ExecutionEnvironment.execute() should not be called when no sinks are defined.

[fschueler]

Currently we call env.execute() after the execution of all program blocks if the ExecutionContext is of type FlinkExecutionContext. In hybrid_flink mode this leads to the problem that execute() can be called even though no sinks are defined.

We should somehow check if an adequate plan exists before calling execute().

[fschueler]

This can not only happen in hybrid_flink mode but whenever collect() is called somewhere in between. Anyone has an idea how we can keep track of the defined sources/sinks?

[aalexandrov]

You can use createProgramPlan and inspect the result for sinks.

[aalexandrov]

Alternatively, use the ExecutionEnvironment wrapper class using the delegate pattern and override the void registerDataSink(DataSink<?> sink) { method in order to keep track of this kind of meta-information.

[fschueler]

Unfortunately, createProgramPlan already fails if no sinks are defined and the variable sinks of the ExecutionEnvironment is private. I guess we will have to go with the alternative.

[fschueler]

Mabe the "final" execute (plus check) could happen in the finally block of the DMLScript in line 685, where the SparkContext is stopped.