Siddharth Seth updated HIVE-10424:
    Attachment: HIVE-10424.1.txt

Patch to factor in running queue + wait queue capacity per node.
Also moves all scheduler on to a single thread - requests go on to a queue and 
are taken off whenever a node becomes available, or has capacity.

Can run with the old 'unlimited' capacity by setting 
llap.task.scheduler.num.schedulable.tasks.per.node to -1

> LLAP: Factor known capacity into scheduling decisions
> -----------------------------------------------------
>                 Key: HIVE-10424
>                 URL: https://issues.apache.org/jira/browse/HIVE-10424
>             Project: Hive
>          Issue Type: Sub-task
>            Reporter: Siddharth Seth
>            Assignee: Siddharth Seth
>             Fix For: llap
>         Attachments: HIVE-10424.1.txt

This message was sent by Atlassian JIRA

Reply via email to