I'm successfully loading my sequence files into a DataFrame with some code like this:
val sc = new SparkContext(conf)
val sqlContext = new org.apache.spark.sql.SQLContext(sc)
sc.sequenceFile[LongWritable,String](src)
val jsonRecs = file.map((record: (String, String)) => new String(record._2))
val df = sqlContext.read.json(jsonRecs)
I'd like to do the same with some text files. The text files have a similar format as the sequence files (A timestamp, a tab char, then the json). But the problem is textFile() returns an RDD[String] instead of an RDD[LongWritable,String] like the sequenceFile() method.
My goal is to be able to test the program with either sequence files or text files as input.
How could I convert the RDD[String] coming from textFile() into an RDD[LongWritable,String]? Or is there a better solution?
Assuming that your text file is a CSV file, you can use following code for reading a CSV file in a Dataframe where spark is the SparkSession:
val df = spark.read.option("header", "false").csv("file.txt")
Like header option there are multiple options you can provide depending upon your requirement. Check this for more details.
Thanks for the responses. It's not a CSV but I guess it could be. It's just the text output of doing this on a sequence file in HDFS:
hdfs dfs -text /path/to/my/file > myFile.txt
Anyway, I found a solution that works for both sequence and text file for my use case. This code ends up setting the variable 'file' to a RDD[String,String] in both cases, and I can work with that.
var file = if (inputType.equalsIgnoreCase("text")) {
sc.textFile(src).map(line => (line.split("\t")(0), line.split("\t")(1)))
} else { // Default to assuming sequence files are input
sc.sequenceFile[String,String](src)
}
Related
I have a requirement where I want to write each individual records in an RDD to an individual file in HDFS.
I did it for the normal filesystem but obviously, it doesn't work for HDFS.
stream.foreachRDD{ rdd =>
if(!rdd.isEmpty()) {
rdd.foreach{
msg =>
val value = msg._2
println(value)
val fname = java.util.UUID.randomUUID.toString
val path = dir + fname
write(path, value)
}
}
}
where write is a function which writes to the filesystem.
Is there a way to do it within spark so that for each record I can natively write to the HDFS, without using any other tool like Kafka Connect or Flume??
EDIT: More Explanation
For eg:
If my DstreamRDD has the following records,
abcd
efgh
ijkl
mnop
I need different files for each record, so different file for "abcd", different for "efgh" and so on.
I tried creating an RDD within the streamRDD but I learnt it's not allowed as the RDD's are not serializable.
You can forcefully repartition the rdd to no. of partitions as many no. of records and then save
val rddCount = rdd.count()
rdd.repartition(rddCount).saveAsTextFile("your/hdfs/loc")
You can do in couple of ways..
From rdd, you can get the sparkCOntext, once you got the sparkCOntext, you can use parallelize method and pass the String as List of String.
For example:
val sc = rdd.sparkContext
sc.parallelize(Seq("some string")).saveAsTextFile(path)
Also, you can use sqlContext to convert the string to DF then write in the file.
for Example:
import sqlContext.implicits._
Seq(("some string")).toDF
I am trying to read multiple csvs into an rdd from a path. This path has many csvs Is there a way I can avoid the headers while reading all the csvs into rdd? or use spotsRDD to omit out the header without having to use filter or deal with each csv individually and then union them?
val path ="file:///home/work/csvs/*"
val spotsRDD= sc.textFile(path)
println(spotsRDD.count())
Thanks
That is pity you are using spark 1.0.0.
You can use CSV Data Source for Apache Spark but this library requires Spark 1.3+ and btw. this library was inlined to Spark 2.x.
But we can analyse and implement something similar.
