A.5. Tungsten Clustering 6.0.0 GA (4 April 2018)

Version End of Life. 4 April 2021

Tungsten Clustering 6.0.0 is a major update to the operation and deployment of multi-master and composite clusters. Within the new framework, a composite multi-master cluster is configured as follows:

  • Clusters within a composite cluster are now managed in a unified fashion, including the overall replication progress across clusters.

  • Cross-site replicators are configured as additional services within the main replicator.

  • Cross-site replicators are managed by the manager as part of a complete composite cluster solution.

  • A new global progress counter displays the current progress for the local and cross-site replication.

  • Connectors are configured by default to provide affinity for the local, and then the remote cluster.

The cluster package name has been changed, and upgrades from older versions to the new configuration and layout are supported.

Behavior Changes

The following changes have been made to Tungsten Clustering and may affect existing scripts and integration tools. Any scripts or environment which make use of these tools should check and update for the new configuration:

  • A new unified cluster deployment is available, the Composite Multimaster. This is an updated version of the Multi-site/Multi-master deployment in previous releases. It encompasses a number of significant changes and improvements:

    • Single, cluster-based, deployment using the new deployment type of composite-multi-master.

    • Unified multimaster cluster status within cctrl.

    • Global progress counter indiciting the current cluster and cross-cluster performance.

    Issues: CT-105, CT-313, CT-431, CT-467

  • The name of the cluster deployment package for Tungsten Clustering has changed. Packages are now named to match the product, for example, tungsten-clustering-6.0.4-27.tar.gz.

    Issues: CT-271, CT-438

  • Support for using Java 7 with Tungsten Clustering has been removed. Java 8 or higher must be used for all deployments.

    Issues: CT-450

  • The behavior of the cctrl has changed to operate better within the new composite deployments. Without the -multi argument, cctrl will cd into the local standalone service. This matches the previous releases for cctrl, but instead all services are still accessible without needing to use the -multi option. With the -multi argument, cctrl will not automatically cd into the local standalone service but will show all available services.

    Issues: CT-524

  • Due to the change in the nature of the services and clustering within SOR and multimaster configurations, the tungsten_provision_slave command has been updated to support cross-cluster provisioning. Because there would now be a conflict of service names, a cross cluster provision should use the --force option. The --service option should still be set to the local service being reset. For example:

    shell> tungsten_provision_slave --source=db4 --service=east --direct --force

    Issues: CT-567

Known Issue

The following issues may affect the operation of Tungsten Clustering and should be taken into account when deploying or updating to this release.

  • During an upgrade installation from a v4 or v4 MSMM deployment, you may get additional, empty, schemas creates within your MySQL database. These schemas are harmless and can safely be removed. For example, if you have two services in your MSMM deployment, east and west, during the upgrade you will get two empty schemas, tungsten_east_from_west and tungsten_west_from_east.

    This will be addressed in a future release.

    Issues: CT-559

  • During a switch operation on a multi-master cluster when the cluster has been put into maintenance mode, the manager will put the cross-site replicators back into the online state.

    This will be addressed in a future release.

    Issues: CT-591

  • When performing a tpm update operation to change the configuration and the cluster is in AUTOMATIC mode, the update will complete correctly but the cluster may be left in MAINTENANCE mode instead of being placed back into AUTOMATIC mode.

    This will be addressed in a future release.

    Issues: CT-595

  • When performing a tpm update in a cluster with an active witness, the host with the witness will not be restarted correctly resulting in the witness being down on that host.

    This will be addressed in a future release.

    Issues: CT-596

  • In a composite multimaster cluster deployment where there are three or more clusters, a failure in the MySQL server in one node in a clsuter could fail to be identified, and ultimately the failover within the environment to fail, either within the cluster or across clusters.

    This will be addressed in a future release.

    Issues: CT-619

Improvements, new features and functionality

  • Installation and Deployment

    • A new utility script, tungsten_prep_upgrade has been provided as part of the standard installation. The script is specifically designed to assist during the upgrade of a multi-site/multi-master deployment from 5.3.0 and earlier to the new Multimaster 6.0.0 deployment.

