Parallel slave failed to retry in retry_event_group() with error
WSREP: Parallel slave worker failed at wsrep_before_command() hook
Fix wsrep transaction cleanup/restart in retry_event_group() to properly
clean up previous transaction by calling wsrep_after_statement().
Also move call to reset error after call to wsrep_after_statement()
to make sure that it remains effective.
Add a MTR test galera_as_slave_parallel_retry to reproduce the error
when the fix is not present.
Other issues which were detected when testing with sysbench:
Check if parallel slave is killed for retry before waiting for prior
commits in THD::wsrep_parallel_slave_wait_for_prior_commit(). This
is required with slave-parallel-mode=optimistic to avoid deadlock
when a slave later in commit order manages to reach prepare phase
before a lock conflict is detected.
Suppress wsrep applier specific warning for slave threads.
Signed-off-by: Julius Goryavsky <julius.goryavsky@mariadb.com>
Ignoring configured server_id should not be a warning because
correct configuration is documented. Changed message to info
level with more detailed message what was configured and
what will be actually used.
Signed-off-by: Julius Goryavsky <julius.goryavsky@mariadb.com>
While applying CTAS log event, we peek the relay log to see if CTAS
contains inserted rows or if it's empty.
The peek function didn't check for end-of-file condition when tried to
get the next event from the log, and thus it hanged.
The fix includes checking for end-of-file while peeking for log events
and considering returned XID_EVENT value as a sign of an empty CTAS.
Signed-off-by: Julius Goryavsky <julius.goryavsky@mariadb.com>
The problem was that when using clang + asan, we do not get a correct value
for the thread stack as some local variables are not allocated at the
normal stack.
It looks like that for example clang 18.1.3, when compiling with
-O2 -fsanitize=addressan it puts local variables and things allocated by
alloca() in other areas than on the stack.
The following code shows the issue
Thread 6 "mariadbd" hit Breakpoint 3, do_handle_one_connection
(connect=0x5080000027b8,
put_in_cache=<optimized out>) at sql/sql_connect.cc:1399
THD *thd;
1399 thd->thread_stack= (char*) &thd;
(gdb) p &thd
(THD **) 0x7fffedee7060
(gdb) p $sp
(void *) 0x7fffef4e7bc0
The address of thd is 24M away from the stack pointer
(gdb) info reg
...
rsp 0x7fffef4e7bc0 0x7fffef4e7bc0
...
r13 0x7fffedee7060 140737185214560
r13 is pointing to the address of the thd. Probably some kind of
"local stack" used by the sanitizer
I have verified this with gdb on a recursive call that calls alloca()
in a loop. In this case all objects was stored in a local heap,
not on the stack.
To solve this issue in a portable way, I have added two functions:
my_get_stack_pointer() returns the address of the current stack pointer.
The code is using asm instructions for intel 32/64 bit, powerpc,
arm 32/64 bit and sparc 32/64 bit.
Supported compilers are gcc, clang and MSVC.
For MSVC 64 bit we are using _AddressOfReturnAddress()
As a fallback for other compilers/arch we use the address of a local
variable.
my_get_stack_bounds() that will return the address of the base stack
and stack size using pthread_attr_getstack() or NtCurrentTed() with
fallback to using the address of a local variable and user provided
stack size.
Server changes are:
- Moving setting of thread_stack to THD::store_globals() using
my_get_stack_bounds().
- Removing setting of thd->thread_stack, except in functions that
allocates a lot on the stack before calling store_globals(). When
using estimates for stack start, we reduce stack_size with
MY_STACK_SAFE_MARGIN (8192) to take into account the stack used
before calling store_globals().
I also added a unittest, stack_allocation-t, to verify the new code.
Reviewed-by: Sergei Golubchik <serg@mariadb.org>
It's read for every command execution, and during slave replication
for every applied event.
It's also planned to be used during write set applying, so it means
mostly every server thread is going to compete for the mutex covering
this variable, especially considering how rarely it changes.
Converting wsrep_ready to atomic relaxes the things.
