1. Test computation pattern on sensible code part
2. Offer a fail safe and unlimited work queue (only limited by disk space)
The impl is straightforward, each Work Queue is a stream (default, elasticsearch, ....), the size of the queue is the concurrency of the pool.
Create a generic computation that receive a record representing a work and that execute the work in the processRecord. Checkpoint on each work.
The topology is a set of sink computations.
The workmanager append records with the serialized work into the stream.