When we look into the com/databricks/spark/csv/DefaultSource.scala there is
val useHeader = parameters.getOrElse("header", "false")
and then in the com/databricks/spark/csv/CsvRelation.scala there is
// If header is set, make sure firstLine is materialized before sending to executors.
val filterLine = if (useHeader) firstLine else null
baseRDD().mapPartitions { iter =>
// When using header, any input line that equals firstLine is assumed to be header
val csvIter = if (useHeader) {
iter.filter(_ != filterLine)
} else {
iter
}
parseCSV(csvIter, csvFormat)
so if we assume the first line is only once in RDD (our csv rows) we can do something like in the example below:
CSV example file:
Latitude,Longitude,Name
48.1,0.25,"First point"
49.2,1.1,"Second point"
47.5,0.75,"Third point"
scala> val csvData = sc.textFile("test.csv")
csvData: org.apache.spark.rdd.RDD[String] = test.csv MapPartitionsRDD[24] at textFile at <console>:24
scala> val header = csvDataRdd.first
header: String = Latitude,Longitude,Name
scala> val csvDataWithoutHeaderRdd = csvDataRdd.mapPartitions{iter => iter.filter(_ != header)}
csvDataWithoutHeaderRdd: org.apache.spark.rdd.RDD[String] = MapPartitionsRDD[25] at mapPartitions at <console>:28
scala> csvDataWithoutHeaderRdd.foreach(println)
49.2,1.1,"Second point"
48.1,0.25,"First point"
47.5,0.75,"Third point"
I want to create a data processing pipeline in AWS to eventually use the processed data for Machine Learning.
I have a Scala script that takes raw data from S3, processes it and writes it to HDFS or even S3 with Spark-CSV. I think I can use multiple files as input if I want to use AWS Machine Learning tool for training a prediction model. But if I want to use something else, I presume it is best if I receive a single CSV output file.
Currently, as I do not want to use repartition(1) nor coalesce(1) for performance purposes, I have used hadoop fs -getmerge for manual testing, but as it just merges the contents of the job output files, I am running into a small problem. I need a single row of headers in the data file for training the prediction model.
If I use .option("header","true") for the spark-csv, then it writes the headers to every output file and after merging I have as many lines of headers in the data as there were output files. But if the header option is false, then it does not add any headers.
Now I found an option to merge the files inside the Scala script with Hadoop API FileUtil.copyMerge. I tried this in spark-shell with the code below.
import org.apache.hadoop.fs.FileUtil
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.Path;
val configuration = new Configuration();
val fs = FileSystem.get(configuration);
FileUtil.copyMerge(fs, new Path("smallheaders"), fs, new Path("/home/hadoop/smallheaders2"), false, configuration, "")
But this solution still just concatenates the files on top of each other and does not handle headers. How can I get an output file with only one row of headers?
I even tried adding df.columns.mkString(",") as the last argument for copyMerge, but this added the headers still multiple times, not once.
you can walk around like this.
1.Create a new DataFrame(headerDF) containing header names.
2.Union it with the DataFrame(dataDF) containing the data.
3.Output the union-ed DataFrame to disk with option("header", "false").
4.merge partition files(part-0000**0.csv) using hadoop FileUtil
In this ways, all partitions have no header except for a single partition's content has a row of header names from the headerDF. When all partitions are merged together, there is a single header in the top of the file. Sample code are the following
//dataFrame is the data to save on disk
//cast types of all columns to String
val dataDF = dataFrame.select(dataFrame.columns.map(c => dataFrame.col(c).cast("string")): _*)
//create a new data frame containing only header names
import scala.collection.JavaConverters._
val headerDF = sparkSession.createDataFrame(List(Row.fromSeq(dataDF.columns.toSeq)).asJava, dataDF.schema)
//merge header names with data
headerDF.union(dataDF).write.mode(SaveMode.Overwrite).option("header", "false").csv(outputFolder)
//use hadoop FileUtil to merge all partition csv files into a single file
val fs = FileSystem.get(sparkSession.sparkContext.hadoopConfiguration)
FileUtil.copyMerge(fs, new Path(outputFolder), fs, new Path("/folder/target.csv"), true, spark.sparkContext.hadoopConfiguration, null)
Output the header using dataframe.schema
( val header = dataDF.schema.fieldNames.reduce(_ + "," + _))
create a file with the header on dsefs
append all the partition files (headerless) to the file in #2 using hadoop Filesystem API
We had a similar issue, following the below approach to get single output file-
Write dataframe to hdfs with headers and without using coalesce or repartition (after the transformations).
dataframe.write.format("csv").option("header", "true").save(hdfs_path_for_multiple_files)
Read the files from the previous step and write back to different location on hdfs with coalesce(1).
dataframe = spark.read.option('header', 'true').csv(hdfs_path_for_multiple_files)
dataframe.coalesce(1).write.format('csv').option('header', 'true').save(hdfs_path_for_single_file)
This way, you will avoid performance issues related to coalesce or repartition while execution of transformations (Step 1).
