Skip to content

Commit

Permalink
CP to V0.1 branch: TF notebook slimming, joyvan pip installs, and new…
Browse files Browse the repository at this point in the history 
… gcr.io locations (#703)

* Refactor tensorflow-notebook-image/Dockerfile (#689)

Remove a lot of bloat - install only the minimal set of packages
required to get started with ML. Any packages required can be installed
by the user in the notebook itself using pip install/conda install

Image size has gone down from 12GB to 3GB for cpu image

Having a lot of packages makes it very challenging to maintain them
because of version conflicts

Run everything as jovyan user - this enables user to run conda install
/ pip install without requiring sudo

Add comments on every step

Fixes #668
Fixes #37
Fixes #472
Conflicts:
	components/tensorflow-notebook-image/Dockerfile
	components/tensorflow-notebook-image/build_image.sh
	components/tensorflow-notebook-image/releaser/components/workflows.libsonnet

* Update various images in kubeflow to kubeflow-images-public (#635)

Point them to kubeflow-images-public instead of kubeflow-images-staging

Related to #534
/cc @jlewi
Conflicts:
	bootstrap/Makefile
	bootstrap/README.md

* Migrate images to kubeflow-images-public (#695)

Related to #534
Conflicts:
	bootstrap/README.md
	docs_dev/images.md
	kubeflow/core/tests/tf-job_test.jsonnet

* Update the hub spawner dropdown for latest NB images (#697)
  • Loading branch information
@k8s-ci-robot
Pete MacKinnon authored and k8s-ci-robot committed Apr 21, 2018
1 parent c8788e2 commit b497fd5
Show file tree
Hide file tree
Showing 15 changed files with 157 additions and 215 deletions.
2 changes: 1 addition & 1 deletion components/k8s-model-server/images/releaser/components/params.libsonnet
View file
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,7 @@
gpu_model: {
http_proxy_image: "gcr.io/kubeflow/http-proxy:1.0",
model_path: "gs://kubeflow-ci-test-models/mnist/",
model_server_image: "gcr.io/kubeflow-images-staging/tf-model-server-gpu:v20180305-pr362-7f250ae-5cc7",
model_server_image: "gcr.io/kubeflow-images-public/tf-model-server-gpu:v20180305-pr362-7f250ae-5cc7",
name: "gpu_model",
namespace: "default",
service_type: "ClusterIP",
Expand Down
2 changes: 1 addition & 1 deletion components/k8s-model-server/images/releaser/environments/releasing/params.libsonnet
View file
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,7 @@ params {
nfsVolumeClaim: "nfs-external",
project: "kubeflow-releasing",
prow_env: "JOB_NAME=tf-serving-release,JOB_TYPE=presubmit,REPO_NAME=kubeflow,REPO_OWNER=kubeflow,BUILD_NUMBER=8cd1,PULL_NUMBER=339,PULL_PULL_SHA=72bc7b1",
registry: "gcr.io/kubeflow-images-staging",
registry: "gcr.io/kubeflow-images-public",
testing_image: "gcr.io/kubeflow-releasing/worker:v20180227-03a5f86-dirty-e3b0c4",
zone: "us-central1-a",
},
Expand Down
299 changes: 111 additions & 188 deletions components/tensorflow-notebook-image/Dockerfile
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2,226 +2,149 @@
# Distributed under the terms of the Modified BSD License.

ARG BASE_IMAGE=ubuntu:latest
# TODO(inc0): When tf-serving-api becomes available for py3, this should be
# removed
FROM python:2.7 as tf-serving-install

RUN pip install tensorflow-serving-api

FROM $BASE_IMAGE

ARG INSTALL_TFMA=yes
ARG TF_PACKAGE=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-1.7.0-cp36-cp36m-linux_x86_64.whl
ARG TF_PACKAGE_PY_27=https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-1.7.0-cp27-none-linux_x86_64.whl

USER root
# Install all OS dependencies for notebook server that starts but lacks all
# features (e.g., download as all possible file formats)

