Cluster transaction coordination

You can configure your high-availability cluster so that when a client session issues a commit, the server blocks the session until the transaction is applied in that session, on a secondary server, or across the cluster. Set the CLUSTER_TXN_SCOPE configuration parameter or run the SET ENVIRONMENT CLUSTER_TXN_SCOPE statement to configure this behavior.

Multistep client operations that are performed on different high-availability cluster servers or in different sessions with high-availability cluster servers can fail because of asynchronous log processing. If a client application loads data onto a cluster server, and then attempts to process the same data on a second cluster server before the data is replicated to the second server, the operation fails. The client transaction must be applied on the second server before its data can be further processed.

Cluster transaction coordination causes client applications to wait for either cluster-wide or secondary-server application of transactions before the transaction commits are returned. This process prevents operation failures and ensures that the steps of multistep processes occur in serial order.

The different scopes for cluster transaction are:
  • SESSION: When a client session issues a commit, the database server blocks the session until the effects of the transaction commit are returned to that session. After control is returned to the session, other sessions at the same database server or on other database servers in the cluster might be unaware of the transaction commit and the transaction's effects.
  • SERVER: When a client session issues a commit, the database server blocks the session until the transaction is applied at the database server from which the client session issued the commit. Other sessions at that database server are aware of the transaction commit and the transaction's effects. Sessions at other database servers in the cluster might be unaware of the transaction's commit and its effects. This behavior is default for high-availability cluster servers.
  • CLUSTER: When a client session issues a commit, the database server blocks the session until the transaction is applied at all database servers in the high-availability cluster, excluding RS secondary servers that are using DELAY_APPLY or STOP_APPLY. Other sessions at any database server in the high-availability cluster, excluding RS secondary servers that are using DELAY_APPLY or STOP_APPLY, are aware of the transaction commit and the transaction's effects.
Cluster transaction coordination was introduced in HCL Informix® version 11.70.xC6. Before HCL Informix® version 11.70.xC6, high-availability cluster servers had the following default behaviors:
  • Primary servers had a cluster transaction scope of SERVER.
  • Read-only secondary servers were in the dirty-read isolation level, and could read uncommitted data.
  • Updatable secondary servers had a cluster transaction scope of SESSION.

Setting a CLUSTER_TXN_SCOPE value to CLUSTER does not change the behavior that is specified by the DRINTERVAL configuration parameter value. When a client application commits a transaction on a primary server, the primary server sends the HDR secondary server logical log buffers at maximum intervals that are specified by the DRINTERVAL configuration parameter. After the primary server sends logical log buffers to the HDR secondary server, it returns control to a session, but the session still does not receive a commit until the transaction is applied on all cluster servers.