[
https://issues.apache.org/jira/browse/DRILL-2698?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Chris Westin updated DRILL-2698:
--------------------------------
Issue Type: New Feature (was: Bug)
> Add allocation policy for (direct memory) limits per-query-per-node
> -------------------------------------------------------------------
>
> Key: DRILL-2698
> URL: https://issues.apache.org/jira/browse/DRILL-2698
> Project: Apache Drill
> Issue Type: New Feature
> Components: Execution - Flow
> Affects Versions: 0.8.0
> Reporter: Chris Westin
> Assignee: Chris Westin
> Fix For: Future
>
>
> We want the allocator to support a policy that limits memory allocation
> per-query-per-node. Current policy is to divide the available direct memory
> by the number of fragments running on a node, and limit each fragment to
> that. This new policy would count fragments that belong to the same query and
> are running on the same node together. So, if query A has 3 fragments running
> on a node, and query B has 5 fragments running on the same node, and there
> are no other queries with fragments on that node, then there are two queries
> running on the node, and their allocation limit is the available direct
> memory divided by two. (Under the original policy, which divides the
> available memory by the number of fragments, A would be allowed 3/8 of the
> memory, and B would be allowed 5/8 of the memory.)
> Note that this policy should be a user-settable boot-time option; the
> original policy would still also be available as an option.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
