如何从Docker容器访问CSV输出？

Question

Python程序生成CSV输出。目前，能够在主机上运行test.py并生成示例_output.csv

然而，当通过Docker容器实现该程序时，很难找到sample_output.csv文件。下面是Dockerfile和requirements.txt文件

numpy==1.19.4
pandas==1.2.0
python-dateutil==2.8.1
pytz==2020.5
scipy==1.5.4
six==1.15.0     // -> requirements.txt



FROM python:3

WORKDIR /demo

COPY requirements.txt ./

RUN pip install --no-cache-dir -r requirements.txt

COPY . .

CMD ["python" , "-u" ,"./test.py"]   // -> Dockerfile

Docker映像可以通过运行Docker build-t imagename生成。但是，在运行docker run imagename时，不会生成csv文件

基于Docker映像运行Docker容器后，希望在查找示例_output.csv文件方面寻求帮助

//test.py

import pandas as pd
import numpy as np
import os 
from scipy.stats import uniform, exponweib
from scipy.special import gamma
from scipy.optimize import curve_fit


N_SUBSYS = 30
STEPS = 100
LIMIT = 101*STEPS
times = np.arange(0, LIMIT, STEPS)

if not os.path.exists('./output'):
    os.mkdir('./output')

    print("Directory")

class WeibullFailure():
    def __init__(self):
        N_TRAINS = 92
        LOWER_BETA = 0.9
        RANGE_BETA = 0.3
        LOWER_LOGSCALE = 4
        RANGE_LOGSCALE = 1.5
        LOWER_SIZE = 4
        RANGE_SIZE = 8
        gensize = N_TRAINS * int(uniform.rvs(LOWER_SIZE, RANGE_SIZE))
        genbeta = uniform.rvs(LOWER_BETA, RANGE_BETA)
        genscale = np.power(10, uniform.rvs(LOWER_LOGSCALE, RANGE_LOGSCALE))
        self.beta = genbeta
        self.eta = genscale
        self.size = gensize

    def generate_failures(self):
        return exponweib.rvs(
            a=1, loc=0, c=self.beta, scale=self.eta, size=self.size
        )

    def __repr__(self):
        string = f"Subsystem ~ ({self.size} Instances)"
        string += f" Weibull({self.eta:.2f}, {self.beta:.4f})"
        return string


def get_cumulative_failures(failure_times, times):
    cumulative_failures = {
        i: np.histogram(ft, times)[0].cumsum()
        for i, ft in failure_times.items()
    }
    cumulative_failures = pd.DataFrame(cumulative_failures, index=times[1:])
    return cumulative_failures


def fit_failures(cumulative_failures, subsystems):
    fitted = {}
    for i, x in cumulative_failures.items():
        size = subsystems[i].size
        popt, _ = curve_fit(
            lambda x, a, b: np.exp(a)*np.power(x, b), x.index, x.values
        )
        fitted[i] = (np.exp(-popt[0]/popt[1])*size, popt[1])
    return fitted


def kl_divergence(p1, p2):
    em_constant = 0.57721  # Euler-Mascheroni constant
    eta1, beta1 = p1
    eta2, beta2 = p2
    e11 = np.log(beta1/np.power(eta1, beta1))
    e12 = np.log(beta2/np.power(eta2, beta2))
    e2 = (beta1 - beta2)*(np.log(eta1) - em_constant/beta1)
    e3 = np.power(eta1/eta2, beta2)*gamma(beta2/beta1 + 1) - 1
    divergence = e11 - e12 + e2 + e3
    return divergence


subsystems = {i: WeibullFailure() for i in range(N_SUBSYS)}
failure_times = {i: s.generate_failures() for i, s in subsystems.items()}

cumulative_failures = get_cumulative_failures(failure_times, times)
fitted = fit_failures(cumulative_failures, subsystems)
divergences = {
    i: kl_divergence(f, [subsystems[i].eta, subsystems[i].beta])
    for i, f in fitted.items()
}

expected_failures = {i: np.power(times[1:]/s.eta, s.beta)*s.size
                     for i, s in subsystems.items()}
expected_failures = pd.DataFrame(expected_failures, index=times[1:])

modeled_failures = {i: np.power(times[1:]/f[0], f[1])*subsystems[i].size
                    for i, f in fitted.items()}
modeled_failures = pd.DataFrame(modeled_failures, index=times[1:])

cols = ['eta', 'fit_eta', 'beta', 'fit_beta', 'kl_divergence', 'n_instance']
out = pd.concat([
    pd.DataFrame({i: [s.size, s.eta, s.beta] for i, s in subsystems.items()},
                 index=['n_instance', 'eta', 'beta']).T,
    pd.DataFrame(fitted, index=['fit_eta', 'fit_beta']).T,
    pd.Series(divergences, name='kl_divergence')
], axis=1)[cols]
out.to_csv('./output/sample_output.csv')

if not os.path.exists('./output/sample_output.csv'):
    print("Hello")

Answer 1

我建议您将输出重定向到控制台，而不是像这样的文件：

from io import StringIO
output = StringIO()
out.to_csv(output)
print(output.getvalue())

然后你就可以拿到你的集装箱了

docker run <container> > output.csv

Answer 2

Docker容器根据定义与主机隔离。当您在容器中运行某个对象时，它会停留在容器中

您可以将主机目录装载到您认为应该显示脚本输出的容器中。您可以使用-v（卷）选项执行此操作：

docker run -v /host/path:/container/path ...

可以指定多个卷：

docker run -v /host/path:/container/path -v /another/host/path:/another/container/path ...

之后，包含所有内容的主机目录将显示在容器中，就像它在主机上一样，如果您的程序将在其中添加或替换某些内容，您将看到它

UPD：查看test.py您的输出文件应该在/demo/output中，因此您可以在那里挂载一些主机目录，例如，您的当前目录：docker run -v $(pwd):/demo/output ...