock.c:
Replaced "Table lock" thread state name, which was used by
threads waiting on thr_lock.c table level lock, with more
elaborate "Waiting for table level lock", to make it
consistent with thread state names which are used while
waiting for metadata locks and table flush.
sql/mdl.cc:
Replaced thread state name "Waiting for table", which was
used by threads waiting for a metadata lock or table flush,
with a set of names which better reflect types of resources
being waited for.
To implement this:
- Adjusted MDL_wait::timed_wait() to take thread state name
as parameter.
- Introduced method of MDL_key class which allows to get
thread state name to be used while waiting for resource
corresponding to the key and changed code to use it.
Added array translating namespaces to thread state names
as part of this change.
sql/mdl.h:
To implement this:
- Adjusted MDL_wait::timed_wait() to take thread state name
as parameter.
- Introduced method of MDL_key class which allows to get
thread state name to be used while waiting for resource
corresponding to the key and changed code to use it.
Added array translating namespaces to thread state names
as part of this change.
sql/sql_base.cc:
Replaced thread state name "Waiting for table", which was
used by threads waiting for table flush, with a more elaborate
"Waiting for table flush".
TABLES <list> WITH READ LOCK are incompatible".
The problem was that FLUSH TABLES <list> WITH READ LOCK
which was issued when other connection has acquired global
read lock using FLUSH TABLES WITH READ LOCK was blocked
and has to wait until global read lock is released.
This issue stemmed from the fact that FLUSH TABLES <list>
WITH READ LOCK implementation has acquired X metadata locks
on tables to be flushed. Since these locks required acquiring
of global IX lock this statement was incompatible with global
read lock.
This patch addresses problem by using SNW metadata type of
lock for tables to be flushed by FLUSH TABLES <list> WITH
READ LOCK. It is OK to acquire them without global IX lock
as long as we won't try to upgrade those locks. Since SNW
locks allow concurrent statements using same table FLUSH
TABLE <list> WITH READ LOCK now has to wait until old
versions of tables to be flushed go away after acquiring
metadata locks. Since such waiting can lead to deadlock
MDL deadlock detector was extended to take into account
waits for flush and resolve such deadlocks.
As a bonus code in open_tables() which was responsible for
waiting old versions of tables to go away was refactored.
Now when we encounter old version of table in open_table()
we don't back-off and wait for all old version to go away,
but instead wait for this particular table to be flushed.
Such approach supported by deadlock detection should reduce
number of scenarios in which FLUSH TABLES aborts concurrent
multi-statement transactions.
Note that active FLUSH TABLES <list> WITH READ LOCK still
blocks concurrent FLUSH TABLES WITH READ LOCK statement
as the former keeps tables open and thus prevents the
latter statement from doing flush.
mysql-test/include/handler.inc:
Adjusted test case after changing status which is set
when FLUSH TABLES waits for tables to be flushed from
"Flushing tables" to "Waiting for table".
mysql-test/r/flush.result:
Added test which checks that "flush tables <list> with
read lock" is compatible with active "flush tables with
read lock" but not vice-versa. This test also covers
bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES
<list> WITH READ LOCK are incompatible".
mysql-test/r/mdl_sync.result:
Added scenarios in which wait for table to be flushed
causes deadlocks to the coverage of MDL deadlock detector.
mysql-test/suite/perfschema/r/dml_setup_instruments.result:
Adjusted test results after removal of COND_refresh
condition variable.
mysql-test/suite/perfschema/r/server_init.result:
Adjusted test and its results after removal of COND_refresh
condition variable.
mysql-test/suite/perfschema/t/server_init.test:
Adjusted test and its results after removal of COND_refresh
condition variable.
mysql-test/t/flush.test:
Added test which checks that "flush tables <list> with
read lock" is compatible with active "flush tables with
read lock" but not vice-versa. This test also covers
bug #52044 "FLUSH TABLES WITH READ LOCK and FLUSH TABLES
<list> WITH READ LOCK are incompatible".
