[
https://issues.apache.org/jira/browse/YARN-1232?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13777932#comment-13777932
]
Karthik Kambatla commented on YARN-1232:
----------------------------------------
The patch adopts an approach very similar to HDFS. When HA is enabled, the
following confs are expected:
# {{yarn.resourcemanager.ha.enabled}} set to true
# {{yarn.resourcemanager.address}} set to LOGICAL_NAME
# {{yarn.resourcemanager.ha.nodes.LOGICAL_NAME}} should list the RM ensemble.
(e.g. "rm1,rm2")
Individual RM-specific configs that provide the RPC addresses are:
# One can set {{yarn.resourcemanager.ha.nodes.id}} on individual hosts to
specify which node in the ensemble, this node corresponds to. (e.g. "rm1")
# {{yarn.resourcemanager.<rpc-suffix>.LOGICAL_NAME.<node-id>}} where
{{rpc-address-conf}} comes from {address, scheduler.address, admin.address,
resource-tracker.address, webapp.address} and {{node-id}} comes from the RM
ensemble ({rm1, rm2})
Implementation details: All the individual RPC services continue to use the
configs they use today. When HA is enabled, RMHAProtocolService sets these
configs from the newly added configs.
Verified the patch manually (pseudo-dist mode) with both today's configs and
newly added configs.
> Configuration support for RM HA
> -------------------------------
>
> Key: YARN-1232
> URL: https://issues.apache.org/jira/browse/YARN-1232
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: resourcemanager
> Reporter: Karthik Kambatla
> Assignee: Karthik Kambatla
> Labels: ha
> Attachments: yarn-1232-1.patch, yarn-1232-2.patch, yarn-1232-3.patch
>
>
> We should augment the configuration to allow users specify two RMs and the
> individual RPC addresses for them. This blocks
> ConfiguredFailoverProxyProvider.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
