FB9-277 - Disable tests fil_path.get_existing_path and get_real_path #10
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…gin (facebook#934) (facebook#934) Summary: JIRA: https://jira.percona.com/browse/FB8-54 JIRA: https://jira.percona.com/browse/FB8-55 JIRA: https://jira.percona.com/browse/FB8-70 JIRA: https://jira.percona.com/browse/FB8-101 This commit adds the following fields to the generic event in audit log: * query_id * database * affected_rows * connection_certificate Reference Patch: facebook@1def6b7 Reference Patch: facebook@ce95a09 Reference Patch: facebook@588be34 Reference Patch: facebook@ba03c70 Reference Patch: facebook@be8c587 Reference Patch: facebook@22b2508 We need some extra info for the shadowing and security logging. This is a simple first step of info that MariaDB actually also exposes. Now we would have the `query_id` and the database name for general events. Making as few changes as possible to accomplish it, so I'm just taking the information from the TDH and exposing it through `mysql_event_general` struct and as a argument to disconnect. Forward the connection certificate to the audit plugin. The connection certificate can then be parsed by the audit plugin and handled appropriately. It made more sense for the certificate to live in the connection events, since they generally don't change between every general event, so the move was done. This is done by caching a BUF_MEM struct on the THD object. Since it's not possible to change certificates on the same connection, this caching should be correct. The BUF_MEM is released on THD::release_resources. If upstream bumps the MYSQL_AUDIT_INTERFACE_VERSION, we should bump ours to be greater or equal to it. Expose the port current mysqld is running on for the audit plugin. If no port, 0 is used. Pull Request resolved: facebook#934 Reviewed By: lloyd Differential Revision: D13874133 Pulled By: lth
…acebook#940) Summary: JIRA: https://jira.percona.com/browse/FB8-93 Reference Patch: facebook@a2b0c32 Reference Patch: facebook@e82ab5a Instead of just showing the errno when running SHOW SLAVE STATUS command, also printing the error name so that user won't have misunderstanding Pull Request resolved: facebook#940 Reviewed By: lloyd Differential Revision: D13924591 Pulled By: lth
…#937) Summary: https://jira.percona.com/browse/FB8-47 The max user name length is increased to 80 chars (utf8mb3) Since the system tables are InnoDB tables and because Primary key is on the full username (80 *3 = 240 bytes) and other columns, InnoDb initialization with 4k page size fails. This is new limitation. Reference Patch: facebook@3dbbcfc This mitigates upstream bug: http://bugs.mysql.com/3083 User name length limit of 16 was too short, so made it 80. 80 is effectively the limit without making additional changes because of MyISAM's 1000 byte limit on key length and the primary key of mysql.columns_priv. To compute the size of the key for that table, the lengths of the char fields are summed and then multipled by 3, resulting in a limit on the max size of user name lengths being 81 characters. Pull Request resolved: facebook#937 Reviewed By: lloyd Differential Revision: D13890369 Pulled By: lth
Summary:
Implements per-database scope read-only setting. The read-only can be
turned on in `alter database` statement. Privilege to set read-only
follows alter-database ACL.
ALTER DATABASE db_name [SUPER_READ_ONLY | READ_ONLY = FALSE | TRUE]]
`SUPER_READ_ONLY = TRUE` will prevent all write transactions from
committing for any user (including super users).
`READ_ONLY = TRUE` will prevent any write transaction from committing
for regular users, while super users can still write to the database.
`SUPER_READ_ONLY = FALSE` will turn off super_read_only on the database,
and the database remains on read_only (for regular users),
`READ_ONLY = FALSE` will turn off read_only on the database,
The READ_ONLY database status can be shown by `show create database`.
