nsivabalan opened a new pull request, #17591:
URL: https://github.com/apache/hudi/pull/17591
### Describe the issue this Pull Request addresses
Fixing unions of rdd for partitioned RLI bootstrap.
For partitioned RLI bootstrap, we create N rdd's for N table partitions and
union them all before we trigger bulk insert in MDT RLI partition. At TB scale,
we are running into StackOverFlow error.
17760 [main] INFO org.apache.spark.SparkContext [] - Starting job: count at
DeepLineageRepro.java:53
Exception in thread "dag-scheduler-event-loop" java.lang.StackOverflowError
at scala.Option.flatMap(Option.scala:271)
at org.apache.spark.rdd.RDD.checkpointRDD(RDD.scala:247)
at org.apache.spark.rdd.RDD.dependencies(RDD.scala:254)
at org.apache.spark.rdd.RDD.isBarrier_$lzycompute(RDD.scala:2055)
at org.apache.spark.rdd.RDD.isBarrier_(RDD.scala:2054)
at org.apache.spark.rdd.RDD.isBarrier(RDD.scala:2050)
at org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2(RDD.scala:2055)
at
org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2$adapted(RDD.scala:2055)
at
scala.collection.IndexedSeqOptimized.prefixLengthImpl(IndexedSeqOptimized.scala:41)
at
scala.collection.IndexedSeqOptimized.exists(IndexedSeqOptimized.scala:49)
at
scala.collection.IndexedSeqOptimized.exists$(IndexedSeqOptimized.scala:49)
at scala.collection.mutable.ArrayBuffer.exists(ArrayBuffer.scala:49)
at org.apache.spark.rdd.RDD.isBarrier_$lzycompute(RDD.scala:2055)
at org.apache.spark.rdd.RDD.isBarrier_(RDD.scala:2054)
at org.apache.spark.rdd.RDD.isBarrier(RDD.scala:2050)
at org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2(RDD.scala:2055)
at
org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2$adapted(RDD.scala:2055)
at
scala.collection.IndexedSeqOptimized.prefixLengthImpl(IndexedSeqOptimized.scala:41)
at
scala.collection.IndexedSeqOptimized.exists(IndexedSeqOptimized.scala:49)
at
scala.collection.IndexedSeqOptimized.exists$(IndexedSeqOptimized.scala:49)
at scala.collection.mutable.ArrayBuffer.exists(ArrayBuffer.scala:49)
at org.apache.spark.rdd.RDD.isBarrier_$lzycompute(RDD.scala:2055)
at org.apache.spark.rdd.RDD.isBarrier_(RDD.scala:2054)
at org.apache.spark.rdd.RDD.isBarrier(RDD.scala:2050)
at org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2(RDD.scala:2055)
at
org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2$adapted(RDD.scala:2055)
at
scala.collection.IndexedSeqOptimized.prefixLengthImpl(IndexedSeqOptimized.scala:41)
at
scala.collection.IndexedSeqOptimized.exists(IndexedSeqOptimized.scala:49)
at
scala.collection.IndexedSeqOptimized.exists$(IndexedSeqOptimized.scala:49)
at scala.collection.mutable.ArrayBuffer.exists(ArrayBuffer.scala:49)
at org.apache.spark.rdd.RDD.isBarrier_$lzycompute(RDD.scala:2055)
at org.apache.spark.rdd.RDD.isBarrier_(RDD.scala:2054)
at org.apache.spark.rdd.RDD.isBarrier(RDD.scala:2050)
at org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2(RDD.scala:2055)
at
org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2$adapted(RDD.scala:2055)
at
scala.collection.IndexedSeqOptimized.prefixLengthImpl(IndexedSeqOptimized.scala:41)
at
scala.collection.IndexedSeqOptimized.exists(IndexedSeqOptimized.scala:49)
at
scala.collection.IndexedSeqOptimized.exists$(IndexedSeqOptimized.scala:49)
at scala.collection.mutable.ArrayBuffer.exists(ArrayBuffer.scala:49)
at org.apache.spark.rdd.RDD.isBarrier_$lzycompute(RDD.scala:2055)
at org.apache.spark.rdd.RDD.isBarrier_(RDD.scala:2054)
at org.apache.spark.rdd.RDD.isBarrier(RDD.scala:2050)
at org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2(RDD.scala:2055)
at
