15.3.8.3 Restrictions on XA Transactions

XA transaction support is limited to the InnoDB storage engine.

For “external XA,” a MySQL server acts as a Resource Manager and client programs act as Transaction Managers. For “Internal XA”, storage engines within a MySQL server act as RMs, and the server itself acts as a TM. Internal XA support is limited by the capabilities of individual storage engines. Internal XA is required for handling XA transactions that involve more than one storage engine. The implementation of internal XA requires that a storage engine support two-phase commit at the table handler level, and currently this is true only for InnoDB.

For XA START, the JOIN and RESUME clauses are recognized but have no effect.

For XA END the SUSPEND [FOR MIGRATE] clause is recognized but has no effect.

The requirement that the bqual part of the xid value be different for each XA transaction within a global transaction is a limitation of the current MySQL XA implementation. It is not part of the XA specification.

An XA transaction is written to the binary log in two parts. When XA PREPARE is issued, the first part of the transaction up to XA PREPARE is written using an initial GTID. A XA_prepare_log_event is used to identify such transactions in the binary log. When XA COMMIT or XA ROLLBACK is issued, a second part of the transaction containing only the XA COMMIT or XA ROLLBACK statement is written using a second GTID. Note that the initial part of the transaction, identified by XA_prepare_log_event, is not necessarily followed by its XA COMMIT or XA ROLLBACK, which can cause interleaved binary logging of any two XA transactions. The two parts of the XA transaction can even appear in different binary log files. This means that an XA transaction in PREPARED state is now persistent until an explicit XA COMMIT or XA ROLLBACK statement is issued, ensuring that XA transactions are compatible with replication.

On a replica, immediately after the XA transaction is prepared, it is detached from the replication applier thread, and can be committed or rolled back by any thread on the replica. This means that the same XA transaction can appear in the events_transactions_current table with different states on different threads. The events_transactions_current table displays the current status of the most recent monitored transaction event on the thread, and does not update this status when the thread is idle. So the XA transaction can still be displayed in the PREPARED state for the original applier thread, after it has been processed by another thread. To positively identify XA transactions that are still in the PREPARED state and need to be recovered, use the XA RECOVER statement rather than the Performance Schema transaction tables.

The following restrictions exist for using XA transactions:

Prior to MySQL 8.0.30, XA transactions are not fully resilient to an unexpected halt with respect to the binary log. If there is an unexpected halt while the server is in the middle of executing an XA PREPARE, XA COMMIT, XA ROLLBACK, or XA COMMIT ... ONE PHASE statement, the server might not be able to recover to a correct state, leaving the server and the binary log in an inconsistent state. In this situation, the binary log might either contain extra XA transactions that are not applied, or miss XA transactions that are applied. Also, if GTIDs are enabled, after recovery @@GLOBAL.GTID_EXECUTED might not correctly describe the transactions that have been applied. Note that if an unexpected halt occurs before XA PREPARE, between XA PREPARE and XA COMMIT (or XA ROLLBACK), or after XA COMMIT (or XA ROLLBACK), the server and binary log are correctly recovered and taken to a consistent state.
Beginning with MySQL 8.0.30, this is no longer an issue; the server implements XA PREPARE as a two-phase operation, which maintains the state of the prepare operation between the storage engine and the server, and imposes order of execution between the storage engine and the binary log, so that state is not broadcast before it is consistent and persistent on the server node.
You should be aware that, when the same transaction XID is used to execute XA transactions sequentially and a break occurs during the processing of XA COMMIT ... ONE PHASE, it may no longer be possible to synchronize the state between the binary log and the storage engine. This can occur if the series of events just described takes place after this transaction has been prepared in the storage engine, while the XA COMMIT statement is still executing. This is a known issue.
The use of replication filters or binary log filters in combination with XA transactions is not supported. Filtering of tables could cause an XA transaction to be empty on a replica, and empty XA transactions are not supported. Also, with the replica's connection metadata repository and applier metadata repository stored in InnoDB tables, which became the default in MySQL 8.0, the internal state of the data engine transaction is changed following a filtered XA transaction, and can become inconsistent with the replication transaction context state.
The error ER_XA_REPLICATION_FILTERS is logged whenever an XA transaction is impacted by a replication filter, whether or not the transaction was empty as a result. If the transaction is not empty, the replica is able to continue running, but you should take steps to discontinue the use of replication filters with XA transactions in order to avoid potential issues. If the transaction is empty, the replica stops. In that event, the replica might be in an undetermined state in which the consistency of the replication process might be compromised. In particular, the gtid_executed set on a replica of the replica might be inconsistent with that on the source. To resolve this situation, isolate the source and stop all replication, then check GTID consistency across the replication topology. Undo the XA transaction that generated the error message, then restart replication.
XA transactions are considered unsafe for statement-based replication. If two XA transactions committed in parallel on the source are being prepared on the replica in the inverse order, locking dependencies can occur that cannot be safely resolved, and it is possible for replication to fail with deadlock on the replica. This situation can occur for a single-threaded or multithreaded replica. When binlog_format=STATEMENT is set, a warning is issued for DML statements inside XA transactions. When binlog_format=MIXED or binlog_format=ROW is set, DML statements inside XA transactions are logged using row-based replication, and the potential issue is not present.