Description
Sometimes the direct dispatcher is impractical to use, as it results in prohibitively many requests being made to the server. But in these cases, users may still want to be able to query metrics that they have already told a Run to upload.
For this case, we need a .flush() function which will empty any queues in the dispatcher (sending their contents to the server). Crucially, control should not flow out of the .flush() function until all those uploads have finished, so that any subsequent queries of that data are guaranteed to be correct.
The use case that has made this requirement apparent is using connectors (which upload lots of things to a run - so many that the direct dispatcher causes problems) with the optimization framework (where we often want to query the contents of a still-open run just after metrics have been sent to it).