Support Questions

Find answers, ask questions, and share your expertise
Announcements
Celebrating as our community reaches 100,000 members! Thank you!

Spark - Cannot mkdir file

avatar
Explorer

Hi,

 

I have an issue with Spark, the job failed with this error message :

 

scala> someDF.write.mode(SaveMode.Append).parquet("file:///data/bbox/tmp")
[Stage 0:>                                                          (0 + 2) / 2]18/06/05 12:37:39 WARN scheduler.TaskSetManager: Lost task 0.0 in stage 0.0 (TID 0, dec-bb-dl03.bbox-dec.lab.oxv.fr, executor 1): java.io.IOException: Mkdirs failed to create file:/data/bbox/tmp/_temporary/0/_temporary/attempt_201806051237_0000_m_000000_0 (exists=false, cwd=file:/yarn/nm/usercache/hdfs/appcache/application_1527756804026_0065/container_e33_1527756804026_0065_01_000002)
        at org.apache.hadoop.fs.ChecksumFileSystem.create(ChecksumFileSystem.java:447)
        at org.apache.hadoop.fs.ChecksumFileSystem.create(ChecksumFileSystem.java:433)
        at org.apache.hadoop.fs.FileSystem.create(FileSystem.java:926)
        at org.apache.hadoop.fs.FileSystem.create(FileSystem.java:907)
        at org.apache.hadoop.fs.FileSystem.create(FileSystem.java:804)
        at parquet.hadoop.ParquetFileWriter.<init>(ParquetFileWriter.java:225)
        at parquet.hadoop.ParquetOutputFormat.getRecordWriter(ParquetOutputFormat.java:311)
        at parquet.hadoop.ParquetOutputFormat.getRecordWriter(ParquetOutputFormat.java:282)
        at org.apache.spark.sql.execution.datasources.parquet.ParquetOutputWriter.<init>(ParquetRelation.scala:94)
        at org.apache.spark.sql.execution.datasources.parquet.ParquetRelation$$anon$3.newInstance(ParquetRelation.scala:286)
        at org.apache.spark.sql.execution.datasources.BaseWriterContainer.newOutputWriter(WriterContainer.scala:129)
        at org.apache.spark.sql.execution.datasources.DefaultWriterContainer.writeRows(WriterContainer.scala:255)
        at org.apache.spark.sql.execution.datasources.InsertIntoHadoopFsRelation$$anonfun$run$1$$anonfun$apply$mcV$sp$3.apply(InsertIntoHadoopFsRelation.scala:148)
        at org.apache.spark.sql.execution.datasources.InsertIntoHadoopFsRelation$$anonfun$run$1$$anonfun$apply$mcV$sp$3.apply(InsertIntoHadoopFsRelation.scala:148)
        at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:66)
        at org.apache.spark.scheduler.Task.run(Task.scala:89)
        at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:242)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
        at java.lang.Thread.run(Thread.java:748)

18/06/05 12:37:39 WARN scheduler.TaskSetManager: Lost task 0.1 in stage 0.0 (TID 2, dec-bb-dl03.bbox-dec.lab.oxv.fr, executor 1): java.io.IOException: Mkdirs failed to create file:/data/bbox/tmp/_temporary/0/_temporary/attempt_201806051237_0000_m_000000_1 (exists=false, cwd=file:/yarn/nm/usercache/hdfs/appcache/application_1527756804026_0065/container_e33_1527756804026_0065_01_000002)
        at org.apache.hadoop.fs.ChecksumFileSystem.create(ChecksumFileSystem.java:447)
        at org.apache.hadoop.fs.ChecksumFileSystem.create(ChecksumFileSystem.java:433)
        at org.apache.hadoop.fs.FileSystem.create(FileSystem.java:926)
        at org.apache.hadoop.fs.FileSystem.create(FileSystem.java:907)
        at org.apache.hadoop.fs.FileSystem.create(FileSystem.java:804)
        at parquet.hadoop.ParquetFileWriter.<init>(ParquetFileWriter.java:225)
        at parquet.hadoop.ParquetOutputFormat.getRecordWriter(ParquetOutputFormat.java:311)
        at parquet.hadoop.ParquetOutputFormat.getRecordWriter(ParquetOutputFormat.java:282)
        at org.apache.spark.sql.execution.datasources.parquet.ParquetOutputWriter.<init>(ParquetRelation.scala:94)
        at org.apache.spark.sql.execution.datasources.parquet.ParquetRelation$$anon$3.newInstance(ParquetRelation.scala:286)
        at org.apache.spark.sql.execution.datasources.BaseWriterContainer.newOutputWriter(WriterContainer.scala:129)
        at org.apache.spark.sql.execution.datasources.DefaultWriterContainer.writeRows(WriterContainer.scala:255)
        at org.apache.spark.sql.execution.datasources.InsertIntoHadoopFsRelation$$anonfun$run$1$$anonfun$apply$mcV$sp$3.apply(InsertIntoHadoopFsRelation.scala:148)
        at org.apache.spark.sql.execution.datasources.InsertIntoHadoopFsRelation$$anonfun$run$1$$anonfun$apply$mcV$sp$3.apply(InsertIntoHadoopFsRelation.scala:148)
        at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:66)
        at org.apache.spark.scheduler.Task.run(Task.scala:89)
        at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:242)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
        at java.lang.Thread.run(Thread.java:748)

