[SPARK-18414] sc.textFile doesn't seem to use LzoTextInputFormat when hadoop-lzo is installed

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Not A Problem
• Affects Version/s: 2.0.1
• Component/s: Input/Output
• Labels: hadoop-lzo
• Assignee: Unassigned
• Reporter: Renan Vicente Gomes da Silva
• Created: 11/Nov/16 16:36
• Updated: 22/Nov/16 15:26
• Resolved: 22/Nov/16 15:26

Description

When reading LZO files using sc.textFile it miss a few files from time to time.

Sample:
val Data = sc.textFile(Files)
listFiles += Data.count()

Considering that Files is a HDFS directory containing LZO files. If executed for example a 1000 times it gets different results a few times.

Now if you use newAPIHadoopFile to force it to use com.hadoop.mapreduce.LzoTextInputFormat it works perfectly, shows the same results in all executions.

Sample:

val Data = sc.newAPIHadoopFile(Files,
classOf[com.hadoop.mapreduce.LzoTextInputFormat],
classOf[org.apache.hadoop.io.LongWritable],
classOf[org.apache.hadoop.io.Text]).map(_._2.toString)
listFiles += Data.count()

Looking at Spark code it looks like it use TextInputFormat by default but is not using com.hadoop.mapreduce.LzoTextInputFormat when hadoop-lzo is installed.

https://github.com/apache/spark/blob/v2.0.1/core/src/main/scala/org/apache/spark/SparkContext.scala#L795-L801