[
https://issues.apache.org/jira/browse/HIVE-17827?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sergey Shelukhin updated HIVE-17827:
------------------------------------
Description:
There are about 5 different duplicate and intersecting structures used in
TezTask to manage the file list (additionalLr, inputOutputX xN (array and list
and/or map?), resourceMap; plus a list in the actual session object); multiple
methods named addExtraResourcesBlah, localizeResourcesBlah; at least 2 places
where TezClient is invoked to add resources to AM, on the same path.
All this mess needs to be changed to have exactly one place that figures out
what needs to be localized, and exactly one place (in TezSessionState so we
don't duplicate state and TezClient usage inside and outside of it) where we
look at what's already there and localize things that are not.
Meanwhile signatures can also be cleaned up, e.g. since branch-1 jobConf is
passed to updateSession and ignored, etc.
was:
There are about 5 different duplicate and intersecting structures used in
TezTask to manage the file list (additionalLr, inputOutputX xN (array and list
and/or map?), resourceMap; plus a list in the actual session object); multiple
methods named addExtraResourcesBlah, localizeResourcesBlah; at least 2 places
where TezClient is invoked to add resources to AM, on the same path.
All this mess needs to be changed to have exactly one place that figures out
what needs to be localized, and exactly one place (in TezSessionState so we
don't duplicate state and TezClient usage inside and outside of it) where we
look at what's already there and localize things that are not.
> refactor TezTask file management
> --------------------------------
>
> Key: HIVE-17827
> URL: https://issues.apache.org/jira/browse/HIVE-17827
> Project: Hive
> Issue Type: Bug
> Reporter: Sergey Shelukhin
>
> There are about 5 different duplicate and intersecting structures used in
> TezTask to manage the file list (additionalLr, inputOutputX xN (array and
> list and/or map?), resourceMap; plus a list in the actual session object);
> multiple methods named addExtraResourcesBlah, localizeResourcesBlah; at least
> 2 places where TezClient is invoked to add resources to AM, on the same path.
> All this mess needs to be changed to have exactly one place that figures out
> what needs to be localized, and exactly one place (in TezSessionState so we
> don't duplicate state and TezClient usage inside and outside of it) where we
> look at what's already there and localize things that are not.
> Meanwhile signatures can also be cleaned up, e.g. since branch-1 jobConf is
> passed to updateSession and ignored, etc.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
