PS-9148: Reworked dictionary / bookshelf thread-safety model #3

percona-ysorokin · 2024-05-02T11:48:31Z

https://perconadev.atlassian.net/browse/PS-9148

Both 'dictionary' and 'bookshelf' classes no longer include their own 'std::shared_mutex' to protect data. Instead, we now have a single 'std::shared_mutex' at the 'query_cache' level.

The return value of the 'get_random()' method in both 'dictionary' and 'bookshelf' classes changed from 'optional_string' to 'std::string_view'. Empty (default constructed) 'std::string_view' is used as an indicator of an unsuccessful operation.
'get_random()' method in the 'query_cache' class still returns a string by value to avoid race conditions.

Changed the behaviour of the 'sql_context::execute_dml()' method - it now throws when SQL errors (like "no table found", etc.) occur.

https://perconadev.atlassian.net/browse/PS-9148 Both 'dictionary' and 'bookshelf' classes no longer include their own 'std::shared_mutex' to protect data. Instead, we now have a single 'std::shared_mutex' at the 'query_cache' level. The return value of the 'get_random()' method in both 'dictionary' and 'bookshelf' classes changed from 'optional_string' to 'std::string_view'. Empty (default constructed) 'std::string_view' is used as an indicator of an unsuccessful operation. 'get_random()' method in the 'query_cache' class still returns a string by value to avoid race conditions. Changed the behaviour of the 'sql_context::execute_dml()' method - it now throws when SQL errors (like "no table found", etc.) occur.

Upstream commit ID : fb-mysql-5.6.35/8cb1dc836b68f1f13e8b2655b2b8cb2d57f400b3 PS-5217 : Merge fb-prod201803 Summary: Original report: https://jira.mariadb.org/browse/MDEV-15816 To reproduce this bug just following below steps, client 1: USE test; CREATE TABLE t1 (i INT) ENGINE=MyISAM; HANDLER t1 OPEN h; CREATE TABLE t2 (i INT) ENGINE=RocksDB; LOCK TABLES t2 WRITE; client 2: FLUSH TABLES WITH READ LOCK; client 1: INSERT INTO t2 VALUES (1); So client 1 acquired the lock and set m_lock_rows = RDB_LOCK_WRITE. Then client 2 calls store_lock(TL_IGNORE) and m_lock_rows was wrongly set to RDB_LOCK_NONE, as below ``` #0 myrocks::ha_rocksdb::store_lock (this=0x7fffbc03c7c8, thd=0x7fffc0000ba0, to=0x7fffc0011220, lock_type=TL_IGNORE) #1 get_lock_data (thd=0x7fffc0000ba0, table_ptr=0x7fffe84b7d20, count=1, flags=2) #2 mysql_lock_abort_for_thread (thd=0x7fffc0000ba0, table=0x7fffbc03bbc0) #3 THD::notify_shared_lock (this=0x7fffc0000ba0, ctx_in_use=0x7fffbc000bd8, needs_thr_lock_abort=true) percona#4 MDL_lock::notify_conflicting_locks (this=0x555557a82380, ctx=0x7fffc0000cc8) percona#5 MDL_context::acquire_lock (this=0x7fffc0000cc8, mdl_request=0x7fffe84b8350, lock_wait_timeout=2) percona#6 Global_read_lock::lock_global_read_lock (this=0x7fffc0003fe0, thd=0x7fffc0000ba0) ``` Finally, client 1 "INSERT INTO..." hits the Assertion 'm_lock_rows == RDB_LOCK_WRITE' failed in myrocks::ha_rocksdb::write_row() Fix this bug by not setting m_locks_rows if lock_type == TL_IGNORE. Closes facebook/mysql-5.6#838 Pull Request resolved: facebook/mysql-5.6#871 Differential Revision: D9417382 Pulled By: lth fbshipit-source-id: c36c164e06c

Upstream commit ID : fb-mysql-5.6.35/77032004ad23d21a4c386f8136ecfbb071ea42d6 PS-6865 : Merge fb-prod201903 Summary: Currently during primary key's value encode, its ttl value can be from either one of these 3 cases 1. ttl column in primary key 2. non-ttl column a. old record(update case) b. current timestamp 3. ttl column in non-key field Workflow #1: first in Rdb_key_def::pack_record() find and store pk_offset, then in value encode try to parse key slice to fetch ttl value by using pk_offset. Workflow #3: fetch ttl value from ttl column The change is to merge #1 and #3 by always fetching TTL value from ttl column, not matter whether the ttl column is in primary key or not. Of course, remove pk_offset, since it isn't used. BTW, for secondary keys, its ttl value is always from m_ttl_bytes, which is stored by primary value encoding. Reviewed By: yizhang82 Differential Revision: D14662716 fbshipit-source-id: 6b4e5f044fd