      Issues: CT-104

  • Command-line Tools

    • The cctrl command now includes a show topology command, that outputs the current toplogy for the cluster or component being viewed.

      Issues: CT-429

    • The tpm diag command has been extended to include multimaster cluster status information, one for each configured service and cross-site service.

      Issues: CT-594

  • Tungsten Connector

    • By default, within composite multi-master clusters, the affinity for the connector is configured to connect to the master for the site on which the connector lives first and then, if that master is not available, connect to the other site

      Issues: CT-448

Bug Fixes

  • Command-line Tools

    • The mm_tpm diag command could complain that an extra replicator is configured and running, even though it would be valid as part of a multi-master deployment.

      Issues: CT-396

    • The mm_trepctl command could fail to display any status information while obtaining the core statistic information from each host.

      Issues: CT-437

  • Tungsten Manager

    • When performing a recover or switch operation within maintenance mode, the cluster would automatically revert to automatic mode just before and immediately after a switch, which could lead to problems correctly recovering a cluster.

      Issues: CT-472

Tungsten Clustering 6.0.0 Includes the following changes made in Tungsten Replicator 6.0.0

Release 6.0.0 is a feature and bugfix release. This release contains the following key fixes:

  • Added PostgreSQL applier support.

  • Added support for custom primary key field selection for source tables that cannot be configured with a primary key within the database.

  • Added a new filter for including whole of transaction metadata information into each event.

  • Added support for extended transaction information within the Kafka applier so that all the messages for a given transaction can be identified.

Behavior Changes

The following changes have been made to Continuent Replicator and may affect existing scripts and integration tools. Any scripts or environment which make use of these tools should check and update for the new configuration:

  • Support for using Java 7 with Tungsten Clustering has been removed. Java 8 or higher must be used for all deployments.

    Issues: CT-450

Improvements, new features and functionality

  • Heterogeneous Replication

    • The Kafka applier now supports the inclusion of transaction information into each Kafka message broadcast, including the list of schema/tables and row counts for the entire transaction, as well as information about whether the message is the first or last message/row within an overall transaction. The transaction information can also be sent as a separate message on an independent Kafka topic.

      Issues: CT-496, CT-586

      For more information, see Optional Configuration Parameters for Kafka (in [Tungsten Replicator 6.0 Manual]).

  • Core Replicator

    • Experimental support for writing row-based data through SQL into PostgreSQL has been added back to the replicator. This includes basic support fr the replication of the data. Currently databases and tables must be created by hand. A future release will incorporate full support for DDL translation.

      Issues: CT-149

  • Filters

    • The pkey has been extended to support the specification of custom primary key fields. This enables fields in the source data to be marked as primary keys even if the source database does not have the keys specified. This is useful for heterogeneous loading of data where a unique key may exist, but cannot be defined due to the application or database that created the tables.

      Issues: CT-481

      For more information, see ???.

    • A new filter, rowaddtxninfo has been added which embeds row counts, both total and per schema/table, to the metadata for a THL event/transaction.

      Issues: CT-497

Bug Fixes

  • Installation and Deployment

    • When performing a tpm reverse, the --replication-port setting would be replaced with it's alias, --oracle-tns-port.

      Issues: CT-597

  • Core Replicator

    • An internal optimization within the replicator that would attempt to optimise row-based information and operations has been removed. The effects of the optimization were actually seen in very few situations, and it duplicated work and operations performed by the pkey filter. Unfortunately the same optimization could also cause issues within heterogeneous deployments with the removal of information.

      Issues: CT-318

    • The internal storage of the MySQL server ID has been updated to support larger server IDs. This works with any MySQL deployment, but has been specifically expanded to work better with some cloud deployments where the server ID cannot be controlled.

      Issues: CT-439

    • The format of some errors and log entries would contain invalid characters.

      Issues: CT-493