Percona XtraDB Cluster limitations¶

The following limitations apply to Percona XtraDB Cluster:

Replication works only with InnoDB storage engine.

Any writes to tables of other types are not replicated.
Unsupported queries:

LOCK TABLES and UNLOCK TABLES is not supported in multi-source setups

Lock functions, such as GET_LOCK(), RELEASE_LOCK(), and so on
Query log cannot be directed to table.

If you enable query logging, you must forward the log to a file:
```
log_output = FILE
```
Use general_log and general_log_file to choose query logging and the log file name.
Maximum allowed transaction size is defined by the wsrep_max_ws_rows and wsrep_max_ws_size variables.

LOAD DATA INFILE processing will commit every 10 000 rows. So large transactions due to LOAD DATA will be split to series of small transactions.
Transaction issuing COMMIT may still be aborted at that stage.

Due to cluster-level optimistic concurrency control, there can be two transactions writing to the same rows and committing in separate Percona XtraDB Cluster nodes, and only one of them can successfully commit. The failing one will be aborted. For cluster-level aborts, Percona XtraDB Cluster gives back deadlock error code:
Error message
```
(Error: 1213 SQLSTATE: 40001  (ER_LOCK_DEADLOCK)).
```
XA transactions are not supported

Due to possible rollback on commit.
Write throughput of the whole cluster is limited by the weakest node.

If one node becomes slow, the whole cluster slows down. If you have requirements for stable high performance, then it should be supported by corresponding hardware.
Minimal recommended size of cluster is 3 nodes.

The 3^rd node can be an arbitrator.
enforce_storage_engine=InnoDB is not compatible with wsrep_replicate_myisam=OFF

wsrep_replicate_myisam is set to OFF by default.
Avoid ALTER TABLE ... IMPORT/EXPORT workloads when running Percona XtraDB Cluster in cluster mode.

It can lead to node inconsistency if not executed in sync on all nodes.
All tables must have a primary key.

This ensures that the same rows appear in the same order on different nodes. The DELETE statement is not supported on tables without a primary key.

See also

MariaDB Galera Documentation: Tables without Primary Keys
When configuring Percona XtraDB Cluster, your server must create a local socket.

You can set up a socket by providing a path, or you can skip creating one explicitly. However, do not leave the variable in my.cnf empty, like this: socket=. If you do, the server won’t create a socket.

State Snapshot Transfer (SST) requires the socket for the following tasks:
- Taking backup using Percona XtraBackup
- Detecting keyring component status
Avoid reusing the names of persistent tables for temporary tables

Although MySQL does allow having temporary tables named the same as persistent tables, this approach is not recommended.

Galera Cluster blocks the replication of those persistent tables the names of which match the names of temporary tables.

With wsrep_debug set to 1, the error log may contain the following message:
Error message
```
... [Note] WSREP: TO BEGIN: -1, 0 : create table t (i int) engine=innodb
... [Note] WSREP: TO isolation skipped for: 1, sql: create table t (i int) engine=innodb.Only temporary tables affected.
```
See also

MySQL Documentation: Problems with temporary tables
Audit Log Filter state synchronization

Percona XtraDB Cluster supports the Audit Log Filter component. While the underlying configuration tables (mysql.audit_log_filter and mysql.audit_log_user) are replicated to all nodes via Galera, the in-memory state of the audit component is local to each node.

To ensure consistent auditing across the cluster, keep the following limitations in mind:
- Manual Cache Refresh: Running an INSERT, UPDATE, or DELETE on the audit filter tables on one node replicates the data, but does not trigger a cache refresh on the other nodes.
- Required Action: You must manually execute SELECT audit_log_filter_flush(); on every node in the cluster after modifying the filter tables to ensure the nodes are using the same rules.
- If the mysql.audit_log_filter or mysql.audit_log_user tables are modified using one of the following user-defined functions (UDFs), then audit_log_filter_flush() must also be run on all the other nodes:
  - audit_log_filter_set_filter
  - audit_log_filter_set_user
  - audit_log_filter_remove_filter
  - audit_log_filter_remove_user
- Session Persistence: The audit log filter is applied at the start of a connection. Even after flushing the cache, existing sessions will continue to be governed by the previous filtering rules. Users must disconnect and reconnect for new filters to take effect.
See also

For more information on the audit log filter, see the Percona Server 8.0 Audit Log Filter overview.

As of version 8.0.21, an INPLACE ALTER TABLE query takes an internal shared lock on the table during the execution of the query. The LOCK=NONE clause is no longer allowed for all of the INPLACE ALTER TABLE queries due to this change.

This change addresses a deadlock, which could cause a cluster node to hang in the following scenario:

An INPLACE ALTER TABLE query in one session or being applied as Total Order Isolation (TOI)
A DML on the same table from another session

Do not use one or more dot characters (.) when defining the values for the following variables:

MySQL and Percona XtraBackup handle the character in different ways and this difference causes unpredictable behavior.