org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2$adapted(RDD.scala:2055)
at
scala.collection.IndexedSeqOptimized.prefixLengthImpl(IndexedSeqOptimized.scala:41)
at
scala.collection.IndexedSeqOptimized.exists(IndexedSeqOptimized.scala:49)
at
scala.collection.IndexedSeqOptimized.exists$(IndexedSeqOptimized.scala:49)
at scala.collection.mutable.ArrayBuffer.exists(ArrayBuffer.scala:49)
at org.apache.spark.rdd.RDD.isBarrier_$lzycompute(RDD.scala:2055)
at org.apache.spark.rdd.RDD.isBarrier_(RDD.scala:2054)
at org.apache.spark.rdd.RDD.isBarrier(RDD.scala:2050)
at org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2(RDD.scala:2055)
at
org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2$adapted(RDD.scala:2055)
at
scala.collection.IndexedSeqOptimized.prefixLengthImpl(IndexedSeqOptimized.scala:41)
at
scala.collection.IndexedSeqOptimized.exists(IndexedSeqOptimized.scala:49)
at
scala.collection.IndexedSeqOptimized.exists$(IndexedSeqOptimized.scala:49)
at scala.collection.mutable.ArrayBuffer.exists(ArrayBuffer.scala:49)
at org.apache.spark.rdd.RDD.isBarrier_$lzycompute(RDD.scala:2055)
at org.apache.spark.rdd.RDD.isBarrier_(RDD.scala:2054)
at org.apache.spark.rdd.RDD.isBarrier(RDD.scala:2050)
at org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2(RDD.scala:2055)
at
org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2$adapted(RDD.scala:2055)
at
scala.collection.IndexedSeqOptimized.prefixLengthImpl(IndexedSeqOptimized.scala:41)
at
scala.collection.IndexedSeqOptimized.exists(IndexedSeqOptimized.scala:49)
at
scala.collection.IndexedSeqOptimized.exists$(IndexedSeqOptimized.scala:49)
at scala.collection.mutable.ArrayBuffer.exists(ArrayBuffer.scala:49)
at org.apache.spark.rdd.RDD.isBarrier_$lzycompute(RDD.scala:2055)
at org.apache.spark.rdd.RDD.isBarrier_(RDD.scala:2054)
at org.apache.spark.rdd.RDD.isBarrier(RDD.scala:2050)
at org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2(RDD.scala:2055)
at
org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2$adapted(RDD.scala:2055)
at
scala.collection.IndexedSeqOptimized.prefixLengthImpl(IndexedSeqOptimized.scala:41)
at
scala.collection.IndexedSeqOptimized.exists(IndexedSeqOptimized.scala:49)
at
scala.collection.IndexedSeqOptimized.exists$(IndexedSeqOptimized.scala:49)
at scala.collection.mutable.ArrayBuffer.exists(ArrayBuffer.scala:49)
at org.apache.spark.rdd.RDD.isBarrier_$lzycompute(RDD.scala:2055)
at org.apache.spark.rdd.RDD.isBarrier_(RDD.scala:2054)
at org.apache.spark.rdd.RDD.isBarrier(RDD.scala:2050)
at org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2(RDD.scala:2055)
at
org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2$adapted(RDD.scala:2055)
at
scala.collection.IndexedSeqOptimized.prefixLengthImpl(IndexedSeqOptimized.scala:41)
at
scala.collection.IndexedSeqOptimized.exists(IndexedSeqOptimized.scala:49)
at
scala.collection.IndexedSeqOptimized.exists$(IndexedSeqOptimized.scala:49)
at scala.collection.mutable.ArrayBuffer.exists(ArrayBuffer.scala:49)
at org.apache.spark.rdd.RDD.isBarrier_$lzycompute(RDD.scala:2055)
at org.apache.spark.rdd.RDD.isBarrier_(RDD.scala:2054)
at org.apache.spark.rdd.RDD.isBarrier(RDD.scala:2050)
at org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2(RDD.scala:2055)
at
org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2$adapted(RDD.scala:2055)
at
scala.collection.IndexedSeqOptimized.prefixLengthImpl(IndexedSeqOptimized.scala:41)
at
scala.collection.IndexedSeqOptimized.exists(IndexedSeqOptimized.scala:49)
at
scala.collection.IndexedSeqOptimized.exists$(IndexedSeqOptimized.scala:49)
at scala.collection.mutable.ArrayBuffer.exists(ArrayBuffer.scala:49)
