Benjamin Bannier created MESOS-8730:
---------------------------------------
Summary: Provide explicit feedback when a resource becomes
unavailable
Key: MESOS-8730
URL: https://issues.apache.org/jira/browse/MESOS-8730
Project: Mesos
Issue Type: Bug
Components: agent, master, storage
Reporter: Benjamin Bannier
With local resource providers we allowed for dynamic changes to agent
resources. This opens up a number of new scenarios
# disappeared resources have been offered to a framework and are used in
tasks, or
# disappeared resources have been reserved, but are currently neither offered
nor used.
In the first case, we cannot in general assume that tasks using disappeared
resource will terminate unexpectedly, and we should handle this explicitly,
e.g., by providing explicit feedback to the framework so it can migrate tasks
of these resources, or by implementing explicit kill policies for such tasks.
In the second case, it is not immediately clear to whom such changes should be
reported since multiple frameworks can share a role. The information should
already be exposed to any watchers of e.g., {{GET_RESOURCE_PROVIDERS}} calls,
but we might want to think about a unified feedback channel handling both
scenarios.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
