Database Reference
In-Depth Information
F
Data movement: Allowing data to be used by additional applications or workloads
on additional hardware. Examples are Reference Data Management, where a
single central server might provide information to many other applications, and also
Business Intelligence/Reporting Systems.
Of course, both of those topics are complex areas, and there are many architectures and
possibilities for doing each of those.
What we will talk about here is data movement, where there is
no
transformation
of the
data—we simply copy the data from one PostgreSQL database server to another. So, we are
specifically avoiding all discussion on ETL tools, EAI tools, inter-database migration, data
warehousing strategies, and so on. Those are valid topics in IT architecture, we just don't
cover them here.
Let's look at the basic architecture. Typically, the individual database servers are referred to
as nodes. The whole group of database servers involved in replication is known as a Cluster.
That is the common usage of the term, though be careful that the term Cluster is also used for
two other quite separate meanings elsewhere in PostgreSQL. First, cluster is sometimes used
to refer to the whole database instance, though I prefer the term "database server". Second,
there is a command named
CLUSTER
, which is designed to sort data into a specific order
within a table.
The first database server is also known as the
Master
,
Primary
,
Provider
,
Sender
, or the
Source
server
.
The second database server is also known as the
Standby
,
Slave
or
Subscriber
, or
Receiver
.
There can be multiple Standbys, if desired.
The replication systems described here are all Single Master, Multiple Standby systems. Multi-
Master architectures are discussed briefly later.
The designations Master and Standby are just roles that any node can take at some point.
To move the Master role to another node, we perform a procedure named Switchover. If the
Master dies, and does not recover, then the more severe role change is known as a Failover.
In many ways, they may be similar, but it helps to use different words for the two situations.
Software that manages the cluster, and in some cases automatically initiates the failover
process, is sometimes referred to as clusterware. Clusterware may also perform other
functions, such as load balancing.
The key aspect of replication is that data changes are captured on the Master, and then
transferred to the Standby nodes. In some cases, a Standby node may send data changes
onto later Standbys, a process known as relay.