mysql-test/t/kill.test:
Adjusted test case after changing status which is set
when FLUSH TABLES waits for tables to be flushed from
"Flushing tables" to "Waiting for table".
mysql-test/t/lock_multi.test:
Adjusted test case after changing status which is set
when FLUSH TABLES waits for tables to be flushed from
"Flushing tables" to "Waiting for table".
mysql-test/t/mdl_sync.test:
Added scenarios in which wait for table to be flushed
causes deadlocks to the coverage of MDL deadlock detector.
sql/ha_ndbcluster.cc:
Adjusted code after adding one more parameter for
close_cached_tables() call - timeout for waiting for
table to be flushed.
sql/ha_ndbcluster_binlog.cc:
Adjusted code after adding one more parameter for
close_cached_tables() call - timeout for waiting for
table to be flushed.
sql/lock.cc:
Removed COND_refresh condition variable. See comment
for sql_base.cc for details.
sql/mdl.cc:
Now MDL deadlock detector takes into account information
about waits for table flushes when searching for deadlock.
To implement this change:
- Declaration of enum_deadlock_weight and
Deadlock_detection_visitor were moved to mdl.h header
to make them available to the code in table.cc which
implements deadlock detector traversal through edges
of waiters graph representing waiting for flush.
- Since now MDL_context may wait not only for metadata
lock but also for table to be flushed an abstract
Wait_for_edge class was introduced. Its descendants
MDL_ticket and Flush_ticket incapsulate specifics
of inspecting waiters graph when following through
edge representing wait of particular type.
We no longer require global IX metadata lock when acquiring
SNW or SNRW locks. Such locks are needed only when metadata
locks of these types are upgraded to X locks. This allows
to use SNW locks in FLUSH TABLES <list> WITH READ LOCK
implementation and keep the latter compatible with global
read lock.
sql/mdl.h:
Now MDL deadlock detector takes into account information
about waits for table flushes when searching for deadlock.
To implement this change:
- Declaration of enum_deadlock_weight and
Deadlock_detection_visitor were moved to mdl.h header
to make them available to the code in table.cc which
implements deadlock detector traversal through edges
of waiters graph representing waiting for flush.
- Since now MDL_context may wait not only for metadata
lock but also for table to be flushed an abstract
Wait_for_edge class was introduced. Its descendants
MDL_ticket and Flush_ticket incapsulate specifics
of inspecting waiters graph when following through
edge representing wait of particular type.
- Deadlock_detection_visitor now has m_table_shares_visited
member which allows to support recursive locking for
LOCK_open. This is required when deadlock detector
inspects waiters graph which contains several edges
representing waits for flushes or needs to come through
the such edge more than once.
sql/mysqld.cc:
Removed COND_refresh condition variable. See comment
for sql_base.cc for details.
sql/mysqld.h:
Removed COND_refresh condition variable. See comment
for sql_base.cc for details.
sql/sql_base.cc:
Changed approach to how threads are waiting for table
to be flushed. Now thread that wants to wait for old
table to go away subscribes for notification by adding
Flush_ticket to table's share and waits using
MDL_context::m_wait object. Once table gets flushed
(i.e. all tables are closed and table share is ready
to be destroyed) all such waiters are notified
individually.
Thanks to this change MDL deadlock detector can take
such waits into account.
To implement this/as result of this change:
- tdc_wait_for_old_versions() was replaced with
tdc_wait_for_old_version() which waits for individual
old share to go away and which is called by open_table()
after finding out that share is outdated. We don't
need to perform back-off before such waiting thanks
to the fact that deadlock detector now sees such waits.
- As result Open_table_ctx::m_mdl_requests became
unnecessary and was removed. We no longer allocate
copies of MDL_request objects on MEM_ROOT when
MYSQL_OPEN_FORCE_SHARED/SHARED_HIGH_PRIO flags are
in effect.