Upstream commit ID : fb-mysql-5.6.35/e025cf1c47e63aada985d78e4083f2e02fba434f PS-7731 : Merge percona-202102 Summary: Today in `SELECT count(*)` MyRocks would still decode every single column due to this check, despite the readset being empty: ``` // bitmap is cleared on index merge, but it still needs to decode columns bool field_requested = decode_all_fields || m_verify_row_debug_checksums || bitmap_is_set(field_map, m_table->field[i]->field_index); ``` As a result MyRocks is significantly slower than InnoDB in this particular scenario. Turns out in index merge, when it tries to reset, it calls ha_index_init with an empty column_bitmap, so our field decoders didn't know it needs to decode anything, so the entire query would return nothing. This is discussed in [this commit](facebook/mysql-5.6@70f2bcd), and [issue 624](facebook/mysql-5.6#624) and [PR 626](facebook/mysql-5.6#626). So the workaround we had at that time is to simply treat empty map as implicitly everything, and the side effect is massively slowed down count(*). We have a few options to address this: 1. Fix index merge optimizer - looking at the code in QUICK_RANGE_SELECT::init_ror_merged_scan, it actually fixes up the column_bitmap properly, but after init/reset, so the fix would simply be moving the bitmap set code up. For secondary keys, prepare_for_position will automatically call `mark_columns_used_by_index_no_reset(s->primary_key, read_set)` if HA_PRIMARY_KEY_REQUIRED_FOR_POSITION is set (true for both InnoDB and MyRocks), so we would know correctly that we need to unpack PK when walking SK during index merge. 2. Overriding `column_bitmaps_signal` and setup decoders whenever the bitmap changes - however this doesn't work by itself. Because no storage engine today actually use handler::column_bitmaps_signal this path haven't been tested properly in index merge. In this case, QUICK_RANGE_SELECT::init_ror_merged_scan should call set_column_bitmaps_no_signal to avoid resetting the correct read/write set of head since head is used as first handler (reuses_handler=true) and subsequent place holders for read/write set updates (reuse_handler=false). 3. Follow InnoDB's solution - InnoDB delays it actually initialize its template again in index_read for the 2nd time (relying on `prebuilt->sql_stat_start`), and during index_read `QUICK_RANGE_SELECT::column_bitmap` is already fixed up and the table read/write set is switched to it, so the new template would be built correctly. In order to make it easier to maintain and port, after discussing with Manuel, I'm going with a simplified version of #3 that delays decoder creation until the first read operation (index_*, rnd_*, range_read_*, multi_range_read_*), and setting the delay flag in index_init / rnd_init / multi_range_read_init. Also, I ran into a bug with truncation_partition where Rdb_converter's tbl_def is stale (we only update ha_rocksdb::m_tbl_def), but it is fine because it is not being used after table open. But my change moves the lookup_bitmap initialization into Rdb_converter which takes a dependency on Rdb_converter::m_tbl_def so now we need to reset it properly. Reference Patch: facebook/mysql-5.6@44d6a8d --------- Porting Note: Due to 8.0's new counting infra (handler::record & handler::record_with_index), this only helps PK counting. Will send out a better fix that works better with 8.0 new counting infra. Reviewed By: Pushapgl Differential Revision: D26265470 fbshipit-source-id: f142be681ab