Signed-off-by: Julius Goryavsky <julius.goryavsky@mariadb.com>
It's possible that MDL conflict handling code is called more
than once for a transaction when:
- it holds more than one conflicting MDL lock
- reschedule_waiters() is executed,
which results in repeated attempts to BF-abort already aborted
transaction.
In such situations, it might be that BF-aborting logic sees
a partially rolled back transaction and erroneously decides
on future actions for such a transaction.
The specific situation tested and fixed is when a SR transaction
applied in the node gets BF-aborted by a started TOI operation.
It's then caught with the server transaction already rolled back,
but with no MDL locks yet released. This caused wrong state
detection for such a transaction during repeated MDL conflict
handling code execution.
Signed-off-by: Julius Goryavsky <julius.goryavsky@mariadb.com>
For TOI events specifically we have a situation where in case of the
same error different nodes may generate different messages. This may
be for two reasons:
- different locale setting between the current client session and
server default (we can reasonably require server locales to be
identical on all nodes, but user can change message locale for the
session)
- non-deterministic course of STATEMENT execution e.g. for ALTER TABLE
On the other hand we may reasonably expect TOI event failures since
they are executed after replication, so we must ensure that voting is
consistent. For that purpose error codes should be sufficiently unique
and deterministic for TOI event failures as DDLs normally deal with
a single object, so we can merely use MySQL error codes to vote on.
Notice that this problem does not happen with regular transactional
writesets, since the originator node will always vote success and
replica nodes are assumed to have the same global locale setting.
As such different error messages indicate different errors even if
the error code is the same (e.g. ER_DUP_KEY can happen on different
rows tables).
Use only MySQL error code (without the error message) for error voting
in case of TOI event failure.
Signed-off-by: Julius Goryavsky <julius.goryavsky@mariadb.com>
It's possible to establish Galera multi-cluster setups connected
through the native replication when every Galera cluster is configured
to have a separate domain ID.
For this setup to work, we need to replace domain ID values in generated
GTID events when they are written at transaction commit to the values
configured by Wsrep replication.
At the same time, it's possible that the GTID event already contains
a correct domain ID if it comes through the native replication from
another Galera cluster.
In this case, when such an event is applied either through a native
replication slave thread or through Wsrep applier, we write GTID event
on transaction start and avoid writing it during transaction commit.
The code contained multiple problems that were fixed:
- applying GTID events didn't work because it's applied without a
running server transaction and Wsrep transaction was not started
- GTID event generation on transaction start didn't contain proper
"standalone" and "is_transactional" flags that the original applied
GTID event contained
- condition determining that GTID event is written on transaction start
to avoid writing it on commit relied on the fact that the GTID event
is the first found in transaction/statement caches, which wasn't the
case and resulted in duplicate GTID events written
- instead of relying on the caches to find a GTID event, a simple check
is introduced that follows the exact rules for checking if event is
written at transaction start as described above
- the test case is improved to check that exact GTID events are
applied after two Galera clusters have synced.
Signed-off-by: Julius Goryavsky <julius.goryavsky@mariadb.com>
This commit fixes sporadic failures in galera_3nodes_sr.GCF-336
test. The following changes have been made here:
1) A small addition to the test itself which should make
it more deterministic by waiting for non-primary state
before COMMIT;
2) More careful handling of the wsrep_ready variable in
the server code (it should always be protected with mutex).
No additional tests are required.
- Fix to avoid mysqltest client getting killed abruptly during
mysql_shutdown(). When Galera replication is shutdown, wait for
THDs with `thd->stmt_da()->is_eof()` to disconnect (these are about
to disconnect anyway).
- Extract duplicate code from `wsrep_stop_replication()` and
`wsrep_shutdown_replication()` in a new function.
- No need to use a custom `shutdown_mysqld.inc` in galera
suite. Delete it, so that the one in `mysql-test/include/` is used.
Signed-off-by: Julius Goryavsky <julius.goryavsky@mariadb.com>
When we commit empty transaction we should allow wsrep
transaction to be on s_must_replay state for DDL that
was killed during certification.
Fix is tested with RQG because deterministic mtr-testcase
was not found.
