Galera wsrep provider is a dlopenable generic replication library which implements the wsrep provider API. It provides very efficient synchronous replication, and multi-master and "parallel applying" capabilities for opaque data buffers. Its primary focus is data consistency. A transaction is either committed on every node, or not at all. It works anywhere TCP works, including on WAN, and can utilize UDP multicast where available. It is used by the MySQL-wsrep patch project to create true multi-master MySQL/InnoDB clusters.
|Tags||replication high availability Clustering Tools Clustering/Distributed Networks|
|Implementation||C C++ STL Boost|
Release Notes: This is the first GA release in the Galera 3.x series. Major new features include the introduction of cluster "segments" to minimize WAN communication between data centers, an optimized writeset format, 128-bit write set checksums, and support for seamless interleaving of external replication events.
Release Notes: This is a major bugfix release. A lot of bugs were fixed since 2.4, some of which could result in a deadlock or other serious malfunctions. Upgrading is recommended.
Release Notes: This release fixed possible data inconsistency due to indeterministic certification index purge and exceptions when using high evs.send_window and evs.user_send_window settings. It also sports some performance and memory optimizations for huge writesets.
Release Notes: This release contains important bugfixes, new features, and performance improvements. Fixed: causal reads guarantee violation at cluster partitioning, exclusive keys didn't produce dependency on shared keys, a race condition in the desync method (could break RSU), IST not working across different EC2 accessibility zones (and similar), and causal reads timeout configuration was ignored. Added: an wsrep_incoming_addresses status variable, the ability to specify several node addresses in a gcomm:// URL. Significant speedups and memory footprint reduction on ranged queries.
Release Notes: This release fixes broken IST between 32-bit and 64-bt architectures, a possible cluster crash on arbitrator reconnect, and a number of other minor issues. It adds support for "shared" keys, a gcs.sync_donor boolean parameter to keep state donor temporally synced with the cluster (useful for non-blocking state transfers like xtrabackup and IST), and support for Solaris 11 (x86 only).