Michael Widenius, founder of MySQL, has announced the availability of the first stable release of the cluster database MariaDB Galera, based on the code base and technology MariaDB synchronous multi-master replication Galera.
The project was created in partnership with Codership, developing system Galera, and is positioned as a competitor to Oracle cluster solutions based on MySQL. Project code is available under the license GPLv2.
The new product extends the DBMS MariaDB, which can act as a transparent replacement of MySQL 5.5, the means for the organization of synchronous data replication across multiple nodes (used topology active-active multi-master, allowing read and write permissions for each node). With synchronous replication, all nodes always contain actual data, ie warranted orphaned transactions because the transaction is recorded only after the spread of the data across all nodes. At the same time, as part of the transaction operations are executed immediately, the delay is waiting for confirmation only occurs when performing “commit”. Transactions at different sites can be executed in parallel. When the transaction is provided guaranteed permanent status of the cluster as a whole, regardless of other currently running transactions.
Replication is performed in parallel to the row level, the transfer only the information about the changes. Management node cluster membership is automatic, bad nodes immediately removed from the cluster without administrator, if necessary, new nodes can be connected on the fly with no additional setup. A cluster can be geographically spread over several data centers, and the customer provides the minimum delay – when placing units in close proximity to the client, all read and write operations will be carried out with maximum speed, the delay will be felt only when the transaction commits.
Of the merits of MariaDB Galera also mentions the possibility of scaling operations as reading and writing. In particular, a possible distribution business writing, if nodes which have already made a record in the chain initiate recording on the remaining nodes. For example, in the first phase is recorded in parallel to the nodes of different data centers, then these nodes to replicate data within each data center, ie over a slow link is only one replication node from each group.