Signed-off-by: Julius Goryavsky <julius.goryavsky@mariadb.com>
wsrep_plugin_init(), wsrep_plugin_deinit(): Remove these dummy functions
in order to fix an error that would be flagged by cmake -DWITH_UBSAN=ON
when using clang.
wsrep_show_ready(), wsrep_show_bf_aborts(): Correct the signature.
- Add selected tables as shared keys for CTAS certification
- Set proper security context on the replayer thread
- Disallow CTAS command retry
Signed-off-by: Julius Goryavsky <julius.goryavsky@mariadb.com>
In some cases "SHOW PROCESSLIST" could show "Reset for next command"
as State, even if the previous query had finished properly.
Fixed by clearing State after end of command and also setting the State
for the "Connect" command.
Other things:
- Changed usage of 'thd->set_command(COM_SLEEP)' to
'thd->mark_connection_idle()'.
- Changed thread_state_info() to return "" instead of NULL. This is
just a safety measurement and in line with the logic of the
rest of the function.
Problem was that with BINLOG-statement you can execute
binlog events on master also (not only in applier).
Fix removes too strict part wsrep_thd_is_applying from
assertion. Note that actual event in test is intentionally
corrupted to test should this error being ignored.
Signed-off-by: Julius Goryavsky <julius.goryavsky@mariadb.com>
There was two related problems:
(1) Galera node that is defined as a slave to async MariaDB
master at restart might do SST (state stransfer) and
part of that it will copy mysql.gtid_slave_pos table.
Problem is that updates on that table are not replicated
on a cluster. Therefore, table from donor that is not
slave is copied and joiner looses gtid position it was
and start executing events from wrong position of the binlog.
This incorrect position could break replication and
causes node to be dropped and requiring user action.
(2) Slave sql thread might start executing events before
galera is ready (wsrep_ready=ON) and that could also
cause node to be dropped from the cluster.
In this fix we enable replication of mysql.gtid_slave_pos
table on a cluster. In this way all nodes in a cluster
will know gtid slave position and even after SST joiner
knows correct gtid position to start.
Furthermore, we wait galera to be ready before slave
sql thread executes any events to prevent too early
execution.
Signed-off-by: Julius Goryavsky <julius.goryavsky@mariadb.com>
Do not start TOI for CREATE TEMPORARY SEQUENCE because
object is local only and not replicated. Similarly,
avoid starting RSU for TEMPORARY SEQUENCEs. Finally,
we need to run commit hooks for TEMPORARY SEQUENCEs
because CREATE TEMPORARY SEQUENCE does implicit
commit for previous changes that need to be replicated
and committed.
Signed-off-by: Julius Goryavsky <julius.goryavsky@mariadb.com>
The problem seems to be a deadlock between KILL command execution
and BF abort issued by an applier, where:
* KILL has locked victim's LOCK_thd_kill and LOCK_thd_data.
* Applier has innodb side global lock mutex and victim trx mutex.
* KILL is calling innobase_kill_query, and is blocked by innodb
global lock mutex.
* Applier is in wsrep_innobase_kill_one_trx and is blocked by
victim's LOCK_thd_kill.
The fix in this commit removes the TOI replication of KILL command
and makes KILL execution less intrusive operation. Aborting the
victim happens now by using awake_no_mutex() and ha_abort_transaction().
If the KILL happens when the transaction is committing, the
KILL operation is postponed to happen after the statement
has completed in order to avoid KILL to interrupt commit
processing.
Notable changes in this commit:
* wsrep client connections's error state may remain sticky after
client connection is closed. This error message will then pop
up for the next client session issuing first SQL statement.
This problem raised with test galera.galera_bf_kill.
The fix is to reset wsrep client error state, before a THD is
reused for next connetion.
* Release THD locks in wsrep_abort_transaction when locking
innodb mutexes. This guarantees same locking order as with applier
BF aborting.
* BF abort from MDL was changed to do BF abort on server/wsrep-lib
side first, and only then do the BF abort on InnoDB side. This
removes the need to call back from InnoDB for BF aborts which originate
from MDL and simplifies the locking.
