[jira] [Created] (UIMA-3129) DUCC RM: Reenable purge bit

["Jim Challenger (JIRA)" <de...@uima.apache.org>]

Jim Challenger created UIMA-3129:
------------------------------------

             Summary: DUCC RM: Reenable purge bit
                 Key: UIMA-3129
                 URL: https://issues.apache.org/jira/browse/UIMA-3129
             Project: UIMA
          Issue Type: Bug
          Components: DUCC
            Reporter: Jim Challenger
            Assignee: Jim Challenger
             Fix For: 1.0-Ducc


We had disabled RM's 'purge' bit because of a race with the Agent.  Changes to the agent eliminate the race.

The 'purge' bit is set when a machine misses some number of heartbeats. RM considers the node "down" and evict the shares, similar to preemption.  Without the 'purge' bit the OR will wait for confirmation from agents that the processes are stopped, but if the agent is not responding, the confirmation may never come and the processes will be "stuck".

The 'purge' bit is used to tell OR that the shares are being evicted because of suspected agent death, so OR can clean up right away and not look for confirmation.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira