It takes a lot of time to download many kinds of software when building containers. hub.docker.com is slow in nature, especially when it encounters modules stored on gcr.io/aws and so on. pip installation of python module is also slow, and the download of conda is like a snail.

There are several ways to speed up the download of container construction:

1. Put it on the "external server" to build, and then transfer it to aliyun and other images. The download speed will be much faster.

• The steps can refer to: Create kubernetes-1.11.0 image service (high speed) in Alibaba cloud
• If the system disk is not enough, refer to: How to add data disk to Docker of container service

2. Add proxy, pip and conda images. The following is a single user image built for jupyterhub.

# Copyright (c) Jupyter Development Team.
# Distributed under the terms of the Modified BSD License.
FROM jupyter/all-spark-notebook:5811dcb711ba

LABEL maintainer="Databook Project,https://github.com/databooks<openthings@163.com>"

USER root

# ====================================================================
# Add proxy, using --build-arg "HTTP_PROXY=http://192.168.199.99:9999"

ENV HTTP_PROXY ${HTTP_PROXY}
ENV HTTPS_PROXY ${HTTP_PROXY}
ENV http_proxy ${HTTP_PROXY}
ENV https_proxy ${HTTP_PROXY}

#Add conda install mirror:

RUN echo $http_proxy && \
    conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/ && \
    conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/ && \
    conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/ && \
    conda config --set show_channel_urls yes

#Add pip install mirror:

RUN echo "[global] \
index-url = http://pypi.tuna.tsinghua/simple \
trusted-host = \
    pypi.tuna.tsinghua \
timeout = 120 \
" > /etc/pip.conf
# ====================================================================

# ====================================================================
USER $NB_UID

RUN pip install --upgrade pip 
RUN pip install bs4 && \
    pip install lxml && \
    pip install ipyleaflet && \
    pip install py4j && \
    pip install pyspark && \
    pip install mlflow && \
    pip install airflow && \
    pip install tushare

RUN conda update -n base conda
RUN conda install -y -c conda-forge nodejs=8.10.0 && \
    conda install -y -c conda-forge tensorflow=1.8.0 && \
    jupyter labextension install jupyter-leaflet

# ====================================================================
ENV HTTP_PROXY ""
ENV HTTPS_PROXY ""
ENV http_proxy ""
ENV https_proxy ""
# ====================================================================

Be careful:

• Note that it should be added to docker. It is useless to set the host, because docker build runs in a separate container.
• When building, use docker build -- build Arg "http_proxy = http://192.168.199.99:9999" to pass in the proxy parameter, and at the end, clear the http_proxy and other parameters. This avoids writing proxy information to the mirror.
• Here, we use the pip and conda images of Tsinghua University, which can be replaced by others. Refer to the following information:
  • Ubuntu configuration pip.conf add domestic source

More references: Databook data book