mysql> alter database test read_only = true;
Query OK, 1 row affected (0.00 sec)
mysql> show create database test;
+----------+---------------------------------------------------------------------------+
| Database | Create Database |
+----------+---------------------------------------------------------------------------+
| test | CREATE DATABASE `test` /*!40100 DEFAULT CHARACTER SET latin1 READ_ONLY */ |
+----------+---------------------------------------------------------------------------+
1 row in set (0.00 sec)
mysql> alter database test super_read_only = true;
Query OK, 1 row affected (0.00 sec)
mysql> show create database test;
+----------+---------------------------------------------------------------------------------+
| Database | Create Database |
+----------+---------------------------------------------------------------------------------+
| test | CREATE DATABASE `test` /*!40100 DEFAULT CHARACTER SET latin1 SUPER_READ_ONLY */ |
+----------+---------------------------------------------------------------------------------+
1 row in set (0.00 sec)
mysql> alter database test super_read_only = false;
Query OK, 1 row affected (0.00 sec)
mysql> show create database test;
+----------+---------------------------------------------------------------------------+
| Database | Create Database |
+----------+---------------------------------------------------------------------------+
| test | CREATE DATABASE `test` /*!40100 DEFAULT CHARACTER SET latin1 READ_ONLY */ |
+----------+---------------------------------------------------------------------------+
1 row in set (0.00 sec)
Details:
* READ_ONLY flag is persisted in db.ops and survives restarts.
* Alter database will only return when no write transaction is
still committing. Uncommitted write transactions will fail after
alter database succeeds.
* DB options are stored in each thread (using hash map), so checking
the read_only options doesn't grab a global lock. Only db_read_only
that is turned on is stored in the local hash map.
* The first time thead local hash map is initialized (once), shared db
option map will be locked and accessed. Since the db options are not
automatically loaded into the shared hash map (cache) at the
beginning, we have to explicitly load the db options during the
first time any thread initializes its local hash map.
* create/alter/delete database DDLs will iterate through thread array
to update local db opt (if needed), similar to how show processlist
works.
* write transactions are checked against local hashmap at commit time
(either explicit commit or auto-commit). This will make the explicit
commit and auto-commits consistent, and avoid problems such that
explicit commit may become a "read-only" transaction if we block
write-statements at the parsing time.
* The list of databases accessed in a write transaction is obtained
through the metadata locks (MDL) that the transaction holds. As I
noted in the code, this may rollback some false positives, such as
the read-only DB in the list may not actually be modified in the
transaction (for cross-db transactions). We do not differentiate
such cases.
Reference patch:
facebook@8194409
facebook@c971129
facebook@e26e9b2
facebook@a9f6f3d
facebook@b1ff9a6
facebook@548c7ca
Differential Revision: D9379904
…ok#972) (facebook#972) Summary: JIRA ticket: https://jira.percona.com/browse/FB8-120 Reference commit: facebook@2f08e3a Reference commit: facebook@a21e84f Reference commit: facebook@5983ba3181a Reference commit: facebook@414870abeee Added new option db_metadata to the create database `CREATE DATABASE [IF NOT EXISTS] db_name DB_METADATA [=] metadata` The new option is stored in the db.opt file along with other database options like charset and read_only. The options accepts any string value. If not specified, it writes empty string as the default value in the db.opt file. To retrieve the option value for a particular database, the show command can be used as follows: `SHOW CREATE DATABASE db_name` Example Usage: create database test2 charset utf8 read_only = true db_metadata '{\'shard_name\':\'myshard_for_test2\'}'; show create database test2; Output: +----------+--------------------------------------------------------------------------------------------------------------------------+ | Database | Create Database | +----------+--------------------------------------------------------------------------------------------------------------------------+ | test2 | CREATE DATABASE `test2` /*!40100 DEFAULT CHARACTER SET utf8 READ_ONLY DB_METADATA {'shard_name':'myshard_for_test2'} */ | +----------+--------------------------------------------------------------------------------------------------------------------------+ Pull Request resolved: facebook#972 Differential Revision: D14883632 Pulled By: yizhang82
Summary: Add DB_METADATA to INFORMATION_SCHEMA.SCHEMATA for each database for easy query by OLM --- 8.0 porting comments 'SCHEMATA_EXT' Information Schema table is implemented as Data Dictionary system view. Reference Patch: facebook@64bf72c276 Differential Revision: D13510326
