Python and MySQL/MariaDB. The queue is an in-memory table. Fair queuing is implemented as a big SQL SELECT statement.

The time scale is slow enough that polling it once a second for new work is sufficient. If it had to go fast, it would require something like a standard fair queuing implementation, where you hash the IP address and put the request onto 1 of N queues, serviced round-robin.

Doesn't matter all that much. For ingestion, you might have one VM running one Java process and one DB process as a microservice (I like Dropwizard and PostgreSQL for this). The worker itself would be another Java process with a 100 thread max thread pool that either periodically checks for new work, or uses the new NOTIFY semantics new to Postgres 9.6. In this case the "secret sauce" is that the worker has some extra smarts to reorder the queue to avoid resource hogs, which is the cool part.