[
https://issues.apache.org/jira/browse/RANGER-1201?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Qiang Zhang updated RANGER-1201:
--------------------------------
Attachment: yarn-ha-zookeeper.jpg
0001-RANGER-1201-support-yarn-HA-V2.patch
When yarn in HA mode, there is only active RM in zookeeper,so we can not get
all the RMs via zookeeper,
and the value of active RM in zookeeper is like "gagcluster-yarn rm1", which
not specified the IP:Port to access RM.
It seems RM Client APIs also do not provide the way to get all RMs or get
IP:Port of the RM which name is "gagcluster-yarn rm1".
So I change the code to have a config to hold the list of RMs and then find
which one is active
and use it for the Lookup / Test Connection in Ranger.
By the way, this solution is still compatible with the scene that yarn is
single node without zookeeper.
> Ranger admin support yarn HA, support standby yarn rest url configuration
> when creating yarn service.
> ------------------------------------------------------------------------------------------------------
>
> Key: RANGER-1201
> URL: https://issues.apache.org/jira/browse/RANGER-1201
> Project: Ranger
> Issue Type: New Feature
> Components: admin
> Reporter: Qiang Zhang
> Assignee: Qiang Zhang
> Priority: Minor
> Labels: features
> Attachments: 0001-RANGER-1201-support-yarn-HA-V2.patch,
> 0001-RANGER-1201-support-yarn-HA.patch, rangerYarnStandby.png,
> yarn-ha-zookeeper.jpg
>
> Original Estimate: 72h
> Time Spent: 72h
> Remaining Estimate: 0h
>
> In ranger admin, when creating yarn service, there is yarn.rest.url for yarn
> active resourcemanger, if yarn is in HA mode, add yarn.rest.url.standby for
> yarn standby resourcemanger, so if yarn active RM is out of service, ranger
> admin can try yarn standby RM to
> support autocomplete for resource names when creating policies of yarn
> service.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