ENV DEBIAN_FRONTEND noninteractive

ENV NB_USER jovyan
ENV NB_UID 1000
ENV HOME /home/$NB_USER
ENV CONDA_DIR=$HOME/.conda
ENV PATH $CONDA_DIR/bin:$PATH

# Use bash instead of sh
SHELL ["/bin/bash", "-c"]

RUN apt-get update && apt-get install -yq --no-install-recommends \
apt-transport-https \
build-essential \
bzip2 \
ca-certificates \
curl \
emacs \
fonts-liberation \
g++ \
git \
graphviz \
inkscape \
jed \
libav-tools \
libcupti-dev \
libsm6 \
libxext-dev \
libxrender1 \
lmodern \
locales \
lsb-release \
openssh-client \
pandoc \
pkg-config \
python \
python-dev \
sudo \
unzip \
vim \
wget \
zip \
zlib1g-dev \
&& apt-get clean && \
rm -rf /var/lib/apt/lists/*
apt-transport-https \
build-essential \
bzip2 \
ca-certificates \
curl \
g++ \
git \
graphviz \
locales \
lsb-release \
sudo \
unzip \
vim \
wget \
zip \
&& apt-get clean && \
rm -rf /var/lib/apt/lists/*

RUN echo "en_US.UTF-8 UTF-8" > /etc/locale.gen && \
locale-gen

# Install Tini
RUN wget --quiet https://github.com/krallin/tini/releases/download/v0.10.0/tini && \
echo "1361527f39190a7338a0b434bd8c88ff7233ce7b9a4876f3315c22fce7eca1b0 *tini" | sha256sum -c - && \
mv tini /usr/local/bin/tini && \
chmod +x /usr/local/bin/tini

# Install ksonnet
RUN wget --quiet https://github.com/ksonnet/ksonnet/releases/download/v0.9.2/ks_0.9.2_linux_amd64.tar.gz && \
tar -zvxf ks_0.9.2_linux_amd64.tar.gz && \
mv ks_0.9.2_linux_amd64/ks /usr/local/bin/ks && \
chmod +x /usr/local/bin/ks

# Configure environment
ENV CONDA_DIR /opt/conda
ENV PATH $CONDA_DIR/bin:$PATH
ENV SHELL /bin/bash
ENV NB_USER jovyan
ENV NB_UID 1000
ENV HOME /home/$NB_USER
ENV LC_ALL en_US.UTF-8
ENV LANG en_US.UTF-8
ENV LANGUAGE en_US.UTF-8

# Create jovyan user with UID=1000 and in the 'users' group
RUN useradd -m -s /bin/bash -N -u $NB_UID $NB_USER && \
mkdir -p $CONDA_DIR && \
chown $NB_USER $CONDA_DIR
chown -R ${NB_USER}:users /usr/local/bin

RUN export CLOUD_SDK_REPO="cloud-sdk-$(lsb_release -c -s)" && \
echo "deb https://packages.cloud.google.com/apt $CLOUD_SDK_REPO main" > /etc/apt/sources.list.d/google-cloud-sdk.list && \
curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | apt-key add - && \
apt-get update && \
apt-get install -y google-cloud-sdk kubectl && \
gcloud config set core/disable_usage_reporting true && \
gcloud config set component_manager/disable_update_check true && \
gcloud config set metrics/environment github_docker_image

RUN chown -R ${NB_USER}:users /home/${NB_USER}

# Run everything below this as $NB_USER
USER $NB_USER

WORKDIR $HOME

# Setup work directory for backward-compatibility
RUN mkdir /home/$NB_USER/work

# Install conda as jovyan and check the md5 sum provided on the download site
ENV MINICONDA_VERSION 4.3.21
# Install Tini - used as entrypoint for container
RUN cd /tmp && \
wget --quiet https://github.com/krallin/tini/releases/download/v0.10.0/tini && \
echo "1361527f39190a7338a0b434bd8c88ff7233ce7b9a4876f3315c22fce7eca1b0 *tini" | sha256sum -c - && \
mv tini /usr/local/bin/tini && \
chmod +x /usr/local/bin/tini

