Spark 3.4 Support about abris

[luisvicenteatprima]

I have tested it on Databricks with Spark 3.4.1 and it works.

[hafizmujadidKhalid]

I tested it with spark 3.5 as well and it works fine.

[kevinwallimann]

Hi @jelmew
Thanks for reporting the issue. Could you please double-check if you are really using v3.5.0, or maybe the latest master version? I'm looking at https://github.com/apache/spark/blob/master/connector/avro/src/main/scala/org/apache/spark/sql/avro/AvroDeserializer.scala#L59 and indeed I can see that a new constructor argument useStableIdForUnionType: Boolean was added. However, this is on the master branch of Spark. I don't see that argument in v3.5.0 https://github.com/apache/spark/blob/v3.5.0/connector/avro/src/main/scala/org/apache/spark/sql/avro/AvroDeserializer.scala#L53-L56, and @hafizmujadidKhalid reported no issues with Spark 3.5.0.

@cerveada See https://issues.apache.org/jira/browse/SPARK-43380. We need to take care of this for the next release of Spark

[cerveada]

Unless Spark changed some of the APIs we use between versions, Abris will work fine. Thanks for reporting your tests, it's valuable information for us and for other Abris users.

[jelmew]

There is some incompatibility on spark 3.5. It seems some constructors have changed, giving Abris problems.

Caused by: java.lang.NoSuchMethodException: org.apache.spark.sql.avro.AvroDeserializer.<init>(org.apache.avro.Schema, org.apache.spark.sql.types.DataType, java.lang.String)
	at java.lang.Class.getConstructor0(Class.java:3082)
	at java.lang.Class.getConstructor(Class.java:1825)
	at org.apache.spark.sql.avro.AbrisAvroDeserializer$$anonfun$1.applyOrElse(AbrisAvroDeserializer.scala:38)
	at org.apache.spark.sql.avro.AbrisAvroDeserializer$$anonfun$1.applyOrElse(AbrisAvroDeserializer.scala:37)
	at scala.runtime.AbstractPartialFunction.apply(AbstractPartialFunction.scala:38)
	at scala.util.Failure.recover(Try.scala:234)
	at org.apache.spark.sql.avro.AbrisAvroDeserializer.<init>(AbrisAvroDeserializer.scala:37)
	at za.co.absa.abris.avro.sql.AvroDataToCatalyst.deserializer$lzycompute(AvroDataToCatalyst.scala:71)
	at za.co.absa.abris.avro.sql.AvroDataToCatalyst.deserializer(AvroDataToCatalyst.scala:71)
	at za.co.absa.abris.avro.sql.AvroDataToCatalyst.nullSafeEval(AvroDataToCatalyst.scala:87)
	at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIteratorForCodegenStage1.processNext(Unknown Source)
	at org.apache.spark.sql.execution.BufferedRowIterator.hasNext(BufferedRowIterator.java:43)
	at org.apache.spark.sql.execution.WholeStageCodegenEvaluatorFactory$WholeStageCodegenPartitionEvaluator$$anon$1.hasNext(WholeStageCodegenEvaluatorFactory.scala:43)
	at scala.collection.Iterator$$anon$10.hasNext(Iterator.scala:460)
	at scala.collection.convert.Wrappers$IteratorWrapper.hasNext(Wrappers.scala:32)
	at com.google.common.collect.Iterators$PeekingImpl.hasNext(Iterators.java:1139)
	at com.databricks.photon.NativeRowBatchIterator.hasNext(NativeRowBatchIterator.java:44)
	at 0xa8e0d62 <photon>.HasNext(external/workspace_spark_3_5/photon/jni-wrappers/jni-row-batch-iterator.cc:50)
	at 0x5e136fb <photon>.OpenImpl(external/workspace_spark_3_5/photon/exec-nodes/file-writer-node.cc:166)
	at com.databricks.photon.JniApiImpl.open(Native Method)
	at com.databricks.photon.JniApi.open(JniApi.scala)
	at com.databricks.photon.JniExecNode.open(JniExecNode.java:71)
	at com.databricks.photon.PhotonWriteResultHandler.$anonfun$getResult$3(PhotonWriteStageExec.scala:121)
	at scala.runtime.java8.JFunction0$mcV$sp.apply(JFunction0$mcV$sp.java:23)
	at com.databricks.photon.PhotonResultHandler.timeit(PhotonResultHandler.scala:30)
	at com.databricks.photon.PhotonResultHandler.timeit$(PhotonResultHandler.scala:28)
	at com.databricks.photon.PhotonWriteResultHandler.timeit(PhotonWriteStageExec.scala:67)
	at com.databricks.photon.PhotonWriteResultHandler.$anonfun$getResult$2(PhotonWriteStageExec.scala:121)
	at org.apache.spark.util.Utils$.tryWithSafeFinallyAndFailureCallbacks(Utils.scala:1542)
	at com.databricks.photon.PhotonWriteResultHandler.getResult(PhotonWriteStageExec.scala:118)
	... 38 more

[jelmew]

Hi @kevinwallimann

This is using databricks runtime 14.2 https://docs.databricks.com/en/release-notes/runtime/14.2.html. They might have backported something from master?

Kind regards,
Jelmer

Edit,

Okay, yup. They included it.

[sauletawil]

+1

[kevinwallimann]

Hi @jelmew
Unfortunately, I don't have access to Spark on Databricks Runtime, but I will reproduce and fix the error by building Spark locally using the latest commit on branch-3.5. Since I've identified a few issues on running the tests for Abris with Spark 3.5.0, I need to fix those first. See #350

[lucafurrer]

+1

[kevinwallimann]

Hi @jelmew
I've released v6.4.0 with the bugfix. https://github.com/AbsaOSS/ABRiS/releases/tag/v6.4.0. Please let me know if it fixes your issue.

[jelmew]

We are testing.

[roicostas]

Same problem here. It seems that AvroDeserializer contructor changed a lot recently, adding more params to the constructor. This way:
3.5.0 has 3 parameters
3.5.1 has 4 parameters
current master version has 5 parameters

We are also using Databricks last LTS version 14.3 which in theory uses 3.5.0 but it should have some customizations and it seems AvroDeserializer is already being provided by one of their jars and it has the new version with 5 parameters.
The jar which provides AvroDeserializer is: file:/databricks/jars/----ws_3_5--connector--avro--avro-hive-2.3__hadoop-3.2_2.12_shaded---606136534--avro-unshaded-hive-2.3__hadoop-3.2_2.12_deploy.jar!/org/apache/spark/sql/avro/AvroDeserializer.class

We printed out available contructors and we checked that the available constructor is the one with 5 parameters.

With databricks 13.3 it works fine

[jelmew]

I agree on that. Databricks 13.3 works fine, but 14.3 lts had the above problem once again