Processing Streaming Data - Real-Time Analytics - page 151

builder.setBolt("processing", new BasicBolt())
.fieldsGrouping("input", new Fields("key1","key2"));
All and Global Groupings
The allGrouping and globalGrouping methods are exact opposites of
each other. The allGrouping method ensures that all tuples in an event
stream are transmitted to all running tasks for a particular bolt. This is
occasionally useful, but it multiplies the number of events by the number of
running tasks, so use it with care.
The globlaGrouping method does the opposite. It ensures that all tuples
from a stream go to the bolt with the lowest numbered identifier. All bolt
tasks receive an identifier number, but this ensures only one of them will
be used. This should also be used with care because it effectively disables
Storm's parallelism.
Direct Groupings
The directGrouping method is a special form of grouping that allows
the output Bolt or Spout to decide which Bolt task receives a particular
Tuple . This requires that the stream be declared to be direct when the
bolt is created. The producer bolt must also use the emitDirect method
instead of the emit method.
The emitDirect method takes an extra parameter that identifies the
destination Bolt . The list of valid identifiers can be obtained from the
TopologyContext when implementing a Bolt . This is shown in greater
detail in the section that details implementing Bolts .
Custom Groupings
When all else fails, it is also possible to implement a custom grouping
method. To do this, create a class that implements the
CustomStreamGrouping interface.
This interface contains two methods. The first method, prepare , is called
when the topology is instantiated and lets the grouping method assemble
any metadata it may need to perform its tasks. In particular, this method
receives the targetTasks variable, which is the list of Bolts that subscribe
to the stream being grouped.