# Install conda as jovyan user and check the md5 sum provided on the download site
ENV MINICONDA_VERSION 4.4.10
RUN cd /tmp && \
mkdir -p $CONDA_DIR && \
wget --quiet https://repo.continuum.io/miniconda/Miniconda3-${MINICONDA_VERSION}-Linux-x86_64.sh && \
echo "c1c15d3baba15bf50293ae963abef853 *Miniconda3-${MINICONDA_VERSION}-Linux-x86_64.sh" | md5sum -c - && \
/bin/bash Miniconda3-${MINICONDA_VERSION}-Linux-x86_64.sh -f -b -p $CONDA_DIR && \
echo "bec6203dbb2f53011e974e9bf4d46e93 *Miniconda3-${MINICONDA_VERSION}-Linux-x86_64.sh" | md5sum -c - && \
/bin/bash Miniconda3-${MINICONDA_VERSION}-Linux-x86_64.sh -f -b -p ${CONDA_DIR} && \
rm Miniconda3-${MINICONDA_VERSION}-Linux-x86_64.sh && \
$CONDA_DIR/bin/conda config --system --prepend channels conda-forge && \
$CONDA_DIR/bin/conda config --system --set auto_update_conda false && \
$CONDA_DIR/bin/conda config --system --set show_channel_urls true && \
$CONDA_DIR/bin/conda update --all && \
conda config --system --prepend channels conda-forge && \
conda config --system --set auto_update_conda false && \
conda config --system --set show_channel_urls true && \
conda update --all && \
conda update conda && \
conda clean -tipsy

# Install Jupyter Notebook and Hub
RUN conda install --quiet --yes \
'nodejs=8.10*' \
'notebook=5.0.*' \
'jupyterhub=0.8.1' \
'jupyterlab=0.31.*' \
# Install base python3 packages
RUN pip install --upgrade pip && \
pip --no-cache-dir install \
# Tensorflow
${TF_PACKAGE} \
# Jupyter Stuff
jupyter \
jupyterhub \
jupyterlab \
# Cleanup
&& conda clean -tipsy

EXPOSE 8888
WORKDIR $HOME

# Configure container startup
ENTRYPOINT ["tini", "--"]
CMD ["start-notebook.sh"]

# Install CUDA Profile Tools and other python packages
RUN pip --no-cache-dir install \
google-cloud-storage \
Pillow \
h5py \
# Install python2 and ipython2 kernel for jupyter notebook
# Install tf packages which only support py2
RUN conda create -n py2 python=2 && \
source activate py2 && \
pip install --upgrade pip && \
pip --no-cache-dir install \
ipykernel \
matplotlib \
numpy \
scipy \
sklearn \
kubernetes \
grpcio \
ktext \
annoy \
nltk \
pydot \
pydot-ng \
graphviz \
# Tensorflow
${TF_PACKAGE_PY_27} \
# Tensorflow packages which only supports python 2
tensorflow-transform \
tensorflow-serving-api \
# ipykernel for python 2 jupyter notebook kernel
&& \
python -m ipykernel.kernelspec

# Install Python 3 packages
# Remove pyqt and qt pulled in for matplotlib since we're only ever going to
# use notebook-friendly backends in these images
python -m ipykernel install --user && \
# tensorflow-model-analysis is only supported for TF 1.6 and above
if [[ $INSTALL_TFMA == "yes" ]]; then \
pip install --no-cache-dir tensorflow-model-analysis && \
jupyter nbextension install --py --symlink tensorflow_model_analysis --user && \
jupyter nbextension enable --py tensorflow_model_analysis --user; \
fi

# Install jupyterlab-manager
RUN conda install --quiet --yes \
'nomkl' \
'ipywidgets=7.1*' \
'pandas=0.22*' \
'numexpr=2.6*' \
'matplotlib=2.0*' \
'scipy=0.19*' \
'seaborn=0.7*' \
'scikit-learn=0.18*' \
'scikit-image=0.12*' \
'sympy=1.0*' \
'cython=0.25*' \
'patsy=0.4*' \
'statsmodels=0.8*' \
'cloudpickle=0.2*' \
'dill=0.2*' \
'numba=0.31*' \
'bokeh=0.12*' \
'sqlalchemy=1.1*' \
'hdf5=1.8.17' \
'h5py=2.6*' \
'vincent=0.4.*' \
'beautifulsoup4=4.5.*' \
'xlrd' && \
conda remove --quiet --yes --force qt pyqt && \
conda clean -tipsy
# nodejs required for jupyterlab-manager
nodejs && \
jupyter labextension install @jupyter-widgets/jupyterlab-manager