…s=0 and a local DDL executed https://perconadev.atlassian.net/browse/PS-9018 Problem ------- In high concurrency scenarios, MySQL replica can enter into a deadlock due to a race condition between the replica applier thread and the client thread performing a binlog group commit. Analysis -------- It needs at least 3 threads for this deadlock to happen 1. One client thread 2. Two replica applier threads How this deadlock happens? -------------------------- 0. Binlog is enabled on replica, but log_replica_updates is disabled. 1. Initially, both "Commit Order" and "Binlog Flush" queues are empty. 2. Replica applier thread 1 enters the group commit pipeline to register in the "Commit Order" queue since `log-replica-updates` is disabled on the replica node. 3. Since both "Commit Order" and "Binlog Flush" queues are empty, the applier thread 1 3.1. Becomes leader (In Commit_stage_manager::enroll_for()). 3.2. Registers in the commit order queue. 3.3. Acquires the lock MYSQL_BIN_LOG::LOCK_log. 3.4. Commit Order queue is emptied, but the lock MYSQL_BIN_LOG::LOCK_log is not yet released. NOTE: SE commit for applier thread is already done by the time it reaches here. 4. Replica applier thread 2 enters the group commit pipeline to register in the "Commit Order" queue since `log-replica-updates` is disabled on the replica node. 5. Since the "Commit Order" queue is empty (emptied by applier thread 1 in 3.4), the applier thread 2 5.1. Becomes leader (In Commit_stage_manager::enroll_for()) 5.2. Registers in the commit order queue. 5.3. Tries to acquire the lock MYSQL_BIN_LOG::LOCK_log. Since it is held by applier thread 1 it will wait until the lock is released. 6. Client thread enters the group commit pipeline to register in the "Binlog Flush" queue. 7. Since "Commit Order" queue is not empty (there is applier thread 2 in the queue), it enters the conditional wait `m_stage_cond_leader` with an intention to become the leader for both the "Binlog Flush" and "Commit Order" queues. 8. Applier thread 1 releases the lock MYSQL_BIN_LOG::LOCK_log and proceeds to update the GTID by calling gtid_state->update_commit_group() from Commit_order_manager::flush_engine_and_signal_threads(). 9. Applier thread 2 acquires the lock MYSQL_BIN_LOG::LOCK_log. 9.1. It checks if there is any thread waiting in the "Binlog Flush" queue to become the leader. Here it finds the client thread waiting to be the leader. 9.2. It releases the lock MYSQL_BIN_LOG::LOCK_log and signals on the cond_var `m_stage_cond_leader` and enters a conditional wait until the thread's `tx_commit_pending` is set to false by the client thread (will be done in the Commit_stage_manager::process_final_stage_for_ordered_commit_group() called by client thread from fetch_and_process_flush_stage_queue()). 10. The client thread wakes up from the cond_var `m_stage_cond_leader`. The thread has now become a leader and it is its responsibility to update GTID of applier thread 2. 10.1. It acquires the lock MYSQL_BIN_LOG::LOCK_log. 10.2. Returns from `enroll_for()` and proceeds to process the "Commit Order" and "Binlog Flush" queues. 10.3. Fetches the "Commit Order" and "Binlog Flush" queues. 10.4. Performs the storage engine flush by calling ha_flush_logs() from fetch_and_process_flush_stage_queue(). 10.5. Proceeds to update the GTID of threads in "Commit Order" queue by calling gtid_state->update_commit_group() from Commit_stage_manager::process_final_stage_for_ordered_commit_group(). 11. At this point, we will have - Client thread performing GTID update on behalf if applier thread 2 (from step 10.5), and - Applier thread 1 performing GTID update for itself (from step 8). Due to the lack of proper synchronization between the above two threads, there exists a time window where both threads can call gtid_state->update_commit_group() concurrently. In subsequent steps, both threads simultaneously try to modify the contents of the array `commit_group_sidnos` which is used to track the lock status of sidnos. This concurrent access to `update_commit_group()` can cause a lock-leak resulting in one thread acquiring the sidno lock and not releasing at all. ----------------------------------------------------------------------------------------------------------- Client thread Applier Thread 1 ----------------------------------------------------------------------------------------------------------- update_commit_group() => global_sid_lock->rdlock(); update_commit_group() => global_sid_lock->rdlock(); calls update_gtids_impl_lock_sidnos() calls update_gtids_impl_lock_sidnos() set commit_group_sidno[2] = true set commit_group_sidno[2] = true lock_sidno(2) -> successful lock_sidno(2) -> waits update_gtids_impl_own_gtid() -> Add the thd->owned_gtid in `executed_gtids()` if (commit_group_sidnos[2]) { unlock_sidno(2); commit_group_sidnos[2] = false; } Applier thread continues.. lock_sidno(2) -> successful update_gtids_impl_own_gtid() -> Add the thd->owned_gtid in `executed_gtids()` if (commit_group_sidnos[2]) { <=== this check fails and lock is not released. unlock_sidno(2); commit_group_sidnos[2] = false; } Client thread continues without releasing the lock ----------------------------------------------------------------------------------------------------------- 12. As the above lock-leak can also happen the other way i.e, the applier thread fails to unlock, there can be different consequences hereafter. 13. If the client thread continues without releasing the lock, then at a later stage, it can enter into a deadlock with the applier thread performing a GTID update with stack trace. Client_thread ------------- #1 __GI___lll_lock_wait #2 ___pthread_mutex_lock #3 native_mutex_lock <= waits for commit lock while holding sidno lock percona#4 Commit_stage_manager::enroll_for percona#5 MYSQL_BIN_LOG::change_stage percona#6 MYSQL_BIN_LOG::ordered_commit percona#7 MYSQL_BIN_LOG::commit percona#8 ha_commit_trans percona#9 trans_commit_implicit percona#10 mysql_create_like_table percona#11 Sql_cmd_create_table::execute percona#12 mysql_execute_command percona#13 dispatch_sql_command Applier thread -------------- #1 ___pthread_mutex_lock #2 native_mutex_lock #3 safe_mutex_lock percona#4 Gtid_state::update_gtids_impl_lock_sidnos <= waits for sidno lock percona#5 Gtid_state::update_commit_group percona#6 Commit_order_manager::flush_engine_and_signal_threads <= acquires commit lock here percona#7 Commit_order_manager::finish percona#8 Commit_order_manager::wait_and_finish percona#9 ha_commit_low percona#10 trx_coordinator::commit_in_engines percona#11 MYSQL_BIN_LOG::commit percona#12 ha_commit_trans percona#13 trans_commit percona#14 Xid_log_event::do_commit percona#15 Xid_apply_log_event::do_apply_event_worker percona#16 Slave_worker::slave_worker_exec_event percona#17 slave_worker_exec_job_group percona#18 handle_slave_worker 14. If the applier thread continues without releasing the lock, then at a later stage, it can perform recursive locking while setting the GTID for the next transaction (in set_gtid_next()). In debug builds the above case hits the assertion `safe_mutex_assert_not_owner()` meaning the lock is already acquired by the replica applier thread when it tries to re-acquire the lock. Solution -------- In the above problematic example, when seen from each thread individually, we can conclude that there is no problem in the order of lock acquisition, thus there is no need to change the lock order. However, the root cause for this problem is that multiple threads can concurrently access to the array `Gtid_state::commit_group_sidnos`. In its initial implementation, it was expected that threads should hold the `MYSQL_BIN_LOG::LOCK_commit` before modifying its contents. But it was not considered when upstream implemented WL#7846 (MTS: slave-preserve-commit-order when log-slave-updates/binlog is disabled). With this patch, we now ensure that `MYSQL_BIN_LOG::LOCK_commit` is acquired when the client thread (binlog flush leader) when it tries to perform GTID update on behalf of threads waiting in "Commit Order" queue, thus providing a guarantee that `Gtid_state::commit_group_sidnos` array is never accessed without the protection of `MYSQL_BIN_LOG::LOCK_commit`.

