The petition is processed by one of the two managers supported, PMES or SGE-oneflow. They both end up triggering the tool executable in the particular virtual machine where the selected tool is encapsulated. Below are explained the differences between both enactors, but anyhow, both include mechanisms to provide elasticity to the system, so that the number of active tool VMs is proportional to the number of job petitions, as long as cloud computational resources allow it.
Eventually, the petition reaches the virtual machine. If the tool corresponds to a workflow orchestrated by PyCOMPSs, the process manager will invoke the application using PyCOMPSs libraries. If not, the tool will be called directly. In any case. tools are generally wrapped by the mg-tool API, that acts as intermediary agent between the MuGVRE and the tool code itself to uniformelly communicate with it. Thought the API, the application or pipelines is eventually executed.