[
https://issues.apache.org/jira/browse/HIVE-18765?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sahil Takiar updated HIVE-18765:
--------------------------------
Description:
Multiple issues:
*Issue 1:*
{{SparkClientImpl#handle(ChannelHandlerContext, Error)}} swallows the cause of
the error message:
{code:java}
LOG.warn("Error reported from remote driver.", msg.cause);
{code}
There should be a '{}' in the message. Without it the {{msg.cause}} info gets
swallowed.
*Issue 2:*
When we send an error during shutdown of the {{RemoteDriver}}, we don't wait
for the error message to be sent. We just send it and then close the RPC
channel. For a graceful shutdown, that doesn't seem ideal. We should at least
wait a bit for the RPC message to be sent before shutting things down. I ran
some tests locally and its pretty easy to hit a situation where the error
message doesn't even get sent to the {{SparkClientImpl}}
was:
{{SparkClientImpl#handle(ChannelHandlerContext, Error)}} swallows the cause of
the error message:
{code}
LOG.warn("Error reported from remote driver.", msg.cause);
{code}
There should be a '{}' in the message. Without it the {{msg.cause}} info gets
swallowed.
> Error message are improperly handled by the RemoteDriver and SparkClientImpl
> ----------------------------------------------------------------------------
>
> Key: HIVE-18765
> URL: https://issues.apache.org/jira/browse/HIVE-18765
> Project: Hive
> Issue Type: Sub-task
> Components: Spark
> Reporter: Sahil Takiar
> Assignee: Sahil Takiar
> Priority: Major
>
> Multiple issues:
> *Issue 1:*
> {{SparkClientImpl#handle(ChannelHandlerContext, Error)}} swallows the cause
> of the error message:
> {code:java}
> LOG.warn("Error reported from remote driver.", msg.cause);
> {code}
> There should be a '{}' in the message. Without it the {{msg.cause}} info gets
> swallowed.
> *Issue 2:*
> When we send an error during shutdown of the {{RemoteDriver}}, we don't wait
> for the error message to be sent. We just send it and then close the RPC
> channel. For a graceful shutdown, that doesn't seem ideal. We should at least
> wait a bit for the RPC message to be sent before shutting things down. I ran
> some tests locally and its pretty easy to hit a situation where the error
> message doesn't even get sent to the {{SparkClientImpl}}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