And the second step provides single output file with one header line.
To merge files in a folder into one file:
import org.apache.hadoop.conf.Configuration
import org.apache.hadoop.fs._
def merge(srcPath: String, dstPath: String): Unit = {
val hadoopConfig = new Configuration()
val hdfs = FileSystem.get(hadoopConfig)
FileUtil.copyMerge(hdfs, new Path(srcPath), hdfs, new Path(dstPath), false, hadoopConfig, null)
}
If you want to merge all files into one file, but still in the same folder (but this brings all data to the driver node):
dataFrame
.coalesce(1)
.write
.format("com.databricks.spark.csv")
.option("header", "true")
.save(out)
Another solution would be to use solution #2 then move the one file inside the folder to another path (with the name of our CSV file).
def df2csv(df: DataFrame, fileName: String, sep: String = ",", header: Boolean = false): Unit = {
val tmpDir = "tmpDir"
df.repartition(1)
.write
.format("com.databricks.spark.csv")
.option("header", header.toString)
.option("delimiter", sep)
.save(tmpDir)
val dir = new File(tmpDir)
val tmpCsvFile = tmpDir + File.separatorChar + "part-00000"
(new File(tmpCsvFile)).renameTo(new File(fileName))
dir.listFiles.foreach( f => f.delete )
dir.delete
}
Try to specify the schema of the header and read all file from the folder using the option drop malformed of spark-csv. This should let you read all the files in the folder keeping only the headers (because you drop the malformed).
Example:
val headerSchema = List(
StructField("example1", StringType, true),
StructField("example2", StringType, true),
StructField("example3", StringType, true)
)
val header_DF =sqlCtx.read
.option("delimiter", ",")
.option("header", "false")
.option("mode","DROPMALFORMED")
.option("inferSchema","false")
.schema(StructType(headerSchema))
.format("com.databricks.spark.csv")
.load("folder containg the files")
In header_DF you will have only the rows of the headers, from this you can trasform the dataframe the way you need.
// Convert JavaRDD to CSV and save as text file
outputDataframe.write()
.format("com.databricks.spark.csv")
// Header => true, will enable to have header in each file
.option("header", "true")
Please follow the link with Integration test on how to write a single header
http://bytepadding.com/big-data/spark/write-a-csv-text-file-from-spark/
I am using https://github.com/databricks/spark-csv , I am trying to write a single CSV, but not able to, it is making a folder.
Need a Scala function which will take parameter like path and file name and write that CSV file.
It is creating a folder with multiple files, because each partition is saved individually. If you need a single output file (still in a folder) you can repartition (preferred if upstream data is large, but requires a shuffle):
df
.repartition(1)
.write.format("com.databricks.spark.csv")
.option("header", "true")
.save("mydata.csv")
or coalesce:
df
.coalesce(1)
.write.format("com.databricks.spark.csv")
.option("header", "true")
.save("mydata.csv")
data frame before saving:
All data will be written to mydata.csv/part-00000. Before you use this option be sure you understand what is going on and what is the cost of transferring all data to a single worker. If you use distributed file system with replication, data will be transfered multiple times - first fetched to a single worker and subsequently distributed over storage nodes.
Alternatively you can leave your code as it is and use general purpose tools like cat or HDFS getmerge to simply merge all the parts afterwards.
