Hi Stefano, thanks for a swift and very detailed reply! Situation for our setup is great, just a few last details to figure out. On Wed, Oct 29, 2014 at 8:43 AM, Stefano Maggiolo <s.maggiolo@xxxxxxxxx> wrote: > Thank you Motiejus, also for moving the conversation :) > > On Tue Oct 28 2014 at 10:34:47 PM Motiejus Jakštys <desired.mta@xxxxxxxxx> > wrote: >> > The Worker is in contact with two other components: ES (that passes the job > metadata and gets back the metadata of the completed job), and the DB > (specifically, only the file storage part). Thanks, I wasn't aware that worker connects to DB. > From a size perspective, the communication with ES is very light. It is also > arguable that a contestant having full access to that communication wouldn't > gain anything practical. Worker communication with DB is equally important as with ES, since in our case DB would stay on-site (on slow hardware and potentially bad uplink), whereas Workers would be in the datacenter. >> * Do the test artifacts stay on the Worker for the evaluation, or are >> they sent back? > > What do you mean by test artifacts? If I understand the question, the > evaluation is going to be done by probably a different Worker. As the file > storage has a cache on the Worker, though, if the evaluation is sent to the > same Worker, it is going to retrieve the executable from the cache instead > than from the DB. Sorry, old-style contests taking over my mind. Please disregard this question, situation is good. > Good question :) > Testcases are transferred only when the worker starts, and the cache is > persistent, so if the worker restart in general there is no need to re-fill > the cache. It is going to happen only if you change testcases. > In my experience usually that is by far the heaviest operation. During the > contest, the heaviest objects are going to be (statically linked) > executables, as a rule of thumb about 2MB per executable. That is all great, thank you. > Hope this helps. If you are going to use this solution, we'd like to have a > report :) If we'll use this setup, I promise you the report. -- Motiejus Jakštys