Tim Armstrong created IMPALA-7323:
-------------------------------------

             Summary: Allow closing scan nodes before all returned batches are 
destroyed
                 Key: IMPALA-7323
                 URL: https://issues.apache.org/jira/browse/IMPALA-7323
             Project: IMPALA
          Issue Type: Improvement
          Components: Backend
            Reporter: Tim Armstrong
            Assignee: Tim Armstrong


If the scan node finishes early because of a limit at the scan or higher up in 
the plan tree, returned batches may still be in use by nodes higher up in the 
operator tree that reference memory still owned by the Scanner object, or 
enqueued in the RowBatchQueue. We should add a method that can be called before 
Close() to attach all of those resources to the trailing batch. E.g. 
AcquireResourcesFinal(). 

This method needs to be separate from Close() because Close() releases all 
resources accounted against the node being closed (e.g. reservations but 
ownership of the returned resources is only done lazily when required by a 
blocking node).



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

Reply via email to