PS-5741: Incorrect use of memset_s in keyring_vault. Fixed the usage of memset_s. The arguments should be: void memset_s(void *dest, size_t dest_max, int c, size_t n) where the 2nd argument is size of buffer and the 3rd is argument is character to fill. --------------------------------------------------------------------------- PS-7769 - Fix use-after-return error in audit_log_exclude_accounts_validate --- *Problem:* `st_mysql_value::val_str` might return a pointer to `buf` which after the function called is deleted. Therefore the value in `save`, after reuturnin from the function, is invalid. In this particular case, the error is not manifesting as val_str` returns memory allocated with `thd_strmake` and it does not use `buf`. *Solution:* Allocate memory with `thd_strmake` so the memory in `save` is not local. --------------------------------------------------------------------------- Fix test main.bug12969156 when WITH_ASAN=ON *Problem:* ASAN complains about stack-buffer-overflow on function `mysql_heartbeat`: ``` ==90890==ERROR: AddressSanitizer: stack-buffer-overflow on address 0x7fe746d06d14 at pc 0x7fe760f5b017 bp 0x7fe746d06cd0 sp 0x7fe746d06478 WRITE of size 24 at 0x7fe746d06d14 thread T16777215 Address 0x7fe746d06d14 is located in stack of thread T26 at offset 340 in frame #0 0x7fe746d0a55c in mysql_heartbeat(void*) /home/yura/ws/percona-server/plugin/daemon_example/daemon_example.cc:62 This frame has 4 object(s): [48, 56) 'result' (line 66) [80, 112) '_db_stack_frame_' (line 63) [144, 200) 'tm_tmp' (line 67) [240, 340) 'buffer' (line 65) <== Memory access at offset 340 overflows this variable HINT: this may be a false positive if your program uses some custom stack unwind mechanism, swapcontext or vfork (longjmp and C++ exceptions *are* supported) Thread T26 created by T25 here: #0 0x7fe760f5f6d5 in __interceptor_pthread_create ../../../../src/libsanitizer/asan/asan_interceptors.cpp:216 #1 0x557ccbbcb857 in my_thread_create /home/yura/ws/percona-server/mysys/my_thread.c:104 #2 0x7fe746d0b21a in daemon_example_plugin_init /home/yura/ws/percona-server/plugin/daemon_example/daemon_example.cc:148 #3 0x557ccb4c69c7 in plugin_initialize /home/yura/ws/percona-server/sql/sql_plugin.cc:1279 percona#4 0x557ccb4d19cd in mysql_install_plugin /home/yura/ws/percona-server/sql/sql_plugin.cc:2279 percona#5 0x557ccb4d218f in Sql_cmd_install_plugin::execute(THD*) /home/yura/ws/percona-server/sql/sql_plugin.cc:4664 percona#6 0x557ccb47695e in mysql_execute_command(THD*, bool) /home/yura/ws/percona-server/sql/sql_parse.cc:5160 percona#7 0x557ccb47977c in mysql_parse(THD*, Parser_state*, bool) /home/yura/ws/percona-server/sql/sql_parse.cc:5952 percona#8 0x557ccb47b6c2 in dispatch_command(THD*, COM_DATA const*, enum_server_command) /home/yura/ws/percona-server/sql/sql_parse.cc:1544 percona#9 0x557ccb47de1d in do_command(THD*) /home/yura/ws/percona-server/sql/sql_parse.cc:1065 percona#10 0x557ccb6ac294 in handle_connection /home/yura/ws/percona-server/sql/conn_handler/connection_handler_per_thread.cc:325 percona#11 0x557ccbbfabb0 in pfs_spawn_thread /home/yura/ws/percona-server/storage/perfschema/pfs.cc:2198 percona#12 0x7fe760ab544f in start_thread nptl/pthread_create.c:473 ``` The reason is that `my_thread_cancel` is used to finish the daemon thread. This is not and orderly way of finishing the thread. ASAN does not register the stack variables are not used anymore which generates the error above. This is a benign error as all the variables are on the stack. *Solution*: Finish the thread in orderly way by using a signalling variable. --------------------------------------------------------------------------- PS-8204: Fix XML escape rules for audit plugin https://jira.percona.com/browse/PS-8204 There was a wrong length specified for some XML escape rules. As a result of this terminating null symbol from replacement rule was copied into resulting string. This lead to quer text truncation in audit log file. In addition added empty replacement rules for '\b' and 'f' symbols which just remove them from resulting string. These symboles are not supported in XML 1.0. --------------------------------------------------------------------------- PS-8854: Add main.percona_udf MTR test Add a test to check FNV1A_64, FNV_64, and MURMUR_HASH user-defined functions.

