[
https://issues.apache.org/jira/browse/HIVE-16459?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Rui Li updated HIVE-16459:
--------------------------
Attachment: HIVE-16459.2.patch
Just noticed we're already cancelling RPCs in
{{RpcDispatcher::channelInactive}}. However, we have a "monitor" inbound
handler that doesn't forward channelInactive event. Therefore
{{RpcDispatcher::channelInactive}} is not triggered. Patch v2 fixes this
monitor.
> Cancel outstanding RPCs when channel closes
> -------------------------------------------
>
> Key: HIVE-16459
> URL: https://issues.apache.org/jira/browse/HIVE-16459
> Project: Hive
> Issue Type: Bug
> Reporter: Rui Li
> Assignee: Rui Li
> Attachments: HIVE-16459.1.patch, HIVE-16459.2.patch
>
>
> In SparkTask, we try to get job infos after the query finishes. Assume the
> job finishes due to remote side crashes and thus closes the RPC. There's a
> race condition: if we try to get job info before we notice the RPC is closed,
> the SparkTask waits for {{hive.spark.client.future.timeout}} (default 60s)
> before it returns, even though we already know the job has failed.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