If you are running Spark with HDFS, I've been solving the problem by writing csv files normally and leveraging HDFS to do the merging. I'm doing that in Spark (1.6) directly:
import org.apache.hadoop.conf.Configuration
import org.apache.hadoop.fs._
def merge(srcPath: String, dstPath: String): Unit = {
val hadoopConfig = new Configuration()
val hdfs = FileSystem.get(hadoopConfig)
FileUtil.copyMerge(hdfs, new Path(srcPath), hdfs, new Path(dstPath), true, hadoopConfig, null)
// the "true" setting deletes the source files once they are merged into the new output
}
val newData = << create your dataframe >>
val outputfile = "/user/feeds/project/outputs/subject"
var filename = "myinsights"
var outputFileName = outputfile + "/temp_" + filename
var mergedFileName = outputfile + "/merged_" + filename
var mergeFindGlob = outputFileName
newData.write
.format("com.databricks.spark.csv")
.option("header", "false")
.mode("overwrite")
.save(outputFileName)
merge(mergeFindGlob, mergedFileName )
newData.unpersist()
Can't remember where I learned this trick, but it might work for you.
I might be a little late to the game here, but using coalesce(1) or repartition(1) may work for small data-sets, but large data-sets would all be thrown into one partition on one node. This is likely to throw OOM errors, or at best, to process slowly.
I would highly suggest that you use the FileUtil.copyMerge() function from the Hadoop API. This will merge the outputs into a single file.
EDIT - This effectively brings the data to the driver rather than an executor node. Coalesce() would be fine if a single executor has more RAM for use than the driver.
EDIT 2: copyMerge() is being removed in Hadoop 3.0. See the following stack overflow article for more information on how to work with the newest version: How to do CopyMerge in Hadoop 3.0?
If you are using Databricks and can fit all the data into RAM on one worker (and thus can use .coalesce(1)), you can use dbfs to find and move the resulting CSV file:
val fileprefix= "/mnt/aws/path/file-prefix"
dataset
.coalesce(1)
.write
//.mode("overwrite") // I usually don't use this, but you may want to.
.option("header", "true")
.option("delimiter","\t")
.csv(fileprefix+".tmp")
val partition_path = dbutils.fs.ls(fileprefix+".tmp/")
.filter(file=>file.name.endsWith(".csv"))(0).path
dbutils.fs.cp(partition_path,fileprefix+".tab")
dbutils.fs.rm(fileprefix+".tmp",recurse=true)
If your file does not fit into RAM on the worker, you may want to consider chaotic3quilibrium's suggestion to use FileUtils.copyMerge(). I have not done this, and don't yet know if is possible or not, e.g., on S3.
This answer is built on previous answers to this question as well as my own tests of the provided code snippet. I originally posted it to Databricks and am republishing it here.
The best documentation for dbfs's rm's recursive option I have found is on a Databricks forum.
spark's df.write() API will create multiple part files inside given path ... to force spark write only a single part file use df.coalesce(1).write.csv(...) instead of df.repartition(1).write.csv(...) as coalesce is a narrow transformation whereas repartition is a wide transformation see Spark - repartition() vs coalesce()
df.coalesce(1).write.csv(filepath,header=True)
will create folder in given filepath with one part-0001-...-c000.csv file
use
cat filepath/part-0001-...-c000.csv > filename_you_want.csv
to have a user friendly filename
This answer expands on the accepted answer, gives more context, and provides code snippets you can run in the Spark Shell on your machine.
More context on accepted answer
The accepted answer might give you the impression the sample code outputs a single mydata.csv file and that's not the case. Let's demonstrate:
val df = Seq("one", "two", "three").toDF("num")
df
.repartition(1)
.write.csv(sys.env("HOME")+ "/Documents/tmp/mydata.csv")
Here's what's outputted:
Documents/
tmp/
mydata.csv/
_SUCCESS
part-00000-b3700504-e58b-4552-880b-e7b52c60157e-c000.csv
N.B. mydata.csv is a folder in the accepted answer - it's not a file!
How to output a single file with a specific name
We can use spark-daria to write out a single mydata.csv file.
import com.github.mrpowers.spark.daria.sql.DariaWriters
DariaWriters.writeSingleFile(
df = df,
format = "csv",
sc = spark.sparkContext,
tmpFolder = sys.env("HOME") + "/Documents/better/staging",
filename = sys.env("HOME") + "/Documents/better/mydata.csv"
)
This'll output the file as follows:
Documents/
better/
mydata.csv
S3 paths
You'll need to pass s3a paths to DariaWriters.writeSingleFile to use this method in S3:
DariaWriters.writeSingleFile(
df = df,
format = "csv",
sc = spark.sparkContext,
tmpFolder = "s3a://bucket/data/src",
filename = "s3a://bucket/data/dest/my_cool_file.csv"
)
See here for more info.
