连接:http://spark.apache.org/docs/latest/programming-guide.htmlhtml
安装好Spark 后,自带了一些demo, 路径在Spark根目录/examples/src/main/python/python
里面有些例子,例如统计字数的 wordcount.pyapache
import sys from operator import add from pyspark import SparkContext import sys reload(sys) sys.setdefaultencoding("utf-8") if __name__ == "__main__": if len(sys.argv) != 2: print >> sys.stderr, "Usage: wordcount <file>" exit(-1) sc = SparkContext(appName="PythonWordCount") lines = sc.textFile(sys.argv[1], 1) counts = lines.flatMap(lambda x: x.split(' ')) \ .map(lambda x: (x, 1)) \ .reduceByKey(add) output = counts.collect() for (word, count) in output: print "%s: %i" % (word, count) sc.stop()
另外参考Spark的python api: http://spark.apache.org/docs/latest/api/python/pyspark.html api
写了一个小demo,就是练习一下api的使用,作业务很方便。针对于大数据文件作统计分析的。好比几十兆上百兆的咱们单机处理,上G的就放在hadoop 的 hdfs上。缓存
下面是一个学生成绩单。四列字段:学生,以及三科成绩。其中学生有重复的(好比额外加分的状况,须要合并分析)。架构
yang 85 90 30 wang 20 60 50 zhang 90 90 90 li 100 54 0 yanf 0 0 0 yang 12 0 0
固然实际中数据要不少,好比不少列,并且几十万行甚至几百万行。这里是一个demo ,至关于在部署前测试。app
在 Spark根目录/example/src/main/python/ 下新建一个 students.py :框架
#coding=utf-8 import sys from operator import add from pyspark import SparkContext import sys reload(sys) sys.setdefaultencoding("utf-8") def map_func(x): s = x.split() return (s[0],[int(s[1]),int(s[2]),int(s[3])]) def f(x): return x rank = sc.parallelize(range(0,sorted.count())) def add(a,b): return [a[r]+ b[r] for r in range(len(a))] def _merge(a,b): print '****' return [a[r]+ b[r] for r in range(len(a))] #the students who has one score is 100 def has100(x): for y in x: if(y==100): return True return False def allIs0(x): if(type(x) == list and sum(x) == 0): return True return False def subMax(x,y): m = [x[1][i] if(x[1][i] > y[1][i]) else y[1][i] for i in range(3)] return('',m) def sumAll(x,y): return ('',[x[1][i]+y[1][i] for i in range(3)]) if __name__ == "__main__": if len(sys.argv) != 2: print >> sys.stderr, "Usage: students <file>" exit(-1) sc = SparkContext(appName="Students") # 加载学生文件,调用map将学生映射成keyValues.其中,key是学生,Value是学生成绩。 # map后的结果如('yang',(85,90,30)) # 以后调用 CombineByKey,将相同窗生的成绩相加(合并)。 # 而后调用cache, 将整个数据缓存,以便屡次进行reduce而无需每次都从新生成。 lines = sc.textFile(sys.argv[1], 1).map(map_func).combineByKey(f,add,_merge).cache() #print lines count = lines.count() # 获取学生中三科成绩有满分的,调用filter来实现 whohas100 = lines.filter(lambda x: filter(has100,x)).collect() # 获取三科中全部成绩都是0的同窗(缺考) whoIs0 = lines.filter(lambda x: filter(allIs0,x)).collect() # 获取每一个学生的成绩总和 sumScore = lines.map(lambda x: (x[0],sum(x[1]))).collect() # 获取三科中,单科最高分 subM = lines.reduce(subMax) # 获取学生单科成绩的总和,求单科平均分用 sumA = lines.reduce(sumAll) # 总分最高的学生 maxScore = max(sumScore,key = lambda x: x[1]) # 总分最低的学生 minScore = min(sumScore,key = lambda x: x[1]) # 全部学生三科成绩平均分 avgA = [x/count for x in sumA[1]] # 根据总分进行排序(默认由小而大) sorted = lines.sortBy(lambda x: sum(x[1])) # 排序并附带序号 sortedWithRank = sorted.zipWithIndex().collect() # 取出成绩最高的前三名同窗,发奖! first3 = sorted.takeOrdered(3,key = lambda x: -sum(x[1])) #print '*'*50 print whohas100 print maxScore print whoIs0 print subM print avgA print sorted.collect() print sortedWithRank print first3 #将结果汇总输出到文件 file = open('/home/yanggaofei/downloads/result.txt','w') file.write('students num:'+`count`+ '\n') file.write('who has a 100 scores:' + str(whohas100) + '\n') file.write('who all is 0:' + str(whoIs0) + '\n') file.write('the max score of each subject:' + str(subM) + '\n') file.write('the avg score of each subject:' + str(avgA) + '\n') file.write('sorted the students:' + str(sorted.collect()) + '\n') file.write('sorted the students with the rank:' + str(sortedWithRank) + '\n') file.write('the first 3 who will get the award:' + str(first3) + '\n') file.close()
好了,运行:ide
[root@cyouemt spark-1.1.1] # ./bin/spark-submit examples/src/main/python/students.py temp/student.txt
运行结果result.txt以下:oop
students num:5 who has a 100 scores:[(u'li', [100, 54, 0])] who all is 0:[(u'yanf', [0, 0, 0])] the max score of each subject:('', [100, 90, 90]) the avg score of each subject:[61, 58, 34] sorted the students:[(u'yanf', [0, 0, 0]), (u'wang', [20, 60, 50]), (u'li', [100, 54, 0]), (u'yang', [97, 90, 30]), (u'zhang', [90, 90, 90])] sorted the students with the rank:[ ((u'yanf', [0, 0, 0]), 0), ((u'wang', [20, 60, 50]), 1), ((u'li', [100, 54, 0]), 2), ((u'yang', [97, 90, 30]), 3), ((u'zhang', [90, 90, 90]), 4)] the first 3 who will get the award:[ (u'zhang', [90, 90, 90]), (u'yang', [97, 90, 30]), (u'li', [100, 54, 0])]
Spark的运行过程会打印出任务执行的开始过程以及结束。表示没研究透,不作陈述。。。
相比hadoop,Spark 是一个内存计算的MapReduce, 经过缓存机制,在性能上要好不少。它自身不带数据系统。可是支持 hdfs,mesos,hbase。文本文件等。从架构和应用角度上看,spark是 一个仅包含计算逻辑的开发库(尽管它提供个独立运行的master/slave服务,但考虑到稳定后以及与其余类型做业的继承性,一般不会被采用),而不 包含任何资源管理和调度相关的实现,这使得spark能够灵活运行在目前比较主流的资源管理系统上,典型的表明是mesos和yarn,咱们称之为 “spark on mesos”和“spark on yarn”。将spark运行在资源管理系统上将带来很是多的收益,包括:与其余计算框架共享集群资源;资源按需分配,进而提升集群资源利用率等