Harsh Panchal created SPARK-36798:
-------------------------------------
Summary: When SparkContext is stopped, metrics system should be
flushed after listeners have finished processing
Key: SPARK-36798
URL: https://issues.apache.org/jira/browse/SPARK-36798
Project: Spark
Issue Type: Bug
Components: Structured Streaming
Affects Versions: 2.3.2
Reporter: Harsh Panchal
In current implementation, when {{SparkContext.stop()}} is called,
{{metricsSystem.report()}} is called before {{listenerBus.stop()}}. In this
case, if some listener is producing some metrics, they would never reach sink.
Background:
We have some ingestion jobs in Spark Structured Streaming. To monitor them,
collect some metrics like number of input rows, trigger time etc. from
{{QueryProgressEvent}} received via {{StreamingQueryListener}}. These metrics
are then pushed to DB by custom sinks registered in {{MetricsSystem}}. We
noticed that these metrics are lost occasionally for last batch.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