Avoiding copyMerge
copyMerge was removed from Hadoop 3. The DariaWriters.writeSingleFile implementation uses fs.rename, as described here. Spark 3 still used Hadoop 2, so copyMerge implementations will work in 2020. I'm not sure when Spark will upgrade to Hadoop 3, but better to avoid any copyMerge approach that'll cause your code to break when Spark upgrades Hadoop.
Source code
Look for the DariaWriters object in the spark-daria source code if you'd like to inspect the implementation.
PySpark implementation
It's easier to write out a single file with PySpark because you can convert the DataFrame to a Pandas DataFrame that gets written out as a single file by default.
from pathlib import Path
home = str(Path.home())
data = [
("jellyfish", "JALYF"),
("li", "L"),
("luisa", "LAS"),
(None, None)
]
df = spark.createDataFrame(data, ["word", "expected"])
df.toPandas().to_csv(home + "/Documents/tmp/mydata-from-pyspark.csv", sep=',', header=True, index=False)
Limitations
The DariaWriters.writeSingleFile Scala approach and the df.toPandas() Python approach only work for small datasets. Huge datasets can not be written out as single files. Writing out data as a single file isn't optimal from a performance perspective because the data can't be written in parallel.
I'm using this in Python to get a single file:
df.toPandas().to_csv("/tmp/my.csv", sep=',', header=True, index=False)
A solution that works for S3 modified from Minkymorgan.
Simply pass the temporary partitioned directory path (with different name than final path) as the srcPath and single final csv/txt as destPath Specify also deleteSource if you want to remove the original directory.
/**
* Merges multiple partitions of spark text file output into single file.
* #param srcPath source directory of partitioned files
* #param dstPath output path of individual path
* #param deleteSource whether or not to delete source directory after merging
* #param spark sparkSession
*/
def mergeTextFiles(srcPath: String, dstPath: String, deleteSource: Boolean): Unit = {
import org.apache.hadoop.fs.FileUtil
import java.net.URI
val config = spark.sparkContext.hadoopConfiguration
val fs: FileSystem = FileSystem.get(new URI(srcPath), config)
FileUtil.copyMerge(
fs, new Path(srcPath), fs, new Path(dstPath), deleteSource, config, null
)
}
import org.apache.hadoop.conf.Configuration
import org.apache.hadoop.fs._
import org.apache.spark.sql.{DataFrame,SaveMode,SparkSession}
import org.apache.spark.sql.functions._
I solved using below approach (hdfs rename file name):-
Step 1:- (Crate Data Frame and write to HDFS)
df.coalesce(1).write.format("csv").option("header", "false").mode(SaveMode.Overwrite).save("/hdfsfolder/blah/")
Step 2:- (Create Hadoop Config)
val hadoopConfig = new Configuration()
val hdfs = FileSystem.get(hadoopConfig)
Step3 :- (Get path in hdfs folder path)
val pathFiles = new Path("/hdfsfolder/blah/")
Step4:- (Get spark file names from hdfs folder)
val fileNames = hdfs.listFiles(pathFiles, false)
println(fileNames)
setp5:- (create scala mutable list to save all the file names and add it to the list)
var fileNamesList = scala.collection.mutable.MutableList[String]()
while (fileNames.hasNext) {
fileNamesList += fileNames.next().getPath.getName
}
println(fileNamesList)
Step 6:- (filter _SUCESS file order from file names scala list)
// get files name which are not _SUCCESS
val partFileName = fileNamesList.filterNot(filenames => filenames == "_SUCCESS")
step 7:- (convert scala list to string and add desired file name to hdfs folder string and then apply rename)
val partFileSourcePath = new Path("/yourhdfsfolder/"+ partFileName.mkString(""))
val desiredCsvTargetPath = new Path(/yourhdfsfolder/+ "op_"+ ".csv")
hdfs.rename(partFileSourcePath , desiredCsvTargetPath)
spark.sql("select * from df").coalesce(1).write.option("mode","append").option("header","true").csv("/your/hdfs/path/")
spark.sql("select * from df") --> this is dataframe
coalesce(1) or repartition(1) --> this will make your output file to 1 part file only
write --> writing data
option("mode","append") --> appending data to existing directory
option("header","true") --> enabling header
csv("<hdfs dir>") --> write as CSV file & its output location in HDFS
repartition/coalesce to 1 partition before you save (you'd still get a folder but it would have one part file in it)
you can use rdd.coalesce(1, true).saveAsTextFile(path)
it will store data as singile file in path/part-00000
Here is a helper function with which you can get a single result-file without the part-0000 and without a subdirectory on S3 and AWS EMR:
def renameSinglePartToParentFolder(directoryUrl: String): Unit = {
import sys.process._
val lsResult = s"aws s3 ls ${directoryUrl}/" !!
