[
https://issues.apache.org/jira/browse/HIVE-12220?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sergey Shelukhin updated HIVE-12220:
------------------------------------
Description:
In the llap-daemon site you need to set, among other things,
llap.daemon.memory.per.instance.mb
and
hive.llap.io.cache.orc.size
The use of hive.llap.io.cache.orc.size caused me some unnecessary problems,
initially I entered the value in MB rather than in bytes. Operator error you
could say but I look at this as a fraction of the other value which is in mb.
Second, is this really tied to ORC? E.g. when we have the vectorized text
reader will this data be cached as well? Or might it be in the future?
I would like to propose instead using hive.llap.io.cache.size.mb for this
setting.
was:
In the llap-daemon site you need to set, among other things,
llap.daemon.memory.per.instance.mb
and
hive.llap.io.cache.orc.size
The use of hive.llap.io.cache.orc.size caused me some unnecessary problems,
initially I entered the value in MB rather than in bytes. Operator error you
could say but I look at this as a fraction of the other value which is in mb.
Second, is this really tied to ORC? E.g. when we have the vectorized text
reader will this data be cached as well? Or might it be in the future?
I would like to propose instead using hive.llap.io.cache.size.mb for this
setting.
NO PRECOMMIT TESTS
> LLAP: Usability issues with hive.llap.io.cache.orc.size
> -------------------------------------------------------
>
> Key: HIVE-12220
> URL: https://issues.apache.org/jira/browse/HIVE-12220
> Project: Hive
> Issue Type: Bug
> Components: llap
> Reporter: Carter Shanklin
> Assignee: Sergey Shelukhin
> Attachments: HIVE-12220.01.patch, HIVE-12220.02.patch,
> HIVE-12220.patch, HIVE-12220.tmp.patch
>
>
> In the llap-daemon site you need to set, among other things,
> llap.daemon.memory.per.instance.mb
> and
> hive.llap.io.cache.orc.size
> The use of hive.llap.io.cache.orc.size caused me some unnecessary problems,
> initially I entered the value in MB rather than in bytes. Operator error you
> could say but I look at this as a fraction of the other value which is in mb.
> Second, is this really tied to ORC? E.g. when we have the vectorized text
> reader will this data be cached as well? Or might it be in the future?
> I would like to propose instead using hive.llap.io.cache.size.mb for this
> setting.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
