使用spark

更新说明

• 免密码登录

for f in  `cat ~/machines`; do
    scp .ssh/id_dsa.pub  $f:~/
    ssh $f  "cat id_dsa.pub >> .ssh/authorized_keys"
done

• 安装

for f in `cat machines `; do
  scp ~/software/soft/spark-2.1.0-bin-hadoop2.7.tgz  $f:~/ ;   
  ssh -t $f "
     sudo mkdir /opt/spark
     cd /opt/spark/;
     sudo tar -zxvf  /home/keke.zhaokk/spark-2.1.0-bin-hadoop2.7.tgz ; 
     sudo ln -s spark-2.1.0-bin-hadoop2.7 current;
     sudo chown -R keke.zhaokk /opt/spark/*;
 "
done

• 修改配置
  注意 SPARK_MASTER_IP=10.11.143.30 需要改为 SPARK_MASTER_HOST=10.11.143.30

• 复制配置

for f in `cat machines `; do
   scp -r $SPARK_HOME/conf/   $f:~/
   ssh  $f "
      mv conf/slaves $SPARK_HOME/conf/
      mv conf/*.sh $SPARK_HOME/conf/
   "
done

---

spark-集群搭建

• 机器

角色	IP	安装软件
worker	10.11.143.24	python(Anaconda3 4.0.0), R, spark-2.X-bin-hadoopX.X
worker	10.11.143.26	python(Anaconda3 4.0.0), R, spark-2.X-bin-hadoopX.X
master	10.11.143.30	python(Anaconda3 4.0.0), R, spark-2.X-bin-hadoopX.X

• 查看使用情况 http://10.11.143.30:8080/
• 集群通过 start-all.sh,stop-all.sh来开启和停止.
• 在脚本使用

spark-shell --total-executor-cores 30  --executor-memory 50g  
 --master spark://10.11.143.30:7077 
# 或者
pyspark --total-executor-cores 10  --executor-memory 50g 
 --master spark://10.11.143.30:7077 

• 在Rstudio中使用

## 配置
library(SparkR)
sc = sparkR.init(
  master = "spark://10.11.143.30:7077",
  sparkEnvir = list(spark.executor.memory="100g",spark.cores.max="20")
  )
sqlContext = sparkRSQL.init(sc)
## 处理数据

### 本地数据
read.df(sqlContext, 
  paste0(Sys.getenv("SPARK_HOME"),
    "/examples/src/main/resources/people.json" ),
  "json") -> people
head(people)

### 数据
createDataFrame( sqlContext, data_dataframe) -> data_spark
head(data_spark)
printSchema(data_spark)

### SQL

registerTempTable(people,"people")
teenagers = sql(sqlContext, "select name from people where age>13 and age<=19")

### 转为data.frame
as.data.frame(teenagers)  -> T

## stop 
sparkR.stop()

• 在jupyter中使用, 见最后
• 调用例子, 在上面任意Linux机器上

$SPARK_HOME/bin/spark-submit --class  org.apache.spark.examples.SparkLR --master spark://10.11.143.30:7077 
$SPARK_HOME/examples/jars/spark-examples_2.11-2.0.0-preview.jar    > out.log

安装说明

• 解压spark-2.0.0-preview-bin-hadoop2.7.tgz到$SPARK_HOME目录下
• 配置master机器, 使得root账号能无密码登录worker机器
• master机器按照如下配置, 然后复制到worker机器上.

$sed -r '/^#/d;/^$/d' spark-env.sh
SPARK_MASTER_IP=10.11.143.30
#SPARK_LOCAL_IP=10.11.143.26  # 26这台机器出现错误, 需要加上这句.

[keke.zhaokk@r73f16622.zui /opt/spark/spark-2.0.0-preview-bin-hadoop2.7/conf]
$sed -r '/^#/d;/^$/d' spark-defaults.conf
spark.master                     spark://10.11.143.30:7077
spark.executor.memory            100g
spark.cores.max                  30

[keke.zhaokk@r73f16622.zui /opt/spark/spark-2.0.0-preview-bin-hadoop2.7/conf]
$sed -r '/^#/d;/^$/d' slaves
10.11.143.24
10.11.143.26
10.11.143.30

$tail -6 /etc/bashrc  

export JAVA_HOME=/opt/taobao/install/ajdk-8.0.0-b60
export PATH=$PATH:$JAVA_HOME/bin                     # worker 机器不需要这句
export SPARK_HOME=/opt/spark/spark-2.0.0-preview-bin-hadoop2.7
export PATH=$PATH:$SPARK_HOME/bin:$SPARK_HOME/sbin   # worker 机器不需要这句

Windows下使用spark(单机版)

• 将winutils.exe 放到%HADOOP_HOME%in 目录下
• 解压 spark-2.0.0-preview-bin-hadoop2.7.tgz 到 %SPARK_HOME%
• 将%SPARK_HOME%in; %SPARK_HOME%sbin; 添加到环境变量PATH中

在jupyter中使用

配置环境变量

import os,sys
os.environ['SPARK_HOME'] = '/opt/spark/spark-2.0.0-preview-bin-hadoop2.7'
os.environ['JAVA_HOME']='/opt/taobao/install/ajdk-8.0.0-b60'
os.environ['PYSPARK_SUBMIT_ARGS']='''
--total-executor-cores 10  --executor-memory 50g  
--master spark://10.11.143.30:7077 pyspark-shell
'''

调用pyspark

import os,sys
spark_home = os.environ.get('SPARK_HOME', None)

if any(map(lambda x: spark_home in x, sys.path)) is False:
    sys.path.insert(0, os.path.join(spark_home ,"python"))
    sys.path.insert(0, os.path.join(spark_home, 'python/lib/py4j-0.10.1-src.zip'))

with open(os.path.join(spark_home, "python/pyspark/shell.py")) as f:
    code = compile(f.read(), "shell.py", 'exec')
    exec(code)

text_file = sc.textFile(spark_home + "/README.md")
word_counts = text_file 
    .flatMap(lambda line: line.split()) 
    .map(lambda word: (word, 1)) 
    .reduceByKey(lambda a, b: a + b)

len(word_counts.collect())

Welcome to
      ____              __
     / __/__  ___ _____/ /__
    _ / _ / _ `/ __/  '_/
   /__ / .__/\_,_/_/ /_/\_   version 2.0.0-preview
      /_/

Using Python version 2.7.8 (default, Nov 27 2014 17:41:17)
SparkSession available as 'spark'.

263

停止调用

sc.stop()