[SPARK-21303] Web-UI shows some Jobs get stuck randomly and stays like that. Neither able to kill

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Not A Problem
• Affects Version/s: 2.1.0, 2.1.1
• Component/s: DStreams

Environment

Kubernetes 1.4.12 on AWS
OS Ubuntu
Spark 2.1.1
Cassandra 3.9

• Assignee: Unassigned
• Reporter: Arun Achuthan
• Created: 04/Jul/17 11:59
• Updated: 12/Jul/17 15:37
• Resolved: 10/Jul/17 06:50

Description

We are running a streaming application which was running without any issues for long. Last few days we are seeing some jobs randomly getting stuck on the web ui. This doesn't stop the application as the following jobs are successful. The stuck jobs remain in the web-ui as stuck with no progress. These are the observations we made. At the time the first job is shown stuck on UI the driver logs mention this

2017-07-04 05:33:20,189 ERROR [dag-scheduler-event-loop] org.apache.spark.scheduler.LiveListenerBus: Dropping SparkListenerEvent because no remaining room in event queue. This likely means one of the SparkListeners is too slow and cannot keep up with the rate at which tasks are being started by the scheduler.

For every other random stuck job the driver logs mention the below at the same time

2017-07-04 05:33:20,194 WARN [dispatcher-event-loop-0] org.apache.spark.scheduler.LiveListenerBus: Dropped 1 SparkListenerEvents since Thu Jan 01 00:00:00 UTC 1970

2017-07-04 05:49:31,571 WARN [dag-scheduler-event-loop] org.apache.spark.scheduler.LiveListenerBus: Dropped 1 SparkListenerEvents since Tue Jul 04 05:34:20 UTC 2017

After the jobs starts getting stuck we are experiencing performance drops as well as scheduling delays within the application. We couldn't find any other significant errors in the driver logs.

Issue Links

• is related to: SPARK-18838 High latency of event processing for large jobs