Dataflow runner deletes the binary specified in flag --worker_binary

Currently a binary compiled for the dataflow runner compiles the source code again in order to produce a worker binary.  In a production environment I will not have the source code around so I need to provide the worker binary myself.

The good news: there is a flag for that.

The bad news: the dataflow runner deletes the binary after launching the pipeline.

Expected behaviour: don't delete the binary. I am the owner of the worker binary. It is not a temporary artifact. I would like to reuse the same binary to launch multiple dataflow pipelines

Better yet, maybe add a flag "–self_worker_binary", so the currently running binary is used as the worker binary. This way I don't have to copy my binary twice into the docker image (once as the entrypoint, and once as data for the command line flag).