Hi, I have a small Mesos cluster of machines. Users launch frameworks onto the cluster that will typically execute tens of tasks that take several hours to run. When all of these tasks are done the framework ends. When offered resources, these frameworks will only accept those that are appropriate. By appropriate I mean that we start up slaves with an attribute "slavetype" set to "worker," "workstation" or "service" and that a resource is only appropriate for the described frameworks if they come from a slave with "slavetype" set to "worker."
A problem that I saw was this: a user launched a framework. It created tasks until all the cluster's appropriate resources were used. At that point I added another slave to the cluster with a "slavetype" of "worker." The slave registered correctly with the master and its resources began being offered to frameworks. In the logs I could see a repeating pattern of offers being sent to existing frameworks, I could see the master processing ACCEPT calls for offers and I could see the resources associated with the new slave being recovered because none of the frameworks they were offered to wanted them. What I never saw was these new resources being offered to the framework that could have used them. Ideally, I would have liked these new resources to have been offered to that framework. (One note, another instance of the same framework was launched after seeing this problem and it was offered these new resources.) Does DRF explain the behavior that I am seeing? If so, can I get the desired behavior by using roles and weights for the framework? If not, is the solution to write my own resource allocator for the master? Is there an allocator out there that might suit my needs? Thanks, Mike
