Alexey Romanenko created BEAM-11016:
---------------------------------------
Summary: SqsIO exception when moving to AWS2 SDK
Key: BEAM-11016
URL: https://issues.apache.org/jira/browse/BEAM-11016
Project: Beam
Issue Type: Bug
Components: io-java-aws
Affects Versions: 2.24.0
Reporter: Alexey Romanenko
[email protected] reported:
I've been attempting to migrate to the AWS2 SDK (version 2.24.0) on Apache
Spark 2.4.7. However, when switching over to the new API and running it I keep
getting the following exceptions:
{code}
2020-09-30 20:38:32,428 ERROR streaming.StreamingContext: Error starting the
context, marking it as stopped org.apache.spark.SparkException: Job aborted due
to stage failure: Task 0 in stage 2.0 failed 4 times, most recent failure: Lost
task 0.3 in stage 2.0 (TID 6, 10.42.180.18, executor 0):
java.lang.NullPointerException at
org.apache.beam.sdk.io.aws2.sqs.SqsUnboundedSource.lambda$new$e39e7721$1(SqsUnboundedSource.java:41)
at
org.apache.beam.vendor.guava.v26_0_jre.com.google.common.base.Suppliers$MemoizingSupplier.get(Suppliers.java:129)
at
org.apache.beam.sdk.io.aws2.sqs.SqsUnboundedSource.getSqs(SqsUnboundedSource.java:74)
at
org.apache.beam.sdk.io.aws2.sqs.SqsUnboundedReader.pull(SqsUnboundedReader.java:165)
at
org.apache.beam.sdk.io.aws2.sqs.SqsUnboundedReader.advance(SqsUnboundedReader.java:108)
at
org.apache.beam.runners.spark.io.MicrobatchSource$Reader.advanceWithBackoff(MicrobatchSource.java:246)
at
org.apache.beam.runners.spark.io.MicrobatchSource$Reader.start(MicrobatchSource.java:224)
at
org.apache.beam.runners.spark.stateful.StateSpecFunctions$1.apply(StateSpecFunctions.java:168)
at
org.apache.beam.runners.spark.stateful.StateSpecFunctions$1.apply(StateSpecFunctions.java:107)
at org.apache.spark.streaming.StateSpec$$anonfun$1.apply(StateSpec.scala:181)
at org.apache.spark.streaming.StateSpec$$anonfun$1.apply(StateSpec.scala:180)
at
org.apache.spark.streaming.rdd.MapWithStateRDDRecord$$anonfun$updateRecordWithData$1.apply(MapWithStateRDD.scala:57)
{code}
Examining the source of SqsUnboundedSource reveals a lambda where it's trying
to chain a few references:
{code}
read.sqsClientProvider().getSqsClient()
{code}
Which is odd as I explicitly set the client provider on the read transform.
This was working well enough with the old SqsIO API to connect and process
messages off the queue.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)