I am using Spark Structured Streaming (2.3) to write parquet data to buckets in the cloud ( Google Cloud Storage).
I am using the following function :
def writeStreaming(data: DataFrame, format: String, options: Map[String, String], partitions: List[String]): DataStreamWriter[Row] = {
var dataStreamWrite = data.writeStream .format(format).options(options).trigger(Trigger.ProcessingTime("120 seconds"))
if (!partitions.isEmpty)
dataStreamWrite = ddataStreamWrite.partitionBy(partitions: _*)
dataStreamWrite
}
Unfortunately, with this approach, I am getting many small files.
I tried to use the trigger approach in order to avoid this, but this didn't work too. Do you have any idea about how to handle this, please ?
Thanks a lot
The reason that you have many small files despite using trigger can be your dataframe having many partitions. To reduce the the parquet to 1 file/ 2 mins, you can coalesce to one partition before writing Parquet files.
var dataStreamWrite = data
.coalesce(1)
.writeStream
.format(format)
.options(options)
.trigger(Trigger.ProcessingTime("120 seconds"))
Related
We have an AWS S3 bucket with millions of documents in a complex hierarchy, and a CSV file with (among other data) links to a subset of those files, I estimate this file will be about 1000 to 10.000 rows. I need to join the data from the CSV file with the contents of the documents for further processing in Spark. In case it matters, we're using Scala and Spark 2.4.4 on an Amazon EMR 6.0.0 cluster.
I can think of two ways to do this. First is to add a transformation on the CSV DataFrame that adds the content as a new column:
val df = spark.read.format("csv").load("<csv file>")
val attempt1 = df.withColumn("raw_content", spark.sparkContext.textFile($"document_url"))
or variations thereof (for example, wrapping it in a udf) don't seem to work, I think because sparkContext.textFile returns an RDD, so I'm not sure it's even supposed to work this way? Even if I get it working, is the best way to keep it performant in Spark?
An alternative I tried to think of is to use spark.sparkContext.wholeTextFiles upfront and then join the two dataframes together:
val df = spark.read.format("csv").load("<csv file>")
val contents = spark.sparkContext.wholeTextFiles("<s3 bucket>").toDF("document_url", "raw_content");
val attempt2 = df.join(contents, df("document_url") === contents("document_url"), "left")
but wholeTextFiles doesn't go into subdirectories and the needed paths are hard to predict, and I'm also unsure of the performance impact of trying to build an RDD of the entire bucket of millions of files if I only need a small fraction of it, since the S3 API probably doesn't make it very fast to list all the objects in the bucket.
Any ideas? Thanks!
I did figure out a solution in the end:
val df = spark.read.format("csv").load("<csv file>")
val allS3Links = df.map(row => row.getAs[String]("document_url")).collect()
val joined = allS3Links.mkString(",")
val contentsDF = spark.sparkContext.wholeTextFiles(joined).toDF("document_url", "raw_content");
The downside to this solution is that it pulls all the urls to the driver, but it's workable in my case (100,000 * ~100 char length strings is not that much) and maybe even unavoidable.
I need to write Cassandra Partitions as parquet file. Since I cannot share and use sparkSession in foreach function. Firstly, I call collect method to collect all data in driver program then I write parquet file to HDFS, as below.
Thanks to this link https://github.com/datastax/spark-cassandra-connector/blob/master/doc/16_partitioning.md
I am able to get my partitioned rows. I want to write partitioned rows into seperated parquet file, whenever a partition is read from cassandra table. I also tried sparkSQLContext that method writes task results as temporary. I think, after all the tasks are done. I will see parquet files.
Is there any convenient method for this?
val keyedTable : CassandraTableScanRDD[(Tuple2[Int, Date], MyCassandraTable)] = getTableAsKeyed()
keyedTable.groupByKey
.collect
.foreach(f => {
import sparkSession.implicits._
val items = f._2.toList
val key = f._1
val baseHDFS = "hdfs://mycluster/parquet_test/"
val ds = sparkSession.sqlContext.createDataset(items)
ds.write
.option("compression", "gzip")
.parquet(baseHDFS + key._1 + "/" + key._2)
})
Why not use Spark SQL everywhere & use built-in functionality of the Parquet to write data by partitions, instead of creating a directory hierarchy yourself?