* Removed wsrep_thd_set_wsrep_aborter() from service_wsrep.h.
The manipulation of the wsrep_aborter can be done solely on
server side. Moreover, it is now debug only variable and
could be excluded from optimized builds.
* Remove LOCK_thd_kill from wsrep_thd_LOCK/UNLOCK to allow more
fine grained locking for SR BF abort which may require locking
of victim LOCK_thd_kill. Added explicit call for
wsrep_thd_kill_LOCK/UNLOCK where appropriate.
* Wsrep-lib was updated to version which allows external
locking for BF abort calls.
Changes to MTR tests:
* Disable galera_bf_abort_group_commit. This test is going to
be removed (MDEV-30855).
* Record galera_gcache_recover_manytrx as result file was incomplete.
Trivial change.
* Make galera_create_table_as_select more deterministic:
Wait until CTAS execution has reached MDL wait for multi-master
conflict case. Expected error from multi-master conflict is
ER_QUERY_INTERRUPTED. This is because CTAS does not yet have open
wsrep transaction when it is waiting for MDL, query gets interrupted
instead of BF aborted. This should be addressed in separate task.
* A new test galera_kill_group_commit to verify correct behavior
when KILL is executed while the transaction is committing.
Co-authored-by: Seppo Jaakola <seppo.jaakola@iki.fi>
Co-authored-by: Jan Lindström <jan.lindstrom@galeracluster.com>
Signed-off-by: Julius Goryavsky <julius.goryavsky@mariadb.com>
This is a backport from 10.5.
The problem seems to be a deadlock between KILL command execution
and BF abort issued by an applier, where:
* KILL has locked victim's LOCK_thd_kill and LOCK_thd_data.
* Applier has innodb side global lock mutex and victim trx mutex.
* KILL is calling innobase_kill_query, and is blocked by innodb
global lock mutex.
* Applier is in wsrep_innobase_kill_one_trx and is blocked by
victim's LOCK_thd_kill.
The fix in this commit removes the TOI replication of KILL command
and makes KILL execution less intrusive operation. Aborting the
victim happens now by using awake_no_mutex() and ha_abort_transaction().
If the KILL happens when the transaction is committing, the
KILL operation is postponed to happen after the statement
has completed in order to avoid KILL to interrupt commit
processing.
Notable changes in this commit:
* wsrep client connections's error state may remain sticky after
client connection is closed. This error message will then pop
up for the next client session issuing first SQL statement.
This problem raised with test galera.galera_bf_kill.
The fix is to reset wsrep client error state, before a THD is
reused for next connetion.
* Release THD locks in wsrep_abort_transaction when locking
innodb mutexes. This guarantees same locking order as with applier
BF aborting.
* BF abort from MDL was changed to do BF abort on server/wsrep-lib
side first, and only then do the BF abort on InnoDB side. This
removes the need to call back from InnoDB for BF aborts which originate
from MDL and simplifies the locking.
* Removed wsrep_thd_set_wsrep_aborter() from service_wsrep.h.
The manipulation of the wsrep_aborter can be done solely on
server side. Moreover, it is now debug only variable and
could be excluded from optimized builds.
* Remove LOCK_thd_kill from wsrep_thd_LOCK/UNLOCK to allow more
fine grained locking for SR BF abort which may require locking
of victim LOCK_thd_kill. Added explicit call for
wsrep_thd_kill_LOCK/UNLOCK where appropriate.
* Wsrep-lib was updated to version which allows external
locking for BF abort calls.
Changes to MTR tests:
* Disable galera_bf_abort_group_commit. This test is going to
be removed (MDEV-30855).
* Record galera_gcache_recover_manytrx as result file was incomplete.
Trivial change.
* Make galera_create_table_as_select more deterministic:
Wait until CTAS execution has reached MDL wait for multi-master
conflict case. Expected error from multi-master conflict is
ER_QUERY_INTERRUPTED. This is because CTAS does not yet have open
wsrep transaction when it is waiting for MDL, query gets interrupted
instead of BF aborted. This should be addressed in separate task.
