1

I am trying to launch a Sparkling Water cloud within Spark using Databricks. I've attached the H2O library (3.16.0.2), PySparkling (pysparkling 0.4.6), and the Sparkling Water jar (sparkling-water-assembly_2.11-2.1.10-all.jar) to the cluster I'm running (Spark 2.1, Auto-updating Scala 1.1.1).

I succesfully import the required libraries below: 

from pysparkling import *
import h2o

Yet when I try to initialize the Sparkling Water cloud using the following commands:

hc = H2OContext.getOrCreate(spark)

or

H2OContext.getOrCreate(sc)

I get the same error:

NameError: name 'H2OContext' is not defined

NameError                                 Traceback (most recent call last)
<command-4043510449425708> in <module>()
----> 1 H2OContext.getOrCreate(sc)

NameError: name 'H2OContext' is not defined

For what it's worth I can initialize the Sparkling Water cloud using this Scala documentation: 

%scala
import org.apache.spark.h2o._
val h2oConf = new H2OConf(sc).set("spark.ui.enabled", "false")
val h2oContext = H2OContext.getOrCreate(sc, h2oConf)

import org.apache.spark.h2o._
h2oConf: org.apache.spark.h2o.H2OConf =
Sparkling Water configuration:
  backend cluster mode : internal
  workers              : None
  cloudName            : sparkling-water-root_app-20171222131625-0000
  flatfile             : true
  clientBasePort       : 54321
  nodeBasePort         : 54321
  cloudTimeout         : 60000
  h2oNodeLog           : INFO
  h2oClientLog         : WARN
  nthreads             : -1
  drddMulFactor        : 10
h2oContext: org.apache.spark.h2o.H2OContext =

Sparkling Water Context:
 * H2O name: sparkling-water-root_app-20171222131625-0000
 * cluster size: 1
 * list of used nodes:
  (executorId, host, port)
  ------------------------
  (x,xx.xxx.xxx.x,54321)
  ------------------------

  Open H2O Flow in browser: http://xx.xxx.xxx.xxx:54321 (CMD + click in Mac OSX)

but this pipeline may not always use Databricks so it needs to be all in PySpark and Databricks doesn't have a corresponding PySpark example.

Thanks in advance.

4

1 回答 1

2

对于 pysparkling，您需要先创建一个 PyPi 库，h2o_pysparkling_2.1因为您使用的是 Spark 2.1 集群。你附加的图书馆pysparkling是不同的。此外，您不需要附加所有其他库，因为该h2o_pysparkling_2.1包已经导入了其他必要的库。

一旦你这样做，你可以运行：

from pysparkling import *

h2oConf = H2OConf(spark)
h2oConf.set("spark.ui.enabled", False)

h2oContext = H2OContext.getOrCreate(spark, h2oConf)
于 2017-12-22T14:43:52.020 回答