…ook#953) (facebook#953) Summary: Jira ticket: https://jira.percona.com/browse/FB8-80 Reference Patch: facebook@564824a Set row_query using the fields unpacked from the row event. Use row_query in show processlist. Pull Request resolved: facebook#953 Reviewed By: lloyd Differential Revision: D14137212 Pulled By: lth
…acebook#965) (facebook#965) Summary: Jira ticket: https://jira.percona.com/browse/FB8-116 Jira ticket: https://jira.percona.com/browse/FB8-152 Jira ticket: https://jira.percona.com/browse/FB8-83 Jira ticket: https://jira.percona.com/browse/FB8-125 Reference commit: facebook@316e0e5 Reference commit: facebook@a85b8d3 Reference commit: facebook@21edb26 Reference commit: facebook@96de6c4 Reference commit: facebook@e1cb53ba0f9 Reference commit: facebook@bd25e29 Reference commit: facebook@e392a18d40e Reference commit: facebook@fa4f588a248 Reference commit: facebook@68f8c9a58fa Reference commit: facebook@4519ab421f8 Reference commit: facebook@42bdfc8 Reference commit: facebook@3950491 Reference commit: facebook@bc17d30 Reference commit: facebook@eb7b1df Reference commit: facebook@6a27bd42dfe Reference commit: facebook@9503743bf59 Reference commit: facebook@7b270a0ad3b Reference commit: facebook@e6f031af712 Reference commit: facebook@7a0da2c9e94 Reference commit: facebook@bb9f4aae4fd Porting notes: The deadlock/ASAN bugfixes were skipped - I wasn't able to reproduce the issue mentioned in the commit messages even without the patches, and the pathes themselves caused early guaranteed crashes in the server. -------- 316e0e5 -------- If enabled, For every trx before the commit event is written we write a rows query event which contains meta data in JSON format wrapped inside a comment. Storing timestamps in the meta data to calculate slave lag in millisecond granularity. -------- a85b8d3 -------- Supporting binlog trx meta data for SBR. The meta data event is written right before every query event in a trx representing a statement. -------- 21edb26 -------- Milli-seconds behind master was not calculated correctly for MTS. With this change we calculate it just like seconds behind master. One difference between these is that sec timestamp is available for every event but msec timestamp is only available once per query/rows event in form of trx metadata inside a Rows_query_event. -------- bd25e29 -------- A separate event for trx metadata was causing an unreasonable increase in binlog size. Also, now we're not writing rows query event if there are no comments in the query and `log_only_query_comments` is enabled. -------- eb7b1df -------- Log meta data event before logging the rows query event for a statement. Pull Request resolved: facebook#965 -------- b8a600a -------- When reset_seconds_behind_master is disabled, in the case where last_master_timestamp is still 0, check to determine if the I/O thread is running before returning 0 or NULL. In the case where replication is broken, the sbm value should be ever increasing, or set to NULL, and not set to 0. Differential Revision: D14652380 Pulled By: yizhang82
Summary: When database or table filter was specified along with verbose mode, mysqlbinlog would print rows_query events of filtered transactions. Fixed that behavior in this change by buffering the rows_query event until the filtering decision is taken in the following rows or query event. Note that the "# at <pos>" comment in the output of mysqlbinlog was always broken for buffered events (like int_var and user_var events). This diff does not fix that behavior for rows_query. The result is that the "at" comment for buffered events are printed consecutively before the actual events. For 8.0: the binlog_gtid.binlog_rows_query_with_filter requires the --binlog-trx-meta-data patch to be ported, so it will currently fail. Reference Patch: facebook@b45040ad5a Differential Revision: D10219146
…acebook#955) (facebook#955) Summary: JIRA : https://jira.percona.com/browse/FB8-102 Referenced patch : facebook@ab9d60b When InnoDB opens a .ibd or log file in O_DIRECT mode, request the os to flush the buffer cache. Otherwise, direct IO is serialized when buffered pages for the file existing within the filesystem cache. Originally Reviewed By: alxyang Pull Request resolved: facebook#955 Reviewed By: lloyd Differential Revision: D14034476 Pulled By: lth
…#947) (facebook#947) Summary: Jira issue: https://jira.percona.com/browse/FB8-49 Reference patch: facebook@915b131 MySQL client doesn't remove padding chars for binary columns. Do the same for RBR. Tools parsing the replication stream vs. the client output can then use memcmp to compare keys. Otherwise, the tools need to handle padding characters correctly. Pull Request resolved: facebook#947 Differential Revision: D13972586 Pulled By: hermanlee
