Apache Spark(TsFile)

February 23, 2023About 5 min

Apache Spark(TsFile)

About Spark-TsFile-Connector

Spark-TsFile-Connector implements the support of Spark for external data sources of Tsfile type. This enables users to read, write and query Tsfile by Spark.

With this connector, you can

load a single TsFile, from either the local file system or hdfs, into Spark
load all files in a specific directory, from either the local file system or hdfs, into Spark
write data from Spark into TsFile

System Requirements

Spark Version	Scala Version	Java Version	TsFile
`2.4.3`	`2.11.8`	`1.8`	`1.0.0`

Note: For more information about how to download and use TsFile, please see the following link: https://github.com/apache/iotdb/tree/master/tsfile. Currently we only support spark version 2.4.3 and there are some known issue on 2.4.7, do no use it

Quick Start

Local Mode

Start Spark with TsFile-Spark-Connector in local mode:

./<spark-shell-path>  --jars  tsfile-spark-connector.jar,tsfile-{version}-jar-with-dependencies.jar,hadoop-tsfile-{version}-jar-with-dependencies.jar

Note:

<spark-shell-path> is the real path of your spark-shell.
Multiple jar packages are separated by commas without any spaces.
See https://github.com/apache/iotdb/tree/master/tsfile for how to get TsFile.

Distributed Mode

Start Spark with TsFile-Spark-Connector in distributed mode (That is, the spark cluster is connected by spark-shell):

. /<spark-shell-path>   --jars  tsfile-spark-connector.jar,tsfile-{version}-jar-with-dependencies.jar,hadoop-tsfile-{version}-jar-with-dependencies.jar  --master spark://ip:7077

Note:

<spark-shell-path> is the real path of your spark-shell.
Multiple jar packages are separated by commas without any spaces.
See https://github.com/apache/iotdb/tree/master/tsfile for how to get TsFile.

Data Type Correspondence

TsFile data type	SparkSQL data type
BOOLEAN	BooleanType
INT32	IntegerType
INT64	LongType
FLOAT	FloatType
DOUBLE	DoubleType
TEXT	StringType

The way to display TsFile is dependent on the schema. Take the following TsFile structure as an example: There are three measurements in the TsFile schema: status, temperature, and hardware. The basic information of these three measurements is listed:

Name	Type	Encode
status	Boolean	PLAIN
temperature	Float	RLE
hardware	Text	PLAIN

The existing data in the TsFile are:

The corresponding SparkSQL table is:

time	root.ln.wf02.wt02.temperature	root.ln.wf02.wt02.status	root.ln.wf02.wt02.hardware	root.ln.wf01.wt01.temperature	root.ln.wf01.wt01.status	root.ln.wf01.wt01.hardware
1	null	true	null	2.2	true	null
2	null	false	aaa	2.2	null	null
3	null	null	null	2.1	true	null
4	null	true	bbb	null	null	null
5	null	null	null	null	false	null
6	null	null	ccc	null	null	null

You can also use narrow table form which as follows: (You can see part 6 about how to use narrow form)

time	device_name	status	hardware	temperature
1	root.ln.wf02.wt01	true	null	2.2
1	root.ln.wf02.wt02	true	null	null
2	root.ln.wf02.wt01	null	null	2.2
2	root.ln.wf02.wt02	false	aaa	null
3	root.ln.wf02.wt01	true	null	2.1
4	root.ln.wf02.wt02	true	bbb	null
5	root.ln.wf02.wt01	false	null	null
6	root.ln.wf02.wt02	null	ccc	null

Scala API

NOTE: Remember to assign necessary read and write permissions in advance.

Example 1: read from the local file system

import org.apache.iotdb.spark.tsfile._
val wide_df = spark.read.tsfile("test.tsfile")  
wide_df.show

val narrow_df = spark.read.tsfile("test.tsfile", true)  
narrow_df.show

Example 2: read from the hadoop file system

import org.apache.iotdb.spark.tsfile._
val wide_df = spark.read.tsfile("hdfs://localhost:9000/test.tsfile") 
wide_df.show

val narrow_df = spark.read.tsfile("hdfs://localhost:9000/test.tsfile", true)  
narrow_df.show

Example 3: read from a specific directory

import org.apache.iotdb.spark.tsfile._
val df = spark.read.tsfile("hdfs://localhost:9000/usr/hadoop") 
df.show

Note 1: Global time ordering of all TsFiles in a directory is not supported now.

Note 2: Measurements of the same name should have the same schema.

Example 4: query in wide form

import org.apache.iotdb.spark.tsfile._
val df = spark.read.tsfile("hdfs://localhost:9000/test.tsfile") 
df.createOrReplaceTempView("tsfile_table")
val newDf = spark.sql("select * from tsfile_table where `device_1.sensor_1`>0 and `device_1.sensor_2` < 22")
newDf.show

import org.apache.iotdb.spark.tsfile._
val df = spark.read.tsfile("hdfs://localhost:9000/test.tsfile") 
df.createOrReplaceTempView("tsfile_table")
val newDf = spark.sql("select count(*) from tsfile_table")
newDf.show

Example 5: query in narrow form

import org.apache.iotdb.spark.tsfile._
val df = spark.read.tsfile("hdfs://localhost:9000/test.tsfile", true) 
df.createOrReplaceTempView("tsfile_table")
val newDf = spark.sql("select * from tsfile_table where device_name = 'root.ln.wf02.wt02' and temperature > 5")
newDf.show

import org.apache.iotdb.spark.tsfile._
val df = spark.read.tsfile("hdfs://localhost:9000/test.tsfile", true) 
df.createOrReplaceTempView("tsfile_table")
val newDf = spark.sql("select count(*) from tsfile_table")
newDf.show

