python分布式 集群_python dask 搭建分布式集群

1、分布式版本安装步骤

1.conda安装：conda install dask distributed-cconda-forge

2.pip 安装：pip install dask distributed --upgrade

3.source安装：

git clone https://github.com/dask/distributed.git

cd distributed

python setup.py install

2、主节点启动方法

dask-scheduler

控制台显示信息如下：

distributed.scheduler - INFO - -----------------------------------------------

distributed.scheduler - INFO - Clear task state

distributed.scheduler - INFO - Scheduler at: tcp://192.168.1.42:8786

distributed.scheduler - INFO - :8787

distributed.scheduler - INFO - Local Directory: C:\Users\User\AppData\Local\Temp\scheduler-gd9uk980

distributed.scheduler - INFO - -----------------------------------------------

3、工作节点启动方法

dask-worker 192.168.1.42:8786

工作节点启动成功后，此时主节点会显示多出信息：

distributed.scheduler - INFO - Register tcp://192.168.1.184:45772

distributed.scheduler - INFO - Starting worker compute stream, tcp://192.168.1.184:45772

distributed.core - INFO - Starting established connection

distributed.scheduler - INFO - Register tcp://192.168.1.183:43405

distributed.scheduler - INFO - Starting worker compute stream, tcp://192.168.1.183:43405

distributed.core - INFO - Starting established connection

distributed.scheduler - INFO - Register tcp://192.168.1.188:38095

distributed.scheduler - INFO - Starting worker compute stream, tcp://192.168.1.188:38095

distributed.core - INFO - Starting established connection

4、官方测试代码

"""分布式dask"""

import time

from dask.distributed import Client

client = Client('192.168.1.42:8786' ,asynchronous=True)

ts = time.time()

A = client.map(square, range(10000))

B = client.map(neg, A)

total = client.submit(sum, B)

print(total.result())

print('cost time :%s'%(time.time()-ts))

cost time :3.793848991394043

5、参考链接

优势：dask内部自动实现了分布式调度、无需用户自行编写复杂的调度逻辑和程序；通过调用简单的方法就可以进行分布式计算、并支持部分模型的并行化处理；内部实现的分布式算法：xgboost、LR、sklearn的部分方法等

用一句话说：dask就是python版本的spark,是一个用Python 语言实现的分布式计算框架

作者：宇智波鼬_adb8

链接：https://www.jianshu.com/p/8ca5d70e0810?utm_campaign=haruki

来源：简书

著作权归作者所有。商业转载请联系作者获得授权，非商业转载请注明出处。