- close_cached_tables() and tdc_wait_for_old_version()
share code which implements waiting for share to be
flushed - the both use TABLE_SHARE::wait_until_flush()
method. Thanks to this close_cached_tables() supports
timeouts and has extra parameter for this.
- Open_table_context::OT_MDL_CONFLICT enum element was
renamed to OT_CONFLICT as it is now also used in cases
when back-off is required to resolve deadlock caused
by waiting for flush and not metadata lock.
- In cases when we discover that current connection tries
to open tables from different generation we now simply
back-off and restart process of opening tables. To
support this Open_table_context::OT_REOPEN_TABLES enum
element was added.
- COND_refresh condition variable became unnecessary and
was removed.
- mysql_notify_thread_having_shared_lock() no longer wakes
up connections waiting for flush as all such connections
can be waken up by deadlock detector if necessary.
sql/sql_base.h:
- close_cached_tables() now has one more parameter -
timeout for waiting for table to be flushed.
- Open_table_context::OT_MDL_CONFLICT enum element was
renamed to OT_CONFLICT as it is now also used in cases
when back-off is required to resolve deadlock caused
by waiting for flush and not metadata lock.
Added new OT_REOPEN_TABLES enum element to be used in
cases when we need to restart open tables process even
in the middle of transaction.
- Open_table_ctx::m_mdl_requests became unnecessary and
was removed.
sql/sql_class.h:
Added assert ensuring that we won't use LOCK_open mutex
with THD::enter_cond(). Otherwise deadlocks can arise in
MDL deadlock detector.
sql/sql_parse.cc:
Changed FLUSH TABLES <list> WITH READ LOCK to take SNW
metadata locks instead of X locks on tables to be flushed.
Since we no longer require global IX lock to be taken
when SNW locks are taken this makes this statement
compatible with FLUSH TABLES WITH READ LOCK statement.
Since SNW locks allow other connections to have table
opened FLUSH TABLES <list> WITH READ LOCK now has to
wait during open_tables() for old version to go away.
Such waits can lead to deadlocks which will be detected
by MDL deadlock detector which now takes waits for table
to be flushed into account.
Also adjusted code after adding one more parameter for
close_cached_tables() call - timeout for waiting for
table to be flushed.
sql/sql_yacc.yy:
FLUSH TABLES <list> WITH READ LOCK now needs only SNW
metadata locks on tables.
sql/sys_vars.cc:
Adjusted code after adding one more parameter for
close_cached_tables() call - timeout for waiting for
table to be flushed.
sql/table.cc:
Implemented new approach to how threads are waiting for
table to be flushed. Now thread that wants to wait for
old table to go away subscribes for notification by
adding Flush_ticket to table's share and waits using
MDL_context::m_wait object. Once table gets flushed
(i.e. all tables are closed and table share is ready
to be destroyed) all such waiters are notified
individually. This change allows to make such waits
visible inside of MDL deadlock detector.
To do it:
- Added list of waiters/Flush_tickets to TABLE_SHARE
class.
- Changed free_table_share() to postpone freeing of
share memory until last waiter goes away and to
wake up subscribed waiters.
- Added TABLE_SHARE::wait_until_flushed() method which
implements subscription to the list of waiters for
table to be flushed and waiting for this event.
Implemented interface which allows to expose waits for
flushes to MDL deadlock detector:
- Introduced Flush_ticket class a descendant of
Wait_for_edge class.
- Added TABLE_SHARE::find_deadlock() method which allows
deadlock detector to find out what contexts are still
using old version of table in question (i.e. to find
out what contexts are waited for by owner of
Flush_ticket).
sql/table.h:
In order to support new strategy of waiting for table flush
(see comment for table.cc for details) added list of
waiters/Flush_tickets to TABLE_SHARE class.
Implemented interface which allows to expose waits for
flushes to MDL deadlock detector:
- Introduced Flush_ticket class a descendant of
Wait_for_edge class.