 

 

We use CDH 5.14 with the Spark included into the CDH (1.6.0), we think about an version incompatibility issue.

 

First I tried to change directory rights (777 or give write right to hadoop group), but it didn't work.

 

Any idea ?

 

Julien.

 

 

 

 

1 ACCEPTED SOLUTION

avatar
Master Collaborator

Hi @JSenzier 

 

Right, this won't work in client mode. It's not about the compatibility of Spark1.6 with CDH version, but the way deploy mode 'client' works. spark-shell on Cloudera installs runs in yarn-client mode by default. Given the use of file:/// (which is generally used for local disks) we recommend running the app in local mode for such local testing or you can turn your script (using maven or sbt) into a jar file and execute this using spark-submit in cluster mode. 

 

$ spark-shell --master local[*]

View solution in original post

4 REPLIES 4

avatar
Master Collaborator

Hi @JSenzier 

 

Right, this won't work in client mode. It's not about the compatibility of Spark1.6 with CDH version, but the way deploy mode 'client' works. spark-shell on Cloudera installs runs in yarn-client mode by default. Given the use of file:/// (which is generally used for local disks) we recommend running the app in local mode for such local testing or you can turn your script (using maven or sbt) into a jar file and execute this using spark-submit in cluster mode. 

 

$ spark-shell --master local[*]

avatar
Explorer

Hi,

 

Thank you for your help, it's working, it's not very easy to understand when we encountered this issue (i didn't understand why yarn tried to create files into _temporary directory first), but with this explanation we can now understand this behaviour, so thank you 😉

avatar
Explorer

My simple ETL code:

def xmlConvert(spark):
etl_time = time.time()
df = spark.read.format('com.databricks.spark.xml').options(rowTag='HistoricalTextData').load(
'file:///home/zangetsu/proj/prometheus-core/demo/demo-1-iot-predictive-maintainance/dataset/data_train')
df = df.withColumn("TimeStamp", df["TimeStamp"].cast("timestamp")).groupBy("TimeStamp").pivot("TagName").sum(
"TagValue").na.fill(0)
df.repartition(1).write.csv(
path="file:///proj/prometheus-core/demo/demo-1-iot-predictive-maintainance/dataset/",
mode="overwrite",
header=True,
sep=",")
print("Time taken to do xml transformation: --- %s seconds ---" % (time.time() - etl_time))


if __name__ == '__main__':
spark = SparkSession \
.builder \
.appName('XML ETL') \
.master("local[*]") \
.config('spark.jars.packages', 'com.databricks:spark-xml_2.11:0.5.0') \
.getOrCreate()

print('Session created')

try:
xmlConvert(spark)

finally:
spark.stop()

Still throwing the issue reported.

avatar
Explorer

And I found a solution by pointint job.local.dir to directory with the code:

spark = SparkSession \
.builder \
.appName('XML ETL') \
.master("local[*]") \
.config('job.local.dir', 'file:/home/zangetsu/proj/prometheus-core/demo/demo-1-iot-predictive-maintainance') \
.config('spark.jars.packages', 'com.databricks:spark-xml_2.11:0.5.0') \
.getOrCreate()

Now all works