) Summary: Jira ticket: https://jira.percona.com/browse/FB8-71 Reference Patch: facebook@4bb6dfa Reference Patch: facebook@ee92950 adds option to show last set GTID in SHOW MASTER LOGS command Unconditionally send gtid in SHOW BINARY LOGS WITH GTID Pull Request resolved: facebook#954 Reviewed By: lloyd Differential Revision: D14137247 Pulled By: lth
…facebook#960) (facebook#960) Summary: Jira issue: https://jira.percona.com/browse/FB8-50 Jira issue: https://jira.percona.com/browse/FB8-51 Reference Patch: facebook@8d520fd Reference Patch: facebook@0e62abd Notice: Network errors per user for I_S.USER_STATISTICS were dropped. Currently there are no stats to capture the network errors while communicating with the client. These stats will help better understand the server perspective while debugging client side issues. Pull Request resolved: facebook#960 Differential Revision: D14193990
…vailable (facebook#958) (facebook#958) Summary: Jira ticket: https://jira.percona.com/browse/FB8-76 Reference Patch: facebook@270854f Reference Patch: facebook@9f00827 Reference Patch: facebook@3e50e46 Reference Patch: facebook@9f18725 When a server is in read_only, we append the current master_host and master_port to the error message, if the server is a slave. The master_host could be either hostname or ip address, depending on how a slave is connected to the master. A new generic error code `ER_OPTION_PREVENTS_STATEMENT_EXTRA_INFO` is added, so that we can append any additional info to the fixed error message. Proposing a way to set a custom message in the `--read-only (super)` error. Note: the variable can only be set when server is in read-only (super) state. p.s. I am hesitant to add a mutex for the variable, so we will get a snapshot of the current pointer of the variable (const char*) when printing the error. Per request, we allow setting read_only_error_msg_extra without being in read_only state. The custom message can be set/reset at any time. D61383 added a new error code to allow appending extra info to read_only error message. The new error code breaks existing dependencies which expects to receive code 1290. Reverting back to use the old error, and extending the 1290 error message instead of adding a new one. This involves updating many files (which was the motivation I added a new one in D61383). Ideally we should have a more elegant solution to send extra info back to clients, which will be worked on separately. This is to fix the broken dependencies. Pull Request resolved: facebook#958 Reviewed By: lloyd Differential Revision: D14137345 Pulled By: lth
) Summary: Jira issue: https://jira.percona.com/browse/FB8-136 Reference Patch: facebook@6168b96 This branch unifies behavior when server faces disk full issue. Previously a server was waiting until disk is free for direct binlog writes but reported an error to a client if it hits the issue while writing to binlog cache which is written to a separate folder and might be mounted to a different partition. Originally Reviewed By: abhinav04sharma Pull Request resolved: facebook#964 Differential Revision: D14312146 Pulled By: hermanlee
) (facebook#961) Summary: Jira issue: https://jira.percona.com/browse/FB8-61 Reference Patch: facebook@c2dbe68 Notice: Changes in `ha_rocksdb.cc` and files in `mysql-test/suite/rocksdb` were not ported. Currently, on lock timeout, we only give ERROR 1205 (HY000): Lock wait timeout exceeded; try restarting transaction which has no information on what type of lock timed out. This change augments the message with the lock type to say something like ERROR 1205 (HY000): Lock wait timeout exceeded; try restarting transaction: Timeout on index: test/t1.GEN_CLUST_INDEX which shows which type of lock timed out as well as the associated object. There two cases to handle here. For error messages coming from the storage engine, the error is stored onto the transaction, which will get forwarded to the SQL layer for display through `handler::get_error_message`.. For MDL locks in the SQL layer, we are directly using the information in the MDL lock to display the error message. I didn't touch NDB because I didn't bother getting that to build. Pull Request resolved: facebook#961 Differential Revision: D14137402 Pulled By: hermanlee
…k#957) (facebook#957) Summary: Jira ticket: https://jira.percona.com/browse/FB8-134 Reference patch: facebook@0fc059a ------ 0fc059a ------ On server shutdown, there exists a race condition that can trigger a segfault if an audit plugin is installed. The root cause because the `THD::audit_class_plugins` array is accessed concurrently without locks. This is the sequence of actions that can trigger a crash: 1. The shutdown thread calls close_connections on all connected clients. 2. The connection thread will call into the audit plugin because it will receive a network error from `close_connections`, and will log the error into the audit plugin. The thread then adds the plugin to `THD::audit_class_plugins`, but it has not updated `audit_class_mask` yet. 3. The shutdown thread sees that `audit_class_mask` is not set, and also adds the plugin into `THD::audit_class_plugins`, but it has only extended the array, without setting the value of the 2nd element of the array. 4. The connection thread then goes ahead and updates `audit_class_mask` and finishes `acquire_plugins`. It then calls `mysql_audit_notify` and sees the array of size 2, and when it loops over the 2nd element, it segfaults because the value is uninitalized. The fix is to add a lock per THD that protects these data structures. They should be relatively uncontended since the shutdown thread is the only thread that calls `acquire_plugins` on another THD. Pull Request resolved: facebook#957 Differential Revision: D14236916 Pulled By: hermanlee