- Added TABLE_SHARE::find_deadlock() method which allows
deadlock detector to find out what contexts are still
using old version of table in question (i.e. to find
out what contexts are waited for by owner of
Flush_ticket).
The problem was that a statement could cause an assert if it was aborted by
KILL QUERY while it waited on a metadata lock. This assert checks that a
statement either sends OK or an error to the client. If the bug was triggered
on release builds, it caused OK to be sent to the client instead of
ER_QUERY_INTERRUPTED.
The root cause of the problem was that there are two separate ways to tell if a
statement is killed: thd->killed and mysys_var->abort. KILL QUERY causes both
to be set, thd->killed before mysys_var->abort. Also, both values are reset
at the end of statement execution. This means that it is possible for
KILL QUERY to first set thd->killed, then have the killed statement reset
both thd->killed and mysys_var->abort and finally have KILL QUERY set
mysys_var->abort. This means that the connection with the killed statement
will start executing the next statement with the two values out of sync - i.e.
thd->killed not set but mysys_var->abort set.
Since mysys_var->abort is used to check if a wait for a metadata lock should
be aborted, the next statement would immediately abort any such waiting.
When waiting is aborted, no OK message is sent and thd->killed is checked to
see if ER_QUERY_INTERRUPTED should be sent to the client. But since
the->killed had been reset, neither OK nor an error message was sent to the
client. This then triggered the assert.
This patch fixes the problem by changing the metadata lock waiting code to
check thd->killed.
No test case added as reproducing the assert is dependent on very exact timing
of two (or more) threads. The patch has been checked using RQG and the grammar
posted on the bug report.
WITH READ LOCK and FLUSH TABLES <list> WITH READ LOCK are
incompatible", which adds information about waits caused by
FLUSH TABLES statement to deadlock detector in MDL subsystem.
Remove API supporting caching of pointers to TABLE_SHARE
object in MDL subsystem and all code related to it.
The problem was that locking requirements of code
implementing this API conflicted with locking requirements
of code which adds information about waits caused by flushes
to deadlock detector in MDL subsystem (the former needed to
lock LOCK_open or its future equivalent while having
write-lock on MDL_lock's rwlock, and the latter needs to be
able to read-lock MDL_lock rwlock while owning LOCK_open or
its future equivalent).
Since caching of pointers to TABLE_SHARE objects in MDL
subsystem didn't bring expected performance benefits we
decided to remove caching API rather than try to come up
with some complex solution for this problem.
DATABASE with open HANDLER"
Remove LOCK_create_db, database name locks, and use metadata locks instead.
This exposes CREATE/DROP/ALTER DATABASE statements to the graph-based
deadlock detector in MDL, and paves the way for a safe, deadlock-free
implementation of RENAME DATABASE.
Database DDL statements will now take exclusive metadata locks on
the database name, while table/view/routine DDL statements take
intention exclusive locks on the database name. This prevents race
conditions between database DDL and table/view/routine DDL.
(e.g. DROP DATABASE with concurrent CREATE/ALTER/DROP TABLE)
By adding database name locks, this patch implements
WL#4450 "DDL locking: CREATE/DROP DATABASE must use database locks" and
WL#4985 "DDL locking: namespace/hierarchical locks".
The patch also changes code to use init_one_table() where appropriate.
The new lock_table_names() function requires TABLE_LIST::db_length to
be set correctly, and this is taken care of by init_one_table().
This patch also adds a simple template to help work with
the mysys HASH data structure.
Most of the patch was written by Konstantin Osipov.
subsystem. Fix a number of caveates that the previous
implementation suffered from, including unprotected
access to shared data and lax resource accounting
(share->ref_count) that could lead to deadlocks.
The new implementation still suffers from a number
of potential deadlocks in some edge cases, and this is
still not enabled by default. Especially since performance
testing has shown that it gives only marginable (not even
exceeding measuring accuracy) improvements.
@todo:
- Remove calls to close_cached_tables() with REFRESH_FAST,
and have_lock, because they break the MDL cache.
