Question

我刚接触到spark 2.0;到目前为止，我一直在使用spark 1.6.1。有人可以帮我用pyspark（python）设置sparkSession吗？我知道在线提供的scala示例类似（here），但我希望能直接使用python语言。

我的具体案例：我在zeppelin spark笔记本中加载来自S3的avro文件。然后建立df＆s并运行各种pyspark＆amp; sql查询它们。我所有的旧查询都使用sqlContext。我知道这是不好的做法，但我用

开始了我的笔记本

sqlContext = SparkSession.builder.enableHiveSupport().getOrCreate()。

我可以用avros阅读

mydata = sqlContext.read.format("com.databricks.spark.avro").load("s3:...

并构建没有问题的数据帧。但是一旦我开始查询数据帧/临时表，我就会继续得到＆＃34; java.lang.NullPointerException＆＃34;错误。我认为这表明存在转换错误（例如，旧的查询在1.6.1中工作但需要针对2.0进行调整）。无论查询类型如何，都会发生错误。所以我假设

1。）sqlContext别名是个坏主意

和

2.）我需要正确设置sparkSession。

因此，如果有人能告诉我这是如何完成的，或者可能解释他们所知道的不同版本的火花之间的差异，我将非常感激。如果我需要详细说明这个问题，请告诉我。如果它令人费解，我道歉。

Answer 1

from pyspark.sql import SparkSession
spark = SparkSession.builder.appName('abc').getOrCreate()

现在要导入一些你可以使用的.csv文件

df=spark.read.csv('filename.csv',header=True)

Answer 2

从这里http://spark.apache.org/docs/2.0.0/api/python/pyspark.sql.html
您可以使用以下方法创建一个spark会话：

>>> from pyspark.conf import SparkConf
>>> SparkSession.builder.config(conf=SparkConf())

Answer 3

正如您在scala示例中所看到的，Spark Session是sql模块的一部分。类似于python。因此，请参阅pyspark sql module documentation

class pyspark.sql.SparkSession（sparkContext，jsparkSession = None）使用Dataset和DataFrame API编程Spark的切入点。一个 SparkSession可用于创建DataFrame，将DataFrame注册为表，在表上执行SQL，缓存表和读取镶木地板文件。要创建SparkSession，请使用以下构建器模式：

>>> spark = SparkSession.builder \
...     .master("local") \
...     .appName("Word Count") \
...     .config("spark.some.config.option", "some-value") \
...     .getOrCreate()

Answer 4

spark  = SparkSession.builder\
                  .master("local")\
                  .enableHiveSupport()\
                  .getOrCreate()

spark.conf.set("spark.executor.memory", '8g')
spark.conf.set('spark.executor.cores', '3')
spark.conf.set('spark.cores.max', '3')
spark.conf.set("spark.driver.memory",'8g')
sc = spark.sparkContext

Answer 5

这是我开发的一个有用的Python SparkSession类：

#!/bin/python
# -*- coding: utf-8 -*-

######################
# SparkSession class #
######################
class SparkSession:

    # - Notes:
    # The main object if Spark Context ('sc' object).
    # All new Spark sessions ('spark' objects) are sharing the same underlying Spark context ('sc' object) into the same JVM,
    # but for each Spark context the temporary tables and registered functions are isolated.
    # You can't create a new Spark Context into another JVM by using 'sc = SparkContext(conf)',
    # but it's possible to create several Spark Contexts into the same JVM by specifying 'spark.driver.allowMultipleContexts' to true (not recommended).
    # - See:
    # https://medium.com/@achilleus/spark-session-10d0d66d1d24
    # https://stackoverflow.com/questions/47723761/how-many-sparksessions-can-a-single-application-have
    # https://stackoverflow.com/questions/34879414/multiple-sparkcontext-detected-in-the-same-jvm
    # https://stackoverflow.com/questions/39780792/how-to-build-a-sparksession-in-spark-2-0-using-pyspark
    # https://stackoverflow.com/questions/47813646/sparkcontext-getorcreate-purpose?noredirect=1&lq=1

    from pyspark.sql import SparkSession

    spark = None   # The Spark Session
    sc = None      # The Spark Context
    scConf = None  # The Spark Context conf

    def _init(self):
        self.sc = self.spark.sparkContext
        self.scConf = self.sc.getConf() # or self.scConf = self.spark.sparkContext._conf

    # Return the current Spark Session (singleton), otherwise create a new oneÒ
    def getOrCreateSparkSession(self, master=None, appName=None, config=None, enableHiveSupport=False):
        cmd = "self.SparkSession.builder"
        if (master != None): cmd += ".master(" + master + ")"
        if (appName != None): cmd += ".appName(" + appName + ")"
        if (config != None): cmd += ".config(" + config + ")"
        if (enableHiveSupport == True): cmd += ".enableHiveSupport()"
        cmd += ".getOrCreate()"
        self.spark = eval(cmd)
        self._init()
        return self.spark

    # Return the current Spark Context (singleton), otherwise create a new one via getOrCreateSparkSession()
    def getOrCreateSparkContext(self, master=None, appName=None, config=None, enableHiveSupport=False):
        self.getOrCreateSparkSession(master, appName, config, enableHiveSupport)
        return self.sc 

    # Create a new Spark session from the current Spark session (with isolated SQL configurations).
    # The new Spark session is sharing the underlying SparkContext and cached data,
    # but the temporary tables and registered functions are isolated.
    def createNewSparkSession(self, currentSparkSession):
        self.spark = currentSparkSession.newSession()
        self._init()
        return self.spark

    def getSparkSession(self):
        return self.spark

    def getSparkSessionConf(self):
        return self.spark.conf

    def getSparkContext(self):
        return self.sc

    def getSparkContextConf(self):
        return self.scConf

    def getSparkContextConfAll(self):
        return self.scConf.getAll()

    def setSparkContextConfAll(self, properties):
        # Properties example: { 'spark.executor.memory' : '4g', 'spark.app.name' : 'Spark Updated Conf', 'spark.executor.cores': '4',  'spark.cores.max': '4'}
        self.scConf = self.scConf.setAll(properties) # or self.scConf = self.spark.sparkContext._conf.setAll()

    # Stop (clears) the active SparkSession for current thread.
    #def stopSparkSession(self):
    #    return self.spark.clearActiveSession()

    # Stop the underlying SparkContext.
    def stopSparkContext(self):
        self.spark.stop() # Or self.sc.stop()

    # Returns the active SparkSession for the current thread, returned by the builder.
    #def getActiveSparkSession(self):
    #    return self.spark.getActiveSession()

    # Returns the default SparkSession that is returned by the builder.
    #def getDefaultSession(self):
    #    return self.spark.getDefaultSession()

如何使用pyspark在Spark 2.0中构建sparkSession？

5 个答案: