[
https://issues.apache.org/jira/browse/HIVE-19708?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
mahesh kumar behera updated HIVE-19708:
---------------------------------------
Description:
* During repl load
** for filesystem based copying of file if the copy fails due to a connection
error to source Name Node, we should recreate the filesystem object.
** the retry logic for local file copy should be triggered using the original
source file path ( and not the CM root path ) since failure can be due to
network issues between DFSClient and NN.
* When listing files in tables / partition to include them in _files, we
should add retry logic when failure occurs. FileSystem object here also should
be recreated since the existing one might be in inconsistent state.
was:
* add a parameter at db level to identify if its a source of replication.
beacon will set this.
* Enable CM root only for databases that are a source of a replication policy,
for other db's skip the CM root functionality.
* prevent database drop if the parameter indicating its source of a
replication, is set.
* as an upgrade to this version, beacon should set the property on all
existing database policies, in affect.
* the parameter should be of the form . – repl.source.for : List < policy ids
>
> Repl copy retrying with cm path even if the failure is due to network issue
> ---------------------------------------------------------------------------
>
> Key: HIVE-19708
> URL: https://issues.apache.org/jira/browse/HIVE-19708
> Project: Hive
> Issue Type: Task
> Components: Hive, HiveServer2, repl
> Affects Versions: 3.1.0
> Reporter: mahesh kumar behera
> Assignee: mahesh kumar behera
> Priority: Major
> Fix For: 3.1.0
>
>
> * During repl load
> ** for filesystem based copying of file if the copy fails due to a
> connection error to source Name Node, we should recreate the filesystem
> object.
> ** the retry logic for local file copy should be triggered using the
> original source file path ( and not the CM root path ) since failure can be
> due to network issues between DFSClient and NN.
> * When listing files in tables / partition to include them in _files, we
> should add retry logic when failure occurs. FileSystem object here also
> should be recreated since the existing one might be in inconsistent state.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