- rework FLUSH TABLES <list> to not use close_cached_tables()
- make sure that whenever we set TABLE_SHARE::version to
0 we free MDL cache references to it.
sql/mdl.cc:
We may cache references to TABLE_SHARE objects in
MDL_lock objects for tables. Create a separate
MDL_lock class to represent a table.
sql/mdl.h:
Adjust the MDL caching API to avoid races.
sql/sql_base.cc:
Move all caching functionality close together.
Implement a solution for deadlocks caused by
close_cached_tables() when MDL cache is enabled (incomplete).
sql/sql_yacc.yy:
Adjust FLUSH rule to do the necessary initialization of
TABLE_LIST elements used in for FLUSH TABLES <list>, and thus
work OK with flush_mdl_cache() function.
locks for DML statements and changes the way MDL locks
are acquired/granted in contended case.
Instead of backing-off when a lock conflict is encountered
and waiting for it to go away before restarting open_tables()
process we now wait for lock to be released without releasing
any previously acquired locks. If conflicting lock goes away
we resume opening tables. If waiting leads to a deadlock we
try to resolve it by backing-off and restarting open_tables()
immediately.
As result both waiting for possibility to acquire and
acquiring of a metadata lock now always happen within the
same MDL API call. This has allowed to make release of a lock
and granting it to the most appropriate pending request an
atomic operation.
Thanks to this it became possible to wake up during release
of lock only those waiters which requests can be satisfied
at the moment as well as wake up only one waiter in case
when granting its request would prevent all other requests
from being satisfied. This solves thundering herd problem
which occured in cases when we were releasing some lock and
woke up many waiters for SNRW or X locks (this was the issue
in bug#52289 "performance regression for MyISAM in sysbench
OLTP_RW test".
This also allowed to implement more fair (FIFO) scheduling
among waiters with the same priority.
It also opens the door for introducing new types of requests
for metadata locks such as low-prio SNRW lock which is
necessary in order to support LOCK TABLES LOW_PRIORITY WRITE.
Notice that after this sometimes can report ER_LOCK_DEADLOCK
error in cases in which it has not happened before.
Particularly we will always report this error if waiting for
conflicting lock has happened in the middle of transaction
and resulted in a deadlock. Before this patch the error was
not reported if deadlock could have been resolved by backing
off all metadata locks acquired by the current statement.
mysql-test/r/mdl_sync.result:
Added test coverage for some aspects of deadlock handling in
metadata locking subsystem.
Adjusted test case after removing back-off in general case
when conflicting metadata lock is encountered during
open_tables() (now this happens only if waiting for
conflicting lock to go away leads to a deadlock).
mysql-test/r/sp_sync.result:
Adjusted test case after removing back-off in general case
when conflicting metadata lock is encountered during
open_tables() (now this happens only if waiting for
conflicting lock to go away leads to a deadlock).
mysql-test/suite/perfschema/r/dml_setup_instruments.result:
Adjusted test results after renaming MDL_context::
m_waiting_for_lock rwlock to m_LOCK_waiting_for.
mysql-test/suite/rpl/r/rpl_sp.result:
Adjusted test case after implementing new approach to
acquiring metadata locks in open_tables(). We no longer
release all MDL locks acquired by statement before waiting
for conflicting lock to go away. As result DROP FUNCTION
statement has to wait for DML statement which managed to
acquire metadata lock on function being dropped and now
waits for other conflicting metadata lock to go away.
mysql-test/suite/rpl/t/rpl_sp.test:
Adjusted test case after implementing new approach to
acquiring metadata locks in open_tables(). We no longer
release all MDL locks acquired by statement before waiting
for conflicting lock to go away. As result DROP FUNCTION
statement has to wait for DML statement which managed to
acquire metadata lock on function being dropped and now
waits for other conflicting metadata lock to go away.
mysql-test/t/mdl_sync.test:
Added test coverage for some aspects of deadlock handling in
metadata locking subsystem.
