Github user dragos commented on the pull request:
https://github.com/apache/spark/pull/10993#issuecomment-179266568
I see the same behavior with master. I think this is a regression
introduced when Akka was removed, and communication has switched to Netty.
Here's what happens: the connection between the driver and each shuffle
server is idle, and controlled by the general `spark.network.timeout`,
defaulting to 120s. That's exactly what can be seen in the logs: the
application disconnects exactly 2 minutes after registration.
We'd need a TCP connection without a timeout, or have heartbeats exchanged
between the two. I'll file a Jira ticket.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