Something like this:
import org.apache.spark.sql.cassandra._
val data = spark.read.cassandraFormat("table", "keyspace").load()
data.write
.option("compression", "gzip")
.partitionBy("col1", "col2")
.parquet(baseHDFS)
In this case, it will create a separate directory for every value of col & col2 as nested directories, with name like this: ${column}=${value}. Then when you read, you may force to read only specific value.
I have 2 instances for the same data.
Hive table called myData in parquet format
Parquet file (not managed by Hive) that is in parquet format
Consider the following code:
val myCoolDataSet = spark
.sql("select * from myData")
.select("col1", "col2")
.as[MyDataSet]
.filter(x => x.col1 == "Dummy")
And this one:
val myCoolDataSet = spark
.read
.parquet("path_to_file")
.select("col1", "col2")
.as[MyDataSet]
.filter(x => x.col1 == "Dummy")
My question is what is better in terms of performance and amount of scanned data?
How spark computes it for the 2 different approaches?
Hive serves as a storage for metadata about the Parquet file. Spark can leverage the information contained therein to perform interesting optimizations. Since the backing storage is the same you'll probably not see much difference, but the optimizations based on the metadata in Hive can give an edge.
I am using Spark 2.1.0 and Kafka 0.9.0.
I am trying to push the output of a batch spark job to kafka. The job is supposed to run every hour but not as streaming.
While looking for an answer on the net I could only find kafka integration with Spark streaming and nothing about the integration with the batch job.
Does anyone know if such thing is feasible ?
Thanks
UPDATE :
As mentioned by user8371915, I tried to follow what was done in Writing the output of Batch Queries to Kafka.
I used a spark shell :
spark-shell --packages org.apache.spark:spark-sql-kafka-0-10_2.11:2.1.0
Here is the simple code that I tried :
val df = Seq(("Rey", "23"), ("John", "44")).toDF("key", "value")
val newdf = df.select(to_json(struct(df.columns.map(column):_*)).alias("value"))
newdf.write.format("kafka").option("kafka.bootstrap.servers", "localhost:9092").option("topic", "alerts").save()
But I get the error :
java.lang.RuntimeException: org.apache.spark.sql.kafka010.KafkaSourceProvider does not allow create table as select.
at scala.sys.package$.error(package.scala:27)
at org.apache.spark.sql.execution.datasources.DataSource.write(DataSource.scala:497)
at org.apache.spark.sql.DataFrameWriter.save(DataFrameWriter.scala:215)
... 50 elided
Have any idea what is this related to ?
Thanks
tl;dr You use outdated Spark version. Writes are enabled in 2.2 and later.
Out-of-the-box you can use Kafka SQL connector (the same as used with Structured Streaming). Include
spark-sql-kafka in your dependencies.
Convert data to DataFrame containing at least value column of type StringType or BinaryType.
Write data to Kafka:
df
.write
.format("kafka")
.option("kafka.bootstrap.servers", server)
.save()
Follow Structured Streaming docs for details (starting with Writing the output of Batch Queries to Kafka).
If you have a dataframe and you want to write it to a kafka topic, you need to convert columns first to a "value" column that contains data in a json format. In scala it is
import org.apache.spark.sql.functions._
val kafkaServer: String = "localhost:9092"
val topicSampleName: String = "kafkatopic"
df.select(to_json(struct("*")).as("value"))
.selectExpr("CAST(value AS STRING)")
.write
.format("kafka")
.option("kafka.bootstrap.servers", kafkaServer)
.option("topic", topicSampleName)
.save()
For this error
java.lang.RuntimeException: org.apache.spark.sql.kafka010.KafkaSourceProvider does not allow create table as select.
at scala.sys.package$.error(package.scala:27)
I think you need to parse the message to Key value pair. Your dataframe should have value column.