Adjusted test case after removing back-off in general case
when conflicting metadata lock is encountered during
open_tables() (now this happens only if waiting for
conflicting lock to go away leads to a deadlock).
mysql-test/t/sp_sync.test:
Adjusted test case after removing back-off in general case
when conflicting metadata lock is encountered during
open_tables() (now this happens only if waiting for
conflicting lock to go away leads to a deadlock).
sql/mdl.cc:
Changed MDL subsystem to support new approach to acquring
metadata locks in open tables and more fair and efficient
scheduling of metadata locks. To implement this:
- Made releasing of the lock and granting it to the most
appropriate pending request atomic operation. As result it
became possible to wake up only those waiters requests from
which can be satisfied at the moment as well as wake-up
only one waiter in case when granting its request would
prevent all other requests from being satisfied.
This solved thundering herd problem which occured in cases
when we were releasing some lock and woke up many waiters
for SNRW or X locks (this was the issue in Bug #52289
"performance regression for MyISAM in sysbench OLTP_RW
test".
To emphasize above changes wake_up_waiters() was renamed
to MDL_context::reschedule_waiters().
- Changed code to add tickets for new requests to the back of
waiters queue and to select tickets to be satisfied from
the head of the queue if possible (this makes scheduling of
requests with the same priority fair). To be able to do
this efficiently we now use for waiting and granted queues
version of I_P_List class which provides fast push_back()
method.
- Members and methods of MDL_context related to sending
and waiting for signal were moved to separate MDL_wait
class.
- Since in order to avoid race conditions we must grant the
lock only to the context which was not chosen as a victim
of deadlock, killed or aborted due to timeout
MDL_wait::set_status() (former awake()) was changed not to
send signal if signal slot is already occupied and to
indicate this fact through its return value. As another
consequence MDL_wait::timed_wait() method was changed to
handle timeout (optionally) and abort due to kill as
signals which make signal slot occupied.
- Renamed MDL_context::acquire_lock_impl() to acquire_lock().
Changed it to be able correctly process requests for shared
locks when there are open HANDLERs, made this method more
optimized for acquisition of shared locks. As part of this
change moved code common between try_acquire_lock() and
acquire_lock() to new try_acquire_lock_impl() method.
Also adjusted acquire_lock()'s code to take into account
the fact that in cases when lock is granted as result of
MDL_context::reschedule_waiters() call (i.e. when it is
granted after waiting for lock conflict to go away)
updating MDL_lock state is responsibility of the thread
calling reschedule_waiters().
- Changed MDL_context::find_deadlock() to send VICTIM
signal even if victim is the context which has initiated
deadlock detection. This is required in order to avoid
races in cases when the same context simultaneously is
chosen as a victim and its request for lock is satisfied.
As result return value of this method became unnecessary
and it was changed to return void.
Adjusted MDL_lock::find_deadlock() method to take into
account that now there can be a discrepancy between
MDL_context::m_waiting_for value being set and real state
of the ticket this member points to.
- Renamed MDL_context::m_waiting_for_lock to m_LOCK_waiting_for
and MDL_context::stop_waiting() to done_waiting_for().
- Finally, removed MDL_context::wait_for_lock() method.
sql/mdl.h:
Changed MDL subsystem to support new approach to acquring
metadata locks in open tables and more fair and efficient
scheduling of metadata locks. To implement this:
- Members and methods of MDL_context related to sending
and waiting for signal were moved to separate MDL_wait
class.
- Since now in order to avoid race conditions we must grant
the lock only to the context which was not chosen as a
victim of deadlock, killed or aborted due to timeout
MDL_wait::set_status (former awake()) was changed not to
send signal if signal slot is already occupied and to
indicate this fact through its return value.
Also NORMAL_WAKE_UP signal became GRANTED, and timeouts
and aborts due to kill became full blown signals rather
than simple return values.
- MDL_wait::timed_wait() now takes extra parameter that
indicates whether signal should be set if timeout is
reached.