at org.apache.spark.rdd.RDD.isBarrier_$lzycompute(RDD.scala:2055)
at org.apache.spark.rdd.RDD.isBarrier_(RDD.scala:2054)
at org.apache.spark.rdd.RDD.isBarrier(RDD.scala:2050)
at org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2(RDD.scala:2055)
at
org.apache.spark.rdd.RDD.$anonfun$isBarrier_$2$adapted(RDD.scala:2055)
at
scala.collection.IndexedSeqOptimized.prefixLengthImpl(IndexedSeqOptimized.scala:41)
at
scala.collection.IndexedSeqOptimized.exists(IndexedSeqOptimized.scala:49)
at
scala.collection.IndexedSeqOptimized.exists$(IndexedSeqOptimized.scala:49)
at scala.collection.mutable.ArrayBuffer.exists(ArrayBuffer.scala:49)
at org.apache.spark.rdd.RDD.isBarrier_$lzycompute(RDD.scala:2055)
at org.apache.spark.rdd.RDD.isBarrier_(RDD.scala:2054)
Response from chatgpt:
this stacktrace is actually very characteristic of a Spark task
deserialization / closure serialization problem, and the pattern
rdd1.union(rdd2).union(rdd3)... is a classic way to trigger it at scale.
What’s really failing?
Your executor is dying with:
java.lang.StackOverflowError
deep inside
java.io.ObjectInputStream.readObject0/defaultReadFields/readArray/...
That means the executor is deserializing something enormous / deeply nested
(the “task binary”: the serialized task + closure + referenced objects + parts
of the RDD lineage metadata), and the Java deserializer hits recursive depth
and blows the thread stack.
This is why you don’t see any “Hudi code” in the trace: the failure happens
before your task even starts running your logic — while Spark is reconstructing
the task on the executor.
Why does chaining 500 unions cause this?
RDD.union doesn’t merge data immediately — it creates a new RDD node with
dependencies on its parents.
When you do:
rdd1.union(rdd2).union(rdd3)...union(rdd500)
you’re building a deeply nested lineage like:
Union( Union( Union( rdd1, rdd2), rdd3), rdd4) ... )
Depth ≈ 500.
At small scale, Spark may get away with it (fewer partitions/tasks, smaller
serialized task payloads). At TB scale:
more partitions → more tasks → more task shipping/deserialization events
scheduler/closure serialization overhead grows
the nested object graph representing the RDD dependencies becomes “deep
enough” that Java serialization recursion overflows
Fixes (best → fallback)
✅ Fix 1: Use SparkContext.union over a collection (flat union)
This avoids the “nested union of unions” structure.
val finalRdd = spark.sparkContext.union(rddsSeq) // Seq[RDD[Record]]
or in Java:
JavaRDD finalRdd = sc.union(rddList);
This tends to create a single UnionRDD with many parents instead of a deep
recursive nesting.
and hence the fix in this patch.
### Summary and Changelog
Fixing rdd unioning during Partitioned RLI bootstrap for large tables.
### Impact
Smooth RLI initialization for partitioned RLI.
### Risk Level
low
### Documentation Update
<!-- Describe any necessary documentation update if there is any new
feature, config, or user-facing change. If not, put "none".
- The config description must be updated if new configs are added or the
default value of the configs are changed.
- Any new feature or user-facing change requires updating the Hudi website.
Please follow the
[instruction](https://hudi.apache.org/contribute/developer-setup#website)
to make changes to the website. -->
### Contributor's checklist
- [ ] Read through [contributor's
guide](https://hudi.apache.org/contribute/how-to-contribute)
- [ ] Enough context is provided in the sections above
- [ ] Adequate tests were added if applicable
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
