Hadoop 2.6和2.7 Apache Terasort，500 on或1 1TB

Question

当运行一个map时，当reducers启动时，它从0到100失败，并显示以下错误：

15/05/12 07:21:27 INFO terasort.TeraSort: starting
15/05/12 07:21:27 WARN util.NativeCodeLoader: Unable to load native-hadoop     library for your platform... using builtin-java classes where applicable
15/05/12 07:21:29 INFO input.FileInputFormat: Total input paths to process :    18000

Spent 1514ms computing base-splits.
Spent 109ms computing TeraScheduler splits.
Computing input splits took 1624ms
Sampling 10 splits of 18000
Making 1 from 100000 sampled records
Computing parititions took 315ms
Spent 1941ms computing partitions.
15/05/12 07:21:30 INFO client.RMProxy: Connecting to ResourceManager at    n1/192.168.2.1:8032
15/05/12 07:21:31 INFO mapreduce.JobSubmitter: number of splits:18000
15/05/12 07:21:31 INFO mapreduce.JobSubmitter: Submitting tokens for job: job_1431389162125_0001
15/05/12 07:21:31 INFO impl.YarnClientImpl: Submitted application application_1431389162125_0001
15/05/12 07:21:31 INFO mapreduce.Job: The url to track the job: http://n1:8088/proxy/application_1431389162125_0001/
15/05/12 07:21:31 INFO mapreduce.Job: Running job: job_1431389162125_0001
15/05/12 07:21:37 INFO mapreduce.Job: Job job_1431389162125_0001 running in uber mode : false
15/05/12 07:21:37 INFO mapreduce.Job:  map 0% reduce 0%
15/05/12 07:21:47 INFO mapreduce.Job:  map 1% reduce 0%
15/05/12 07:22:01 INFO mapreduce.Job:  map 2% reduce 0%
15/05/12 07:22:13 INFO mapreduce.Job:  map 3% reduce 0%
15/05/12 07:22:25 INFO mapreduce.Job:  map 4% reduce 0%
15/05/12 07:22:38 INFO mapreduce.Job:  map 5% reduce 0%
15/05/12 07:22:50 INFO mapreduce.Job:  map 6% reduce 0%
15/05/12 07:23:02 INFO mapreduce.Job:  map 7% reduce 0%
15/05/12 07:23:15 INFO mapreduce.Job:  map 8% reduce 0%
15/05/12 07:23:27 INFO mapreduce.Job:  map 9% reduce 0%
15/05/12 07:23:40 INFO mapreduce.Job:  map 10% reduce 0%
15/05/12 07:23:52 INFO mapreduce.Job:  map 11% reduce 0%
15/05/12 07:24:02 INFO mapreduce.Job:  map 100% reduce 100%
15/05/12 07:24:06 INFO mapreduce.Job: Job job_1431389162125_0001 failed with  state FAILED due to: Task failed task_1431389162125_0001_r_000000
Job failed as tasks failed. failedMaps:0 failedReduces:1

这是默认配置，每次都会失败。

我注释掉插入到xml中的任何配置都会发现这个问题，但是我仍然有问题，只有在reducing开始的时候，作业才会失败。

Answer 1

Yarn负责资源管理，还为可以使用MapReduce的批处理工作负载和实时工作负载提供服务。

有一些内存设置可以在Yarn容器级别设置，也可以在映射器和减速器级别设置。内存是以纱线容器大小的增量请求的。Mapper和reducer任务在容器中运行。

mapreduce.map.memory.mb and mapreduce.reduce.memory.mb

以上参数描述了map-reduce任务的内存上限，如果该任务订阅的内存超过该上限，对应的容器将被终止。

这些参数决定了可以分别分配给mapper和reduce任务的最大内存量。让我们看一个例子: Mapper受配置参数mapreduce.map.memory.mb中定义的内存上限的限制。

但是，如果yarn.scheduler.minimum-allocation-mb的值大于这个mapreduce.map.memory.mb的值，那么将考虑yarn.scheduler.minimum-allocation-mb，并分配该大小的容器。

此参数需要仔细设置，如果设置不当，可能会导致性能不佳或出现OutOfMemory错误。

mapreduce.reduce.java.opts and mapreduce.map.java.opts

此属性值需要小于mapreduce.map.memory.mb/mapreduce.reduce.memory.mb，中定义的map/reduce任务的上限，因为它应该适合map/reduce任务的内存分配。