Databases Reference
In-Depth Information
row is stored redundantly, so you can lose a node without losing data, and the cluster
remains functional. Although MySQL Cluster still isn't a complete solution for every
type of application, as we mentioned in the previous chapter, it has been improved
rapidly in recent releases and now has a huge list of new features and characteristics:
disk storage for nonindexed data, online scaling by adding data nodes,
ndbinfo
tables
for managing the cluster, scripts for provisioning and managing the cluster, multi-
threaded operation, push-down joins (now called adaptive query localization), the
ability to handle BLOBs and tables with many columns, centralized user management,
and NoSQL access through the NDB API as well as the
memcached
protocol. Upcoming
releases will include the ability to run in eventual-consistency mode, with per-
transaction conflict detection and resolution across a WAN, for active-active replica-
tion between datacenters. In short, MySQL Cluster is an impressive piece of technology.
There are also at least two providers of add-on products to simplify cluster deployment
and management: Oracle support contracts for MySQL Cluster include its MySQL
Cluster Manager product, and Severalnines offers a Cluster Control product (
http://
www.severalnines.com
)
. This product is also capable of helping deploy and manage
replication clusters.
Percona XtraDB Cluster
Percona XtraDB Cluster is a relatively new technology that adds synchronous replica-
tion and clustering capabilities to the XtraDB (InnoDB) storage engine itself, rather
than through a new storage engine or an external server. It is built on Galera,
9
a library
that replicates writes across nodes in a cluster. Like MySQL Cluster, Percona XtraDB
Cluster offers synchronous multi-master replication,
10
with true write-anywhere ca-
pabilities. Also like MySQL Cluster, it can provide high availability as well as guarantee
zero data loss (durability, the D in ACID) when a node fails, and of course nodes can
fail without causing the whole cluster to fail.
The underlying technology, Galera, uses a technique called
write-set replication
. Write
sets are actually encoded as row-based binary log events for the purpose of transmitting
them between nodes and updating the other nodes in the cluster, though the binary
log is not required to be enabled.
Percona XtraDB Cluster is very fast. Cross-node replication can actually be faster than
not clustering, because writing to remote RAM is faster than writing to the local disk
in full durability mode. You have the option of relaxing durability on each node for
performance, if you wish, and relying on the presence of multiple nodes with copies of
9. The Galera technology is developed by Codership Oy (
http://www.codership.com
) and is available as a
patch for standard MySQL and InnoDB. Percona XtraDB Cluster includes a modified version of that
patchset, as well as other features and functionality. Percona XtraDB Cluster is a Galera-based solution
that's ready to use out of the box.
10. You can also use it in a master-replica configuration by writing to just one node, but there's no difference
in the cluster configuration for this mode of operation.