- Enabled fast push_back() operation in MDL_context::m_tickets
list to make move_ticket_after_trans_sentinel() method more
efficient.
- Removed MDL_context::wait_for_lock() method.
- Renamed MDL_context::m_waiting_for_lock to m_LOCK_waiting_for
and MDL_context::stop_waiting() to done_waiting_for().
- MDL_context::acquire_lock_impl() became acquire_lock().
- Introduced MDL_context::try_acquire_lock_impl() as a
place for code shared by try_acquire_lock and
acquire_lock().
- Due to fact that now VICTIM signal is sent even if victim
is the context which has initiated deadlock detection
find_deadlock() no longer needs a return value.
sql/sql_base.cc:
Implemented new approach to acquiring metadata locks in
open_tables(). We no longer perform back-off when conflicting
metadata lock is encountered. Instead we wait for this lock
to go away while holding all locks which were acquired so
far. Back-off is only used in situation when further waiting
will cause a deadlock which could be avoided by performing
back-off and restarting open_tables() process. Absence of
waiting between back-off and restart of acquiring metadata
locks can't lead to livelocks as MDL subsystem was changed
to make release of lock and granting it to waiting lock
an atomic action, so back-off will automatically give way
to other participants of deadlock loop.
Accordingly:
- open_table_get_mdl_lock() and open_and_process_routine()
were changed to wait for conflicting metadata lock to
go away without back-off. Only if such wait leads to a
deadlock back-off is requested. As part of this change
new error handler class was introduced which converts,
if possible, ER_LOCK_DEADLOCK error to a request for
back-off and re-start of open_tables() process.
- Open_table_context::recover_from_failed_open() was changed
not to wait in case of metadata lock conflict. Instead we
immediately proceed to re-acquiring locks.
- Open_table_context::request_backoff_action() now always
emits error if back-off is requested in the middle of
transaction as we can't be sure that releasing lock
which were acquired only by current statement will
resolve a deadlock. Before this patch such situations were
successfully detected thanks to the fact that we called
MDL_context::wait_for_lock() method in
recover_from_failed_open().
- In order to avoid deadlocks open_tables() code was adjusted
to flush open HANDLERs for which there are pending requests
for X locks before restarting the process of acquiring
metadata locks.
- Changed close_tables_for_reopen() not to reset MDL_request
for tables belonging to the tail of prelocking list. It is
no longer necessary as these MDL_request objects won't be
used for any waiting.
- Adjusted comment in tdc_wait_for_old_version() to avoid
mentioning removed MDL_context::wait_for_lock() method.
sql/sql_base.h:
As we no longer wait for conflicting metadata lock away in
Open_table_context::recover_from_failed_open() method,
Open_table_context::OT_WAIT_MDL_LOCK action was renamed to
OT_MDL_CONFLICT.
Also Open_table_context::m_failed_mdl_request became
unnecessary and was removed.
sql/sql_plist.h:
Extended I_P_List template to support efficient push_back()
operation if it is parameterized with an appropriate policy
class.
sql/sql_show.cc:
Adjusted code after removal of MDL_context::wait_for_lock()
method. Now if one needs to acquire metadata lock with waiting
one has to use a variant of MDL_context::acquire_lock() method.
Conflicts:
Text conflict in mysql-test/r/archive.result
Contents conflict in mysql-test/r/innodb_bug38231.result
Text conflict in mysql-test/r/mdl_sync.result
Text conflict in mysql-test/suite/binlog/t/disabled.def
Text conflict in mysql-test/suite/rpl_ndb/r/rpl_ndb_binlog_format_errors.result
Text conflict in mysql-test/t/archive.test
Contents conflict in mysql-test/t/innodb_bug38231.test
Text conflict in mysql-test/t/mdl_sync.test
Text conflict in sql/sp_head.cc
Text conflict in sql/sql_show.cc
Text conflict in sql/table.cc
Text conflict in sql/table.h
