@Documented @Retention(value=RUNTIME) @Target(value=METHOD) public @interface GridifySetToValue
GridifySetToValueannotation is the way to grid-enable existing code with specific semantics.
This annotation can be applied to any public method that needs to be grid-enabled, static or non-static. When this annotation is applied to a method, the method execution is considered to be grid-enabled. In general, the execution of this method can be transferred to another node in the grid, potentially splitting it into multiple subtasks to be executed in parallel on multiple grid nodes. But from the caller perspective this method still looks and behaves like a local apply. This is achieved by utilizing AOP-based interception of the annotated code and injecting all the necessary grid logic into the method execution.
GridifySetToValue used as extended version of
It provides automated gridification of two popular types of mathematical
functions that allow for "embarrassingly" simple parallelization (for example,
find prime numbers in collection or find maximum in collection). The goal of this
addition is to provide very simple and easy to use real ForkJoin gridification
for some often used use cases. Note that in a standard
the user has to write
ComputeTask to provide ForkJoin behavior.
In a proposed design - the ForkJoin will be achieved automatically.
Let F be the function or a method in Java language and St be a set of values
t or a typed Collection<T> in Java. Let also
the result value. These are the types of functions that will be supported (defined in
terms of their properties):
F (S 1, S 2, ..., S k) => R, where F (S 1, S 2, ..., S k) == F (F (S 1), F (S 2), ..., F (S k)) which basically defines a function over a set that produces the same result when applied recursively over its results for a subsets of the original set.
In definition above we used Collection<T> for the purpose of example. The following Java collections will have to be supported as return values or method parameters:
Note that when using
@GridifySetToValue annotation the state of the whole instance will be
serialized and sent out to remote node. Therefore the class must implement
Serializable interface. If you cannot make the class
then you must implement custom grid task which will take care of proper state
initialization. In either case, Ignite must be able to serialize the state passed to remote node.
Example for these types of functions would be any aggregate function returning single value from the multiple inputs. For example:
... @GridifySetToValue(threshold = 2) public static Number findMaxValue(Collection<Number> input); ...Method
findMaxValuefindMaxValue simply traverses the input collection and finds the maximum element in. If we split this collection into two sub-collections, for example, and find maximums in each - we can then apply the same method to the collection of two just found maximums to find the overall maximum elements of the original collection. The formal definition of (or requirement for) such method:
-javaagent:[path to jboss-aop-jdk50-4.x.x.jar]
-Djboss.aop.class.path=[path to ignite.jar]
GridifySpringEnhancer.enhance(java.lang.Object)for the object that has method with
Note that this method of weaving is rather inconvenient and AspectJ or JBoss AOP is recommended over it. Spring AOP can be used in situation when code augmentation is undesired and cannot be used. It also allows for very fine grained control of what gets weaved.
|Modifier and Type||Optional Element and Description|
Name of the Ignite instance to use.
Optional interceptor class.
Optional node filter to filter nodes participated in job split.
Optional parameter that defines a split size.
Optional parameter that defines the minimal value below which the execution will NOT be grid-enabled.
Optional gridify task execution timeout.
public abstract long timeout
0which indicates that task will not timeout.
public abstract int threshold
public abstract int splitSize
This value should be greater than zero and can be less, equal or greater than
value. In most cases, however, the optimal value for the split size is the
For example, if input collection size is 100, number of nodes 10 and split size is set to 5 - Ignite
will submit 2 sub-collections of 5 elements each to each node (provided in order by load
By default (when split size is zero) - Ignite will automatically determine the split size based on number of nodes and collection size - if collection size is available (not an iterator). If collection size cannot be determined - the split size will default to threshold. If threshold is not set - a runtime exception will be thrown.