如何阻止Dask LocalCluster工作进程的打印内容输出到控制台？

我有一个运行时会打印内容的Python函数，希望通过Dask运行它，但不希望工作进程的打印内容直接输出到控制台。简化示例代码如下：

def dask_function(i):
    print('Ignore me!')
    return i**2

cluster = LocalCluster(12, processes=True)
client = Client(cluster)

dask_delays = []
for i in range(10):
    dask_delays.append(dask.delayed(dask_function)(i))

dask_outs = client.compute(dask_delays)

运行这段代码后，控制台会打印10次“Ignore me!”，请问有什么方法可以阻止这种输出？

方法1：重定向工作进程的标准输出/错误流 #

创建LocalCluster时，通过worker_kwargs参数将工作进程的stdout和stderr重定向到空设备（Linux/macOS用/dev/null，Windows用NUL），彻底丢弃打印内容：

import os
from dask.distributed import LocalCluster, Client
import dask

def dask_function(i):
    print('Ignore me!')
    return i**2

# 根据系统选择空设备路径
null_device = '/dev/null' if os.name != 'nt' else 'NUL'

cluster = LocalCluster(
    n_workers=12,
    processes=True,
    worker_kwargs={
        'stdout': open(null_device, 'w'),
        'stderr': open(null_device, 'w')
    }
)
client = Client(cluster)

dask_delays = [dask.delayed(dask_function)(i) for i in range(10)]
dask_outs = client.compute(dask_delays)

方法2：在函数层面抑制打印 #

如果能修改目标函数，直接移除print语句是最直接的方式。若不想改动原函数，可使用装饰器临时重定向函数内的输出：

import contextlib
import io
from dask.distributed import LocalCluster, Client
import dask

def suppress_print(func):
    def wrapper(*args, **kwargs):
        # 将输出重定向到内存缓冲区，不输出到控制台
        with contextlib.redirect_stdout(io.StringIO()), contextlib.redirect_stderr(io.StringIO()):
            return func(*args, **kwargs)
    return wrapper

@suppress_print
def dask_function(i):
    print('Ignore me!')
    return i**2

cluster = LocalCluster(12, processes=True)
client = Client(cluster)

dask_delays = [dask.delayed(dask_function)(i) for i in range(10)]
dask_outs = client.compute(dask_delays)

方法3：通过Dask日志配置过滤输出 #

Dask工作进程的print内容会被日志系统捕获，调整工作进程的日志级别为WARNING及以上，即可隐藏INFO级别的打印内容：

import logging
from dask.distributed import LocalCluster, Client
import dask

# 设置Dask工作进程日志级别，忽略INFO及以下的输出
logging.getLogger('distributed.worker').setLevel(logging.WARNING)

def dask_function(i):
    print('Ignore me!')
    return i**2

cluster = LocalCluster(12, processes=True)
client = Client(cluster)

dask_delays = [dask.delayed(dask_function)(i) for i in range(10)]
dask_outs = client.compute(dask_delays)

内容的提问来源于stack exchange，提问作者Tom Warner