Example 6: write in wide form

// we only support wide_form table to write
import org.apache.iotdb.spark.tsfile._

val df = spark.read.tsfile("hdfs://localhost:9000/test.tsfile") 
df.show
df.write.tsfile("hdfs://localhost:9000/output")

val newDf = spark.read.tsfile("hdfs://localhost:9000/output")
newDf.show

Example 7: write in narrow form

// we only support wide_form table to write
import org.apache.iotdb.spark.tsfile._

val df = spark.read.tsfile("hdfs://localhost:9000/test.tsfile", true) 
df.show
df.write.tsfile("hdfs://localhost:9000/output", true)

val newDf = spark.read.tsfile("hdfs://localhost:9000/output", true)
newDf.show

Appendix A: Old Design of Schema Inference

The way to display TsFile is related to TsFile Schema. Take the following TsFile structure as an example: There are three measurements in the Schema of TsFile: status, temperature, and hardware. The basic info of these three Measurements is:

Name	Type	Encode
status	Boolean	PLAIN
temperature	Float	RLE
hardware	Text	PLAIN

The existing data in the file are:

A set of time-series data

There are two ways to show a set of time-series data:

the default way

Two columns are created to store the full path of the device: time(LongType) and delta_object(StringType).

time : Timestamp, LongType
delta_object : Delta_object ID, StringType

Next, a column is created for each Measurement to store the specific data. The SparkSQL table structure is:

time(LongType)	delta_object(StringType)	status(BooleanType)	temperature(FloatType)	hardware(StringType)
1	root.ln.wf01.wt01	True	2.2	null
1	root.ln.wf02.wt02	True	null	null
2	root.ln.wf01.wt01	null	2.2	null
2	root.ln.wf02.wt02	False	null	"aaa"
2	root.sgcc.wf03.wt01	True	null	null
3	root.ln.wf01.wt01	True	2.1	null
3	root.sgcc.wf03.wt01	True	3.3	null
4	root.ln.wf01.wt01	null	2.0	null
4	root.ln.wf02.wt02	True	null	"bbb"
4	root.sgcc.wf03.wt01	True	null	null
5	root.ln.wf01.wt01	False	null	null
5	root.ln.wf02.wt02	False	null	null
5	root.sgcc.wf03.wt01	True	null	null
6	root.ln.wf02.wt02	null	null	"ccc"
6	root.sgcc.wf03.wt01	null	6.6	null
7	root.ln.wf01.wt01	True	null	null
8	root.ln.wf02.wt02	null	null	"ddd"
8	root.sgcc.wf03.wt01	null	8.8	null
9	root.sgcc.wf03.wt01	null	9.9	null

unfold delta_object column

Expand the device column by "." into multiple columns, ignoring the root directory "root". Convenient for richer aggregation operations. To use this display way, the parameter "delta_object_name" is set in the table creation statement (refer to Example 5 in Section 5.1 of this manual), as in this example, parameter "delta_object_name" is set to "root.device.turbine". The number of path layers needs to be one-to-one. At this point, one column is created for each layer of the device path except the "root" layer. The column name is the name in the parameter and the value is the name of the corresponding layer of the device. Next, one column is created for each Measurement to store the specific data.

Then SparkSQL Table Structure is as follows:

time(LongType)	group(StringType)	field(StringType)	device(StringType)	status(BooleanType)	temperature(FloatType)	hardware(StringType)
1	ln	wf01	wt01	True	2.2	null
1	ln	wf02	wt02	True	null	null
2	ln	wf01	wt01	null	2.2	null
2	ln	wf02	wt02	False	null	"aaa"
2	sgcc	wf03	wt01	True	null	null
3	ln	wf01	wt01	True	2.1	null
3	sgcc	wf03	wt01	True	3.3	null
4	ln	wf01	wt01	null	2.0	null
4	ln	wf02	wt02	True	null	"bbb"
4	sgcc	wf03	wt01	True	null	null
5	ln	wf01	wt01	False	null	null
5	ln	wf02	wt02	False	null	null
5	sgcc	wf03	wt01	True	null	null
6	ln	wf02	wt02	null	null	"ccc"
6	sgcc	wf03	wt01	null	6.6	null
7	ln	wf01	wt01	True	null	null
8	ln	wf02	wt02	null	null	"ddd"
8	sgcc	wf03	wt01	null	8.8	null
9	sgcc	wf03	wt01	null	9.9	null

TsFile-Spark-Connector displays one or more TsFiles as a table in SparkSQL By SparkSQL. It also allows users to specify a single directory or use wildcards to match multiple directories. If there are multiple TsFiles, the union of the measurements in all TsFiles will be retained in the table, and the measurement with the same name have the same data type by default. Note that if a situation with the same name but different data types exists, TsFile-Spark-Connector does not guarantee the correctness of the results.

The writing process is to write a DataFrame as one or more TsFiles. By default, two columns need to be included: time and delta_object. The rest of the columns are used as Measurement. If user wants to write the second table structure back to TsFile, user can set the "delta_object_name" parameter(refer to Section 5.1 of Section 5.1 of this manual).

Appendix B: Old Note NOTE: Check the jar packages in the root directory of your Spark and replace libthrift-0.9.2.jar and libfb303-0.9.2.jar with libthrift-0.9.1.jar and libfb303-0.9.1.jar respectively.