A continuous job is a job that produces usable output before it ends. A continuous flow graph, unlike a regular Ab Initio graph, is intended to run for an indefinite period of time, continually taking in new input and producing new, usable output while the graph keeps running. A continuous flow graph might or might not go on forever.
The advantages of continuous flow graphs include better performance and latency. There is no overhead for starting the job each time a new batch of data arrives. Results of the job are available sooner than for a non-continuous graph.
A continuous flow graph includes
• One or more subscribers. A subscriber is the only allowed data source.
• A publisher at the end of every data flow.
• Any continuous or continuously enabled component can be in the middle, between a subscriber and a publisher.
Restrictions for continuous flow :
• All components in the graph must be continuous components or they must be continuously enabled.
• All components with no output flows must be publishers. There must be at least one publisher in the graph for the graph to determine when a checkpoint can be committed.
• All subscribers must issue checkpoints and compute points in the same sequence.
• The graph must execute in a single phase. More than one phase is not allowed.
• All data in a continuous flow graph must come from a subscriber component. The source of data cannot be an Input File or Input Table component.
