[
https://issues.apache.org/jira/browse/YARN-7622?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Greg Phillips updated YARN-7622:
--------------------------------
Attachment: YARN-7622.004.patch
[~wilfreds] - Thanks again for the review. The two fields which triggered the
findbugs synchronization issues have been reverted to volatile (Atomic types
removed), and an additional allocation file check has been added to ensure only
hdfs or local file schemes are provided. A unit test has been added to confirm
that providing an invalid URI scheme will fail prior to {{serviceInit}}.
In the case where an absolute path for the allocation file is provided and the
file doesn't exist the original behavior would throw an exception inside the
{{serviceInit}} thread at the call to lastModified(). This exception is handled
within the thread without breaking the loop. {{getAllocationFile}} only checks
for file existence if it is a resource accessible to the classloader, otherwise
a potentially bogus URL would still throw an exception within the
{{serviceInit}} thread.
> Allow fair-scheduler configuration on HDFS
> ------------------------------------------
>
> Key: YARN-7622
> URL: https://issues.apache.org/jira/browse/YARN-7622
> Project: Hadoop YARN
> Issue Type: Improvement
> Components: fairscheduler, resourcemanager
> Reporter: Greg Phillips
> Assignee: Greg Phillips
> Priority: Minor
> Attachments: YARN-7622.001.patch, YARN-7622.002.patch,
> YARN-7622.003.patch, YARN-7622.004.patch
>
>
> The FairScheduler requires the allocation file to be hosted on the local
> filesystem on the RM node(s). Allowing HDFS to store the allocation file will
> provide improved redundancy, more options for scheduler updates, and RM
> failover consistency in HA.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
