[
https://issues.apache.org/jira/browse/YARN-3673?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Subru Krishnan updated YARN-3673:
---------------------------------
Attachment: YARN-3673-YARN-2915-v2.patch
Updated patch (v2) that addresses [~jianhe]'s feedback and fix for Yetus
warnings.
bq. If FederationRMFailoverProxyProvider is used for standalone RM, we only
have a single RM, how does it do failover ?
It does _not_ do failover for standalone RM. It still obtains the address for
the standalone RM from FederationStateStore and connects to it using a
appropriate RetryPolicies (like retryForeverWithFixedSleep).
bq. My point is that it looks to me the cache in the singleton is not much
used. Please correct me if I’m wrong. Because it always removes the object from
the cache before doing failover.
You are right, cache is not used for failover but pooling is independent of
caching with Singleton access pattern the only commonality. Cache is across all
store implementations (managed by facade) while pooling is dependent on store
(SQL for example supports it). We are fine to bypass cache for failover as it's
a rare event.
> Create a FailoverProxy for Federation services
> ----------------------------------------------
>
> Key: YARN-3673
> URL: https://issues.apache.org/jira/browse/YARN-3673
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: nodemanager, resourcemanager
> Reporter: Subru Krishnan
> Assignee: Subru Krishnan
> Attachments: YARN-3673-YARN-2915-v1.patch,
> YARN-3673-YARN-2915-v2.patch
>
>
> This JIRA proposes creating a failover proxy for Federation based on the
> cluster membership information in the StateStore that can be used by both
> Router & AMRMProxy
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