* A new test galera_kill_group_commit to verify correct behavior
when KILL is executed while the transaction is committing.
Co-authored-by: Seppo Jaakola <seppo.jaakola@iki.fi>
Co-authored-by: Jan Lindström <jan.lindstrom@galeracluster.com>
Signed-off-by: Julius Goryavsky <julius.goryavsky@mariadb.com>
Problem for Galera is the fact that sequences are not really
transactional. Sequence operation is committed immediately
in sql_sequence.cd and later Galera could find out that
we have changes but actual statement is not there anymore.
Therefore, we must make some restrictions what kind
of sequences Galera can support.
(1) Galera cluster supports only sequences implemented
by InnoDB storage engine. This is because Galera replication
supports currently only InnoDB.
(2) We do not allow LOCK TABLE on sequence object and
we do not allow sequence creation under LOCK TABLE, instead
lock is released and we issue warning.
(3) We allow sequences with NOCACHE definition or with
INCREMEMENT BY 0 CACHE=n definition. This makes sure that
sequence values are unique accross Galera cluster.
Signed-off-by: Julius Goryavsky <julius.goryavsky@mariadb.com>
After d7d3ad69 we should use KILL_CONNECTION_HARD to interrupt
debug_sync waits. Test case uses debug_sync and then disconnects
connection from cluster.
Updated wsrep-lib to version in which server_state
wait_until_state() and sst_received() were changed to report
errors via return codes instead of throwing exceptions. Added
error handling accordingly.
Tested manually that failure in sst_received() which was
caused by server misconfiguration (unknown configuration variable
in server configuration) does not cause crash due to uncaught
exception.
Test MDEV-26575 fails when it runs after MDEV-25389. This is because
the latter simulates a failure while an applier thread is
created in `start_wsrep_THD()`. The failure was not handled correctly
and would not cleanup the created THD from the global
`server_threads`. A subsequent shutdown would hang and eventually fail
trying to close this THD.
Reviewed-by: Jan Lindström <jan.lindstrom@mariadb.com>
If repl.max_ws_size is set too low following CREATE TABLE could fail
during commit. In this case wsrep_commit_empty should allow rolling
it back if provider state is s_aborted.
Furhermore, original ER_ERROR_DURING_COMMIT does not really tell anything
clear for user. Therefore, this commit adds a new error
ER_TOO_BIG_WRITESET. This will change some test cases output.
In test user has set WSREP_ON=OFF this causes streaming replication
recovery to fail and this caused call to unireg_abort(). However,
this call is not necessary and we can let transaction to fail. Naturally,
if real user does this he needs to bootstrap his cluster.
wsrep_server_incoming_address function always returned value of the
wsrep_node_incoming_address even when actual incoming address
was resolved to inc_addr variable. Fixed by returning inc_addr
if it does contain incoming address.
Using parallel slave applying can cause deadlock between between DDL and
other events. GTID with lower seqno can be blocked in galera when node
entered TOI mode, but DDL GTID which has higher node can be blocked
before previous GTIDs are applied locally.
Fix is to check prior commits before entering TOI.
Reviewed-by: Jan Lindström <jan.lindstrom@mariadb.com>
* Fix test galera.MW-44 to make it work with --ps-protocol
* Skip test galera.MW-328C under --ps-protocol This test
relies on wsrep_retry_autocommit, which has no effect
under ps-protocol.
* Return WSREP related errors on COM_STMT_PREPARE commands
Change wsrep_command_no_result() to allow sending back errors
when a statement is prepared. For example, to handle deadlock
error due to BF aborted transaction during prepare.
* Add sync waiting before statement prepare
When a statement is prepared, tables used in the statement may be
opened and checked for existence. Because of that, some tests (for
example galera_create_table_as_select) that CREATE a table in one node
and then SELECT from the same table in another node may result in errors
due to non existing table.
To make tests behave similarly under normal and PS protocol, we add a
call to sync wait before preparing statements that would sync wait
during normal execution.
Reviewed-by: Jan Lindström <jan.lindstrom@mariadb.com>
