Hi Stefan,
Flink uses only one broadcast variable for all parallel tasks on one
machine.
Flink can also load the broadcast variable into a custom data structure.
Have a look at the getBroadcastVariableWithInitializer() method:
/**
* Returns the result bound to the broadcast variable identified by the
* given {@code name}. The broadcast variable is returned as a shared data
structure
* that is initialized with the given {@link BroadcastVariableInitializer}.
* <p>
* IMPORTANT: The broadcast variable data structure is shared between the
parallel
* tasks on one machine. Any access that modifies its internal
state needs to
* be manually synchronized by the caller.
*
* @param name The name under which the broadcast variable is registered;
* @param initializer The initializer that creates the shared data
structure of the broadcast
* variable from the sequence of elements.
* @return The broadcast variable, materialized as a list of elements.
*/
<T, C> C getBroadcastVariableWithInitializer(String name,
BroadcastVariableInitializer<T, C> initializer);
Right now, there is no easy way to run multiple tasks one after the other
that I am aware of.
However, we are working on materializing intermediate results. Once this
feature is available, it should be easy to do the grep steps one by one.
Cheers, Fabian