anishek updated HIVE-19188:
    Affects Version/s: 3.0.0

> Replication Resiliency: for listing files for a partition /  table
> ------------------------------------------------------------------
>                 Key: HIVE-19188
>                 URL: https://issues.apache.org/jira/browse/HIVE-19188
>             Project: Hive
>          Issue Type: Bug
>          Components: HiveServer2
>    Affects Versions: 3.0.0
>            Reporter: anishek
>            Priority: Major
> When for listing files for a partition /  table fails during dump we should 
> have ability to *retry with wait,* so the whole dump process does not fail. 
> additionally  the filesystem object used to list the files might have to be 
> recreated since only IOException is thrown by the object for various 
> operations and we dont know if the internal DFSClient it holds has a broken 
> connection due to network error.  This has to be done using 
> Filesystem.newInstance() since caching of fs object might be enabled.
> currently caching of the filesystem object is depended upon
> * Scheme
> * Authority
> * UGI
> * And a long called unique defaults to 0 unless explicitly provided, which is 
> never the case from Hive since we call default constructor which uses 0L by 
> default internally.

This message was sent by Atlassian JIRA

Reply via email to