Summary: We still have cases where we want to connect to mysql via a port without grants. Differential Revision: D14333193
…k#946) (facebook#946) Summary: JIRA: https://jira.percona.com/browse/FB8-48 Reference Patch: facebook@8225c64 Reference Patch: facebook@f5466d6 Reference Patch: facebook@87e3650 Port the slave_run_triggers_for_rbr feature from mariadb 10.1.1. When using statement based replication slave executes the sql statments which runs the slave side triggers. Since in row based replication, slave applies the row events directly to the storage engine, triggers on the slave table are not executed. Add functionality to run triggers on slave side when executing row based events. The following triggers are invoked: * Update_row_event runs an UPDATE trigger * Delete_row_event runs a DELETE trigger * Write_row_event action depends on whether the operation will require foreign key checks: 1) when FK checks are not necessary, the operation will invoke a DELETE trigger if the record to be modified existed in the table. After that, an INSERT trigger will be invoked. 2) when FK checks are necessary, either an UPDATE or or a combination of DELETE and INSERT triggers will be invoked. slave_run_triggers_for_rbr option controls the feature. Default value is NO which don't invoke trigger for row-based events; Setting the option to YES will cause the SQL slave thread to invoke triggers for row based events; setting it to LOGGING will also cause the changes made by the triggers to be written into the binary log. There is a basic protection against triggers being invoked both on the master and slave. If the master modifies a table that has triggers, it will produce row-based binlog events with the "triggers were invoked for this event" flag. The slave will not invoke any triggers for flagged events. optionally disable binlogging while executing triggers Online schema change (OSC) creates triggers only on master side. Since RBR logs row changes made by triggers, sql_thread will hit errors due to missing tables on slaves. Add a session variable SQL_LOG_BIN_TRIGGERS to optionally disable binlogging for trigger statements so that RBR and OSC are fine with each other. disable_sql_log_bin_triggers flag is added in the TABLE_LIST struct to track tables that are opened during trigger execution. This flag is used to skip writing Table_map_log_events for such tables. With sql_log_bin_triggers enabled, the trigger changes which may be necessary for slave are not propagated. To avoid this, slave_run_triggers_for_rbr option must be enabled on slave. Pull Request resolved: facebook#946 Reviewed By: lloyd Differential Revision: D13972562 Pulled By: lth
… FULL (facebook#942) (facebook#942) Summary: JIRA: https://jira.percona.com/browse/FB8-82 Reference Patch: facebook@b1fec60 Some update and upsert (insert ... on duplicate key update) queries used to log columns which have not changed in the after image (when using RBR) even when the image format is not FULL. Pull Request resolved: facebook#942 Reviewed By: lloyd Differential Revision: D13930763 Pulled By: lth
… while there are active un-acked trxns (facebook#966) (facebook#966) Summary: Jira issue: https://jira.percona.com/browse/FB8-131 Reference Patch: facebook@2d47615 Reference Patch: facebook@d1ea741 Reference Patch: facebook@f5085f4 Reference Patch: facebook@9204c65 Option to shutdown master if semi-sync is being switched off while there are active un-acked trxns Semi-sync master can be switched off manually or because of a timeout while waiting for an ack. This change introduces a new variable rpl_semi_sync_master_crash_if_active_trxs which when set to true will cause the master to shutdown (after printing an error msg) if semi-sync is being switched off while there are active un-acked transactions. This prevents any un-acked trx from committing on the master. Also delete pid file when crashing after timeout with unacked trx. If rpl_semi_sync_master_crash_if_active_trxs = true, the server crashes after a timeout if there are any active unacked transactions. mysqld_safe expects the pid file to be deleted if the expectation is to not bring the intance back up again. Pull Request resolved: facebook#966 Differential Revision: D14308888 Pulled By: hermanlee