Let say if you have a dataframe with student_id, scores.
df.show()
>> student_id | scores
1 | 99.00
2 | 98.00
then you should modify your dataframe to
value
{"student_id":1,"score":99.00}
{"student_id":2,"score":98.00}
To convert you can use similar code like this
df.select(to_json(struct($"student_id",$"score")).alias("value"))
I've got a fairly simple job coverting log files to parquet. It's processing 1.1TB of data (chunked into 64MB - 128MB files - our block size is 128MB), which is approx 12 thousand files.
Job works as follows:
val events = spark.sparkContext
.textFile(s"$stream/$sourcetype")
.map(_.split(" \\|\\| ").toList)
.collect{case List(date, y, "Event") => MyEvent(date, y, "Event")}
.toDF()
df.write.mode(SaveMode.Append).partitionBy("date").parquet(s"$path")
It collects the events with a common schema, converts to a DataFrame, and then writes out as parquet.
The problem I'm having is that this can create a bit of an IO explosion on the HDFS cluster, as it's trying to create so many tiny files.
Ideally I want to create only a handful of parquet files within the partition 'date'.
What would be the best way to control this? Is it by using 'coalesce()'?
How will that effect the amount of files created in a given partition? Is it dependent on how many executors I have working in Spark? (currently set at 100).
you have to repartiton your DataFrame to match the partitioning of the DataFrameWriter
Try this:
df
.repartition($"date")
.write.mode(SaveMode.Append)
.partitionBy("date")
.parquet(s"$path")
In Python you can rewrite Raphael's Roth answer as:
(df
.repartition("date")
.write.mode("append")
.partitionBy("date")
.parquet("{path}".format(path=path)))
You might also consider adding more columns to .repartition to avoid problems with very large partitions:
(df
.repartition("date", another_column, yet_another_colum)
.write.mode("append")
.partitionBy("date)
.parquet("{path}".format(path=path)))
The simplest solution would be to replace your actual partitioning by :
df
.repartition(to_date($"date"))
.write.mode(SaveMode.Append)
.partitionBy("date")
.parquet(s"$path")
You can also use more precise partitioning for your DataFrame i.e the day and maybe the hour of an hour range. and then you can be less precise for writer.
That actually depends on the amount of data.
You can reduce entropy by partitioning DataFrame and the write with partition by clause.
I came across the same issue and I could using coalesce solved my problem.
df
.coalesce(3) // number of parts/files
.write.mode(SaveMode.Append)
.parquet(s"$path")
For more information on using coalesce or repartition you can refer to the following spark: coalesce or repartition
Duplicating my answer from here: https://stackoverflow.com/a/53620268/171916
This is working for me very well:
data.repartition(n, "key").write.partitionBy("key").parquet("/location")
It produces N files in each output partition (directory), and is (anecdotally) faster than using coalesce and (again, anecdotally, on my data set) faster than only repartitioning on the output.
If you're working with S3, I also recommend doing everything on local drives (Spark does a lot of file creation/rename/deletion during write outs) and once it's all settled use hadoop FileUtil (or just the aws cli) to copy everything over:
import java.net.URI
import org.apache.hadoop.fs.{FileSystem, FileUtil, Path}
// ...
def copy(
in : String,
out : String,
sparkSession: SparkSession
) = {
FileUtil.copy(
FileSystem.get(new URI(in), sparkSession.sparkContext.hadoopConfiguration),
new Path(in),
FileSystem.get(new URI(out), sparkSession.sparkContext.hadoopConfiguration),
new Path(out),
false,
sparkSession.sparkContext.hadoopConfiguration
)
}
how about trying running scripts like this as map job consolidating all the parquet files into one:
$ hadoop jar /usr/hdp/2.3.2.0-2950/hadoop-mapreduce/hadoop-streaming-2.7.1.2.3.2.0-2950.jar \
-Dmapred.reduce.tasks=1 \
-input "/hdfs/input/dir" \
-output "/hdfs/output/dir" \
-mapper cat \
-reducer cat