[SPARK-18546] UnsafeShuffleWriter corrupts encrypted shuffle files when merging

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 2.1.0
• Fix Version/s: 2.1.0
• Component/s: Spark Core
• Target Version/s: 2.1.0
• Assignee: Marcelo Masiero Vanzin
• Reporter: Marcelo Masiero Vanzin
• Created: 22/Nov/16 21:19
• Updated: 08/Dec/16 07:42
• Resolved: 30/Nov/16 22:11

Description

The merging algorithm in UnsafeShuffleWriter does not consider encryption, and when it tries to merge encrypted files the result data cannot be read, since data encrypted with different initial vectors is interleaved in the same partition data. This leads to exceptions when trying to read the files during shuffle:

com.esotericsoftware.kryo.KryoException: com.ning.compress.lzf.LZFException: Corrupt input data, block did not start with 2 byte signature ('ZV') followed by type byte, 2-byte length)
	at com.esotericsoftware.kryo.io.Input.fill(Input.java:142)
	at com.esotericsoftware.kryo.io.Input.require(Input.java:155)
	at com.esotericsoftware.kryo.io.Input.readInt(Input.java:337)
	at com.esotericsoftware.kryo.util.DefaultClassResolver.readClass(DefaultClassResolver.java:109)
	at com.esotericsoftware.kryo.Kryo.readClass(Kryo.java:610)
	at com.esotericsoftware.kryo.Kryo.readClassAndObject(Kryo.java:721)
	at org.apache.spark.serializer.KryoDeserializationStream.readObject(KryoSerializer.scala:228)
	at org.apache.spark.serializer.DeserializationStream.readKey(Serializer.scala:169)
	at org.apache.spark.util.collection.ExternalAppendOnlyMap$DiskMapIterator.readNextItem(ExternalAppendOnlyMap.scala:512)
	at org.apache.spark.util.collection.ExternalAppendOnlyMap$DiskMapIterator.hasNext(ExternalAppendOnlyMap.scala:533)
...

(This is our internal branch so don't worry if lines don't necessarily match.)

Issue Links

• is blocked by: SPARK-18547 Decouple I/O encryption key propagation from UserGroupInformation
• links to: [Github] Pull Request #15982 (vanzin)