) (facebook#979) Summary: Jira issue: https://jira.percona.com/browse/FB8-186 Reference Patch: facebook@16b02494411 This will allow us to override some glibc functions like localtime_r. Originally Reviewed By: hermanlee Pull Request resolved: facebook#979 Reviewed By: lloyd Differential Revision: D14335103 Pulled By: lth
facebook#959) Summary: Jira ticket: https://jira.percona.com/browse/FB8-53 Reference Patch: facebook@f29eb03 Reference Patch: facebook@c631017 'THD' class extended with the following convenience methods: - 'has_net_vio()' - 'get_net_vio()' - 'has_net_vio_ssl_arg()' - 'get_net_vio_ssl_arg()' - 'get_peer_certificate_info()' Implemented new 'Fill_authinfo_list' class derived from 'Do_THD_Impl' which populates 'INFORMATION_SCHEMA.AUTHINFO' table. This is a change to help the implementation of SSL based authentication efforts. Currently the certificate details are unavailable to the sys admin. The v3 extensions usually carry important information for identifying the user. This change adds a new table to the information schema to expose auth details. It is organized by process ID like processlist table. Pull Request resolved: facebook#959 Reviewed By: lloyd Differential Revision: D14137362 Pulled By: lth
…cebook#967) (facebook#967) Summary: Jira ticket: https://jira.percona.com/browse/FB8-167 Reference Patch: facebook@b01ff6a Currently, when force index is used, full table scans are highly penalized, but it is still possible to do a full index scan which is expensive. In many cases, we only use force index when we know there is better way to traverse the index, and not necessarily because we want a full index scan. To address these cases, add a session variable called optimizer_force_index_for_range, which when turned on, will try to search for a 'range' plan before falling back to an index plan. Pull Request resolved: facebook#967 Reviewed By: lloyd Differential Revision: D14387089 Pulled By: lth
…facebook#950) Summary: JIRA: https://jira.percona.com/browse/FB8-84 Reference Patch: facebook@478fa2a Reference Patch: facebook@8a269ee4f97 Reference Patch: facebook@c53bf248a27 Reference Patch: facebook@ae0d38ae0c8 The new global variable `max_nonsuper_connections` will enforce the limit for the sum of all non-admin connections. This will be useful to limit regular user connections while still allow super users to connect to server. Pull Request resolved: facebook#950 Reviewed By: lloyd Differential Revision: D13987618 Pulled By: lth
…g can be rollbacked (facebook#977) (facebook#977) Summary: Jira issue: https://jira.percona.com/browse/FB8-105 Reference Patch: facebook@cd39ae1 Reference Patch: facebook@0623a07 Stop slave immediately in MTS if partial trx in the relay log can be rollbacked In MTS stop slave can take a minute to complete if the last transaction is partially downloaded from the master. The slave waits for a minute for the master to send the rest of the trx and then finally gives up. Strictly, this wait is only required if the partial trx cannot be rollbacked safely (e.g. trx on non-transactional engine, DDLs etc.). This change checks if there are no jobs queued in the worker threads and if the partial trx can be rollbacked, if yes, it immediately stops the slave. Originally Reviewed By: tianx Pull Request resolved: facebook#977 Reviewed By: lloyd Differential Revision: D14387911 Pulled By: lth
…ebook#982) (facebook#982) Summary: Jira issue: https://jira.percona.com/browse/FB8-128 Reference Patch: facebook@2dd16b3 Reference Pull: facebook#681 Originally Reviewed By: lth Pull Request resolved: facebook#982 Reviewed By: lloyd Differential Revision: D14421316 Pulled By: lth
…r of rows in a single rows event (facebook#974) (facebook#974) Summary: Jira issue: https://jira.percona.com/browse/FB8-187 Reference Patch: facebook@dfe6746cdfd Number of rows in a single event can be indirectly controlled with binlog_rows_event_max_size. This new variable will give more direct control to the user. Originally Reviewed By: anirbanr-fb Pull Request resolved: facebook#974 Reviewed By: lloyd Differential Revision: D14421324 Pulled By: lth
facebook#970) Summary: Jira issue: https://jira.percona.com/browse/FB8-172 Reference Patch: facebook@b90e801 Reference Patch: facebook@ecef4b2 There are cases where it is always unacceptable for a client to be doing full table scans. To prevent this from happening, add a new variable optimizer_full_scan. When it is turned off, it will return a new error code ER_FULL_SCAN_DISABLED. For best results, use with optimizer_force_index_for_range. Currently, it does not fallback to alternative plans because we check at the end of query planning. EXPLAIN queries with derived tables do not populate select_options in the JOIN structure properly. This means that we could error if we tried to run an explain on a query with derived tables. Instead of reading from the JOIN struct, read from the LEX struct on the THD, as this is where the flag is original set on during query parsing. This bug also means that in upstream, we increment some status variables tracking full table scans despite the fact that only an explain statement was done. This seems to have been fixed in 8.0 though. Pull Request resolved: facebook#970 Reviewed By: lloyd Differential Revision: D14567846 Pulled By: lth
