[ 
https://issues.apache.org/jira/browse/HIVE-16859?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16054712#comment-16054712
 ] 

Thejas M Nair commented on HIVE-16859:
--------------------------------------

Should we support a key=value format, so that this is extendable for adding 
additional params in future ?
Does it make sense to encode the 'settings' at a metadata level instead of per 
file level ?


> CM uri encoding
> ---------------
>
>                 Key: HIVE-16859
>                 URL: https://issues.apache.org/jira/browse/HIVE-16859
>             Project: Hive
>          Issue Type: Bug
>          Components: HiveServer2
>    Affects Versions: 3.0.0
>            Reporter: anishek
>            Assignee: anishek
>
> Currently for hive replication, the cm root uri is configured via 
> "hive.repl.cmrootdir". This configuration needs to have the same value on 
> both the primary and replica hive warehouse. 
> CM uri should be encoded such that the cm root of the source should be part 
> of the URI itself. so the cmfs uri's should be following
> {code}
> cmfs:hdfs://[authority]/[actual_location]#[checksum_of_file]_[encoded_cm_root_on_primary]
> {code}
> so that we can detect what is the root location of the source cm root at any 
> target replica warehouse. Since the filesystem configurations can be 
> different for the  primary and replica warehouse there might be additional 
> configurations will be required to create {{FileSystem}} objects to talk to 
> respective filesystems. if we want to support that we can add an additional 
> configuration stating the primary cm root location on the replica warehouse 
> along with other fs related configurations and in that case this bug might be 
> irrelevant.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Reply via email to