…n read() syscall over network https://jira.percona.com/browse/PS-8592 Description ----------- GR suffered from problems caused by the security probes and network scanner processes connecting to the group replication communication port. This usually is not a problem, but poses a serious threat when another member tries to join the cluster by initialting a connection to the member which is affected by external processes using the port dedicated for group communication for longer durations. On such activites by external processes, the SSL enabled server stalled forever on the SSL_accept() call waiting for handshake data. Below is the stacktrace: Thread 55 (Thread 0x7f7bb77ff700 (LWP 2198598)): #0 in read () #1 in sock_read () #2 in BIO_read () #3 in ssl23_read_bytes () percona#4 in ssl23_get_client_hello () percona#5 in ssl23_accept () percona#6 in xcom_tcp_server_startup(Xcom_network_provider*) () When the server stalled in the above path forever, it prohibited other members to join the cluster resulting in the following messages on the joiner server's logs. [ERROR] [MY-011640] [Repl] Plugin group_replication reported: 'Timeout on wait for view after joining group' [ERROR] [MY-011735] [Repl] Plugin group_replication reported: '[GCS] The member is already leaving or joining a group.' Solution -------- This patch adds two new variables 1. group_replication_xcom_ssl_socket_timeout It is a file-descriptor level timeout in seconds for both accept() and SSL_accept() calls when group replication is listening on the xcom port. When set to a valid value, say for example 5 seconds, both accept() and SSL_accept() return after 5 seconds. The default value has been set to 0 (waits infinitely) for backward compatibility. This variable is effective only when GR is configred with SSL. 2. group_replication_xcom_ssl_accept_retries It defines the number of retries to be performed before closing the socket. For each retry the server thread calls SSL_accept() with timeout defined by the group_replication_xcom_ssl_socket_timeout for the SSL handshake process once the connection has been accepted by the first accept() call. The default value has been set to 10. This variable is effective only when GR is configred with SSL. Note: - Both of the above variables are dynamically configurable, but will become effective only on START GROUP_REPLICATION. ------------------------------------------------------------------------- PS-8844: Fix the failing main.mysqldump_gtid_purged https://jira.percona.com/browse/PS-8844 This patch fixes the test failure of main.mysqldump_gtid_purged that failed due to the uninitialized variable $redirect_stderr in the start_proc_in_background.inc.