Summary: Jira issue: https://jira.percona.com/browse/FB8-139 Reference Patch: facebook@18c1e44 Enables traces to be passed in as query attributes by clients and traces (instruction costs) to be collected on server side. Originally Reviewed By: abhinav04sharma Pull Request resolved: facebook#981 Reviewed By: lloyd Differential Revision: D14387874 Pulled By: lth
inikep
pushed a commit
that referenced
this pull request
May 26, 2023
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 fbshipit-source-id: d96ebcef966 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 fbshipit-source-id: 8e7fdb8 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
Jun 1, 2023
inikep
pushed a commit
that referenced
this pull request
Jun 1, 2023
Summary: Rebaseline type relate tests: 1. SHOW INDEX format changes 2. No more LSM_TREE in Index_type as the corresponding 3. More rows in EXPLAIN and rows column is now in column #10 (instead of #9) and there is another filtered column that needs to be masked 4. Using utf8 results warnings but I decided to keep the warning as a reminder in the future. TODO: we need to decide what to do with binary unpacking with regarding to utf8/utf8mb3/utf8mb4. 5. YEAR(2) is deprecated 6. GROUP BY now needs ORDER BY for ordered output 7. MySQL 8 no longer allows specifying NULL for PK so I'm adding a extra_pk_col_opts to the col_opt_*.tests to only specify supported flags for PK 8. Binary output are now hex instead of character (for example, 0 => 0x30) Reviewed By: lloyd Differential Revision: D17503619
inikep
pushed a commit
that referenced
this pull request
Jun 1, 2023
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 fbshipit-source-id: d96ebcef966 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 fbshipit-source-id: 8e7fdb8 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
Jun 14, 2023
inikep
pushed a commit
that referenced
this pull request
Jun 14, 2023
Summary: Rebaseline type relate tests: 1. SHOW INDEX format changes 2. No more LSM_TREE in Index_type as the corresponding 3. More rows in EXPLAIN and rows column is now in column #10 (instead of #9) and there is another filtered column that needs to be masked 4. Using utf8 results warnings but I decided to keep the warning as a reminder in the future. TODO: we need to decide what to do with binary unpacking with regarding to utf8/utf8mb3/utf8mb4. 5. YEAR(2) is deprecated 6. GROUP BY now needs ORDER BY for ordered output 7. MySQL 8 no longer allows specifying NULL for PK so I'm adding a extra_pk_col_opts to the col_opt_*.tests to only specify supported flags for PK 8. Binary output are now hex instead of character (for example, 0 => 0x30) Reviewed By: lloyd Differential Revision: D17503619
inikep
pushed a commit
that referenced
this pull request
Jun 14, 2023
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 fbshipit-source-id: d96ebcef966 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 fbshipit-source-id: 8e7fdb8 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
Jun 19, 2023
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 fbshipit-source-id: d96ebcef966 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 fbshipit-source-id: 8e7fdb8 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
Jun 23, 2023
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 fbshipit-source-id: d96ebcef966 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 fbshipit-source-id: 8e7fdb8 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
Apr 25, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
May 7, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
May 8, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
May 9, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
May 10, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
May 13, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
May 15, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
May 16, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
May 17, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
May 17, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
May 21, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
May 21, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
May 30, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
Jun 28, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
Jul 2, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
Jul 19, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
Jul 19, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
Jul 30, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
Jul 31, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
Aug 2, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
inikep
pushed a commit
that referenced
