In the case where it uses VM images to deploy:

Sometimes, e.g. due to RabbitMQ server restarts, some extractor services are stopped by the OS since it respawns too fast. The symptom: the VMs are up, but there are no RabbitMQ connections from them, so they use cloud compute resources, but do not process jobs in the queue.

Need to handle this case. Could check for such VMs, then restart the service.