val partFilename = lsResult.split("\n").map(_.split(" ").last).filter(_.contains("part-0000")).last
s"aws s3 rm ${directoryUrl}/_SUCCESS" !
s"aws s3 mv ${directoryUrl}/${partFilename} ${directoryUrl}" !
}
val targetPath = "s3://my-bucket/my-folder/my-file.csv"
df.coalesce(1).write.csv(targetPath)
renameSinglePartToParentFolder(targetPath)
Write to a single part-0000... file.
Use AWS CLI to list all files and rename the single file accordingly.
by using Listbuffer we can save data into single file:
import java.io.FileWriter
import org.apache.spark.sql.SparkSession
import scala.collection.mutable.ListBuffer
val text = spark.read.textFile("filepath")
var data = ListBuffer[String]()
for(line:String <- text.collect()){
data += line
}
val writer = new FileWriter("filepath")
data.foreach(line => writer.write(line.toString+"\n"))
writer.close()
def export_csv(
fileName: String,
filePath: String
) = {
val filePathDestTemp = filePath + ".dir/"
val merstageout_df = spark.sql(merstageout)
merstageout_df
.coalesce(1)
.write
.option("header", "true")
.mode("overwrite")
.csv(filePathDestTemp)
val listFiles = dbutils.fs.ls(filePathDestTemp)
for(subFiles <- listFiles){
val subFiles_name: String = subFiles.name
if (subFiles_name.slice(subFiles_name.length() - 4,subFiles_name.length()) == ".csv") {
dbutils.fs.cp (filePathDestTemp + subFiles_name, filePath + fileName+ ".csv")
dbutils.fs.rm(filePathDestTemp, recurse=true)
}}}
There is one more way to use Java
import java.io._
def printToFile(f: java.io.File)(op: java.io.PrintWriter => Unit)
{
val p = new java.io.PrintWriter(f);
try { op(p) }
finally { p.close() }
}
printToFile(new File("C:/TEMP/df.csv")) { p => df.collect().foreach(p.println)}
In Spark, we can use textFile to load file into lines and try to do some operations with these lines as follows.
val lines = sc.textFile("xxx")
val counts = lines.filter(line => lines.contains("a")).count()
However, in my situation, I would like to load the file into blocks because the data in files and the block will be kind of as follow. Blocks will be separated with empty line in files.
user: 111
book: 222
comments: like it!
Therefore, I hope the textFile function or any other solutions can help me load the file with blocks, which may be achieved as follows.
val blocks = sc.textFile("xxx", 3 line)
Does anyone face this situation before? Thanks
I suggest you to implement your own file reader function from Hdfs. Look attextFile function, it's built on top of hadoopFile function and it uses TextInputFormat:
def textFile(path: String, minPartitions: Int = defaultMinPartitions): RDD[String] = {
assertNotStopped()
hadoopFile(path, classOf[TextInputFormat], classOf[LongWritable], classOf[Text],
minPartitions).map(pair => pair._2.toString).setName(path)
}
But this TextInputFormat can be customized via hadoop properties as described in this answer. In your case delimiter could be:
conf.set("textinputformat.record.delimiter", "\n\n")