[jira] [Commented] (KYLIN-1050) Protect coprocessor from exhausting mem or cpu of region server

["vipul jhawar (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/KYLIN-1050?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14935133#comment-14935133 ] 

vipul jhawar commented on KYLIN-1050:
-------------------------------------

Hi li

A few comments on this as input to consider in scope.
There can be different kinds of queries being triggered by users for instance, it could be a long query which may require a lot of scanning and filtering a.k.a a scheduled query kind of thing which actually may not require a timeout and someone may be open to running the co processor as long as he wants.
There will be other sets of queries which will require the user or a system level setting in the UI, which specified that if my query is not completing in N seconds, pls timeout so that i am not waiting forever and hogging the resources. So, if we give an option to pass the timeout param from the query in the REST api to the coprocessor and then the co processor timeout accordingly, it would provide a lot of flexibility as i know that for certain queries i am ok with higher timeout so there may not be a lower bound timeout but a higher bound can be set as default if required.

> Protect coprocessor from exhausting mem or cpu of region server
> ---------------------------------------------------------------
>
>                 Key: KYLIN-1050
>                 URL: https://issues.apache.org/jira/browse/KYLIN-1050
>             Project: Kylin
>          Issue Type: Bug
>            Reporter: liyang
>
> As reported by vipul jhawar -- http://mail-archives.apache.org/mod_mbox/kylin-dev/201509.mbox/ajax/%3CCAHNrOjG0q%3Dbb2BuNfn5TSg3qqh-m%3DsUuVnpEUASqvGkFpc%2BBkQ%40mail.gmail.com%3E
> HBase coprocessor may run long, consuming lots of mem and/or cpu on region server for whatever reason.  When this happens, the client side RPC timeout, but the server side is leftover to run, continue occupying the resource.
> Need a mechanism to stop the server side in such scenario. Either by a timeout, or by mem consumption, or both.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)