[jira] [Created] (MESOS-9012) Master cannot reclaim task resources when scheduler is unavailable

["Benno Evers (JIRA)" <ji...@apache.org>]

Benno Evers created MESOS-9012:
----------------------------------

             Summary: Master cannot reclaim task resources when scheduler is unavailable
                 Key: MESOS-9012
                 URL: https://issues.apache.org/jira/browse/MESOS-9012
             Project: Mesos
          Issue Type: Bug
            Reporter: Benno Evers


Currently, task status updates from the agent include the `latest_state` field, which contains the latest task state known to the agent, regardless of which status updates the scheduler acknowledged.

This allows the master to reclaim the tasks' resources as soon as it finishes, without waiting for a framework acknowledgement, which can save significant time if the framework is currently disconnected or otherwise not available.

However, this feature does not currently work correctly, because we currently only send new status updates from the agent to master after the previous one has been acknowledged, so the master has to wait until that or until task reconciliation happens until it gets the information that the task has died.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)