[SPARK-47842] Spark job relying over Hudi are blocked after one or zero commit

• Type: Bug
• Status: Open
• Priority: Blocker
• Resolution: Unresolved
• Affects Version/s: 3.3.0
• Component/s: PySpark, Structured Streaming
• Labels: pull-request-available

Environment

Hudi version : 0.12.1-amzn-0
Spark version : 3.3.0
Hive version : 3.1.3
Hadoop version : 3.3.3 amz
Storage (HDFS/S3/GCS..) : S3
Running on Docker? (yes/no) : no (EMR 6.9.0)
Additional context

• Flags: Important

Language

English italian

• Assignee: Unassigned
• Reporter: alessandro pontis
• Created: 13/Apr/24 14:00
• Updated: 26/Apr/24 00:22

Description

Hello, we are facing the fact that some pyspark job that rely on Hudi seems to be blocked, in fact if we go over the spark console we can see the situation in the attachment
we can see that we have 71 completed jobs but those are CDC process that should read from Kafka topic continuously. We verified yet that there are messages queued over the kafka topic. If you kill the application and then restart in some cases the job will act normally and other times the job still remain stacked.

Our deploy condition are the following:
We read INSERT, UPDATE and DELETE operation from a Kafka topic and we replicate them in a target hudi table stored on Hive via a pyspark job running 24/7

 

PYSPARK WRITE
df_source.writeStream.foreachBatch(foreach_batch_write_function)
 {{ FOR EACH BATCH FUNCTION:
#management of delete messages
batchDF_deletes.write.format('hudi') \
.option('hoodie.datasource.write.operation', 'delete') \
.options(**hudiOptions_table) \
.mode('append') \
.save(S3_OUTPUT_PATH)

#management of update and insert messages
batchDF_upserts.write.format('org.apache.hudi') \
.option('hoodie.datasource.write.operation', 'upsert') \
.options(**hudiOptions_table) \
.mode('append') \
.save(S3_OUTPUT_PATH)}}
 
SPARK SUBMIT
spark-submit --master yarn --deploy-mode cluster --num-executors 1 --executor-memory 1G --executor-cores 2 --conf spark.dynamicAllocation.enabled=false --packages org.apache.spark:spark-sql-kafka-0-10_2.12:3.1.2 --conf spark.serializer=org.apache.spark.serializer.KryoSerializer --conf spark.sql.hive.convertMetastoreParquet=false --jars /usr/lib/hudi/hudi-spark-bundle.jar <path_to_script>

Issue Links

• links to: GitHub Pull Request #19