Aggregates Continuous Handlers

• The StoredMeasureHandler implementation (core product) receives transaction events coming from the TransactionStream. This is done for all measures that are only impacted by facts that contribute to the locations they are computed on.
• Other handlers are usually custom implementations to support either of these particular types:
  • a real-time event that is not related to facts contribution (e.g. market data feed that impacts post-processed measures). These types of handlers do not rely on the TransactionStream but on a custom stream.
  • a measure that is impacted by facts that do not contribute to the location it is computed on (for example a customized location shifting post-processor). These types of handlers do rely on the TransactionStream but implement their own logic for computing the impact.

​

Handlers (and Streams) Registration

final IActivePivotInstanceDescription cube =
    StartBuilding.cube("myCube")
        .withPostProcessor("myCustomMeasure")
        .withPluginKey("MY_CUSTOM_PP_KEY")
        .withContinuousQueryHandlers("STORED", "MY_CUSTOM_HANDLER_KEY")
        .withDimensions(dimensionAdder) // whatever you need to describe your cube ...
        .build();

Do not forget the STORED handler when defining custom handlers!
When explicitly configuring handlers for a given measure on a stream other than the TransactionStream, remember to add the default handler (plugin key = STORED) that is responsible for the measure sensitivity to real-time events for facts contribution. This handler is not added by default on top of the custom handlers (it is only the default when there is no explicit configuration).

@Override
public String getStreamKey() {
  return TransactionStream.PLUGIN_TYPE;
}

A measure is not allowed to define several handlers impacted by the same stream. If a measure is impacted in several ways (for example must be recomputed if the primitive measure for its location changes but also if the primitive measure for a shifted location to the previous day changes), then a custom handler implementation has to take into account the combined logic.

​

A Concrete Example

• contributors.COUNT
• Delta.SUM
• PnL.SUM

• Whenever an ActivePivot transaction is committed, the STORED handler computes impacted cells and pushes updated values to the parent continuous query’s listeners.
• Whenever Market Data are updated, the MARKETDATA handler computes impacted cells and pushes updated values to the parent continuous query’s listeners.

​

Continuous Handlers and Chains of Post-Processors

​

Implementing a Custom Handler

There is only one handler instance per handler (plugin) key within a given cube instance, except for those that inherit from IMeasureDependentAggregatesContinuousHandler.
This means that, if multiple post-processors need to use the same handler logic on different data (different hierarchies, data source name), be careful of how the handler is implemented.
Typically, this means ensuring that it does not hold any state or cache anything beyond the execution scope / lifespan of each handler method.
Then, when it really makes sense to re-use some common logic, you should rely on class inheritance: have a parent abstract class implement the common logic and keep specific handler implementations (with separate plugin keys!) for different types of data.

1. The original location (i.e. the one passed as a parameter of computeImpact())). This corresponds to the location that is asked for in the query.
2. The collection of impacted locations. These are the sub-locations of the original location for which we need to recompute the measure because the value might have changed.
3. The collection of removable points - point locations (not range locations with null coordinates) which might be “removed” as a result of the event concerned. Note that range locations with collection coordinates are allowed.

Retrieve a collection of non-range locations from a collection of range locations The easiest way to construct a collection of point locations from a collection of range locations is to use the LocationUtil.expandAll() function:
LocationUtil.expandAll(pivot.getHierarchies(),myCollectionOfLocations).

Don’t forget to register your new handler as a valid extended plugin value

@QuartetExtendedPluginValue(
    intf = IMeasureDependentAggregatesContinuousHandler.class,
    key = TimeLineHandler.PLUGIN_KEY)

​

Measure Dependent Handler