[
https://issues.apache.org/jira/browse/FLINK-5931?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15897351#comment-15897351
]
Stephan Ewen commented on FLINK-5931:
-------------------------------------
Currently, Flink-on-YARN stores critical files in HDFS, like typically
checkpoints, and JobGraphs.
We would need another storage location for the cases where the file system is
unavailable.
> Make Flink highly available even if defaultFS is unavailable
> ------------------------------------------------------------
>
> Key: FLINK-5931
> URL: https://issues.apache.org/jira/browse/FLINK-5931
> Project: Flink
> Issue Type: Improvement
> Reporter: Haohui Mai
> Assignee: Haohui Mai
>
> In order to use Flink in mission-critical environments, Flink must be
> available even if the {{defaultFS}} is unavailable.
> We have deployed HDFS in HA mode in our production environment. In our
> experience we have experienced performance degradations and downtime when the
> HDFS cluster is being expanded or under maintenances. Under this case it is
> desirable to deploy jobs through alternative filesystem (e.g., S3).
> This jira is to track the improvements to Flink to enable Flink to continue
> to operate even {{defaultFS}} is unavailable.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
