Let's say I'm not intending to use multiple instances of each role, but rather increase the VM size when load increases.
Idea: Use internal endpoints between each role, for sending a notification that role A has put a job for role B in the queue&table (using both, for the roles to ascertain that a job is done at least once and at most once)?
Problem it solves: no role would have to be constantly polling any queue, but instead only polling it upon a notification through the channel between the internal endpoints, and only poll until the queue is empty.
Is this a reasonable solution, or would it bring possible problems?
Or is this just unnecessary overhead given the low transaction costs ($ 0.005 per 100.000 reads?). It would take about 1000.000 transactions per day to reach the cost of an XS worker role instance.