this pull request
Aug 6, 2024
Summary: [Porting Notes] We want to dump raft logs to vanilla async replicas regardless of whether it's the relay log or binlog. Effectively after this change we'll dump relay logs on the followers and binlogs on the leader. When the raft role changes, the logs to the dumped are also changed. Dump_log class is introduced as a thin wrapper/continer around mysql_bin_log or rli->relay_log and is inited with mysql_bin_log to emulate vanilla mysql behavior. Dump threads use the global dump_log object instead of mysql_bin_log directly. We switch the log in dump log only when raft role changes (in binlog_change_to_binlog() and binlog_change_to_apply_log()). During raft role change we take all log releated locks (LOCK_log, LOCK_index, LOCK_binlog_end_pos, and dump log lock) to serialize it with other log operations like dumping logs. Related doc - https://fb.quip.com/oTVAAdgEi4zY This diff contains below 7 patches: D23013977 D24766787 D24716539 D24900223 D24955284 D25174166 D25775525 Reviewed By: luqun Differential Revision: D26141496 ------------------------------------------------------------------------------- Passing raw_log pointer to wait_with_heartbeat() and wait_without_heartbeat() Summary: When enable_raft_plugin is OFF Dump_log::lock() is a no-op. Which means that when enable_raft_plugin is OFF there can be a race between log switching and dump threads. This could lead to a scenario where the raw_log that wait_next_event() is working on might be different than what wait_with_heartbeat()/wait_without_heartbeat() is working on. This can cause deadlocks because wait_with_heartbeat()/wait_without_heartbeat()'s mysql_cond_wait would unlock and then lock a different log's LOCK_binlog_end_pos mutex which would then never be unlocked by wait_next_event(). Reviewed By: anirbanr-fb Differential Revision: D32152658 ----------------------------------------------------------------------------------------- Fix rpl_raft_dump_raft_logs Summary: This tests completes but fails because the following warning exists: ``` 2022-08-30T16:28:00.159525Z 11 [ERROR] [MY-013114] [Repl] Slave I/O for channel '': Got fatal error 1236 from master when reading data from binary log: 'Slave has more GTIDs than the master has, using the master's SERVER_UUID. This may indicate that the end of the binary log was truncated or that the last binary log file was lost, e.g., after a power or disk failure when sync_binlog != 1. The master may or may not have rolled back transactions that were already replicated to the slave. Suggest to replicate any transactions that master has rolled back from slave to master, and/or commit empty transactions on master to account for transactions that have been', Error_code: MY-013114 ``` Since the MTR result file is valid, we can suppress this error. Reviewed By: yichenshen Differential Revision: D39141846 ------------------------------------------------------------------------------- Fix heap overflow in group_relay_log_name handling Summary: We were accessing group_relay_log_name in Query_log_event::do_apply_event_worker() but it's assigned only after the coordinator thread encounters an end event (i.e. xid event or a query event with "COMMIT" or "ROLLBACK" query). This was causing a race between accessing group_relay_log_name in the worker thread and writing it on the coordinator thread. We don't need to set transaction position in events other than end event, so now we set transaction position in query event only if it's an end event. The race is eliminated because group_relay_log_name is set before enqueuing the event to the worker thread (in both dep repl and vanilla mts). Reviewed By: lth Differential Revision: D28767430 ------------------------------------------------------------------------------- fix memory during MYSQL_BIN_LOG::open_existing_binlog Summary: asandebug complain there are memory leaks during MYSQL_BIN_LOG open Direct leak of 50 byte(s) in 1 object(s) allocated from: #0 0x67460ef in malloc #1 0x93f0777 in my_raw_malloc(unsigned long, int) #2 0x93f064a in my_malloc(unsigned int, unsigned long, int) #3 0x93f0eb0 in my_strdup(unsigned int, char const*, int) #4 0x8af01a6 in MYSQL_BIN_LOG::open(unsigned int, char const*, char const*, unsigned int) #5 0x8af8064 in MYSQL_BIN_LOG::open_binlog(char const*, char const*, unsigned long, bool, bool, bool, Format_description_log_event*, unsigned int, RaftRotateInfo*, bool) #6 0x8b00c00 in MYSQL_BIN_LOG::new_file_impl(bool, Format_description_log_event*, RaftRotateInfo*) #7 0x8d65e47 in rotate_relay_log(Master_info*, bool, bool, bool, RaftRotateInfo*) #8 0x8d661c0 in rotate_relay_log_for_raft(RaftRotateInfo*) #9 0x8c7696a in process_raft_queue #10 0xa0fa1fd in pfs_spawn_thread(void*) #11 0x7f8c9a12b20b in start_thread release these memory before assign them Reviewed By: Pushapgl Differential Revision: D28819752
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Both errors are caused by filesystem permissions.