ARG TF_PACKAGE=tf-nightly
RUN echo Installing $TF_PACKAGE && pip install --upgrade --quiet --no-cache-dir $TF_PACKAGE
# Install common packages from requirements.txt for both python2 and python3
COPY --chown=jovyan:users requirements.txt $HOME/requirements.txt
RUN pip --no-cache-dir install -r $HOME/requirements.txt && \
source activate py2 && \
pip --no-cache-dir install -r $HOME/requirements.txt

# Hack recommended in tf-serving-api to use it with py3. Remove when proper pip package is available.
COPY --from=tf-serving-install /usr/local/lib/python2.7/site-packages/tensorflow_serving /opt/conda/lib/python3.6/site-packages/tensorflow_serving
# Copy over init scripts
COPY --chown=jovyan:users start-singleuser.sh start-notebook.sh start.sh /usr/local/bin/
COPY --chown=jovyan:users jupyter_notebook_config.py $HOME/.jupyter/
RUN chmod a+rx /usr/local/bin/*

ENV CLOUD_SDK_VERSION 168.0.0
RUN export CLOUD_SDK_REPO="cloud-sdk-$(lsb_release -c -s)" && \
echo "deb https://packages.cloud.google.com/apt $CLOUD_SDK_REPO main" > /etc/apt/sources.list.d/google-cloud-sdk.list && \
curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | apt-key add - && \
apt-get update && \
apt-get install -y google-cloud-sdk=${CLOUD_SDK_VERSION}-0 kubectl && \
gcloud config set core/disable_usage_reporting true && \
gcloud config set component_manager/disable_update_check true && \
gcloud config set metrics/environment github_docker_image

# Activate ipywidgets extension in the environment that runs the notebook server
RUN jupyter nbextension enable --py widgetsnbextension --sys-prefix

RUN curl -L -o bazel.sh https://github.com/bazelbuild/bazel/releases/download/0.11.1/bazel-0.11.1-installer-linux-x86_64.sh && chmod a+x ./bazel.sh && ./bazel.sh && rm ./bazel.sh
SHELL ["/bin/bash", "-c"]

RUN git clone https://github.com/tensorflow/models.git /home/$NB_USER/tensorflow-models && git clone https://github.com/tensorflow/benchmarks.git /home/$NB_USER/tensorflow-benchmarks
# Import matplotlib the first time to build the font cache.
ENV XDG_CACHE_HOME /home/$NB_USER/.cache/

# Create a conda environment for Python 2. We want to include as many of the
# packages from our root environment as we reasonably can, so we explicitly
# list that environment, then include everything unless it is Conda (which
# can only be in the root environment), Jupyterhub (which requires Python 3),
# or Python itself. We also want to include the pip packages, but we cannot
# install those via conda, so we list them, drop any conda packages, and
# then install them via pip. We do this on a best-effort basis, so if any
# packages from the Python 3 environment cannot be installed with Python 2,
# then we just skip them.
RUN conda_packages=$(conda list -e | cut -d '=' -f 1 | grep -v '#' | sort) && \
pip_packages=$(pip --no-cache-dir list --format=freeze | cut -d '=' -f 1 | grep -v '#' | sort) && \
pip_only_packages=$(comm -23 <(echo "${pip_packages}") <(echo "${conda_packages}")) && \
conda create -n ipykernel_py2 python=2 --file <(echo "${conda_packages}" | grep -v conda | grep -v python | grep -v jupyterhub) && \
source activate ipykernel_py2 && \
python -m ipykernel install --user && \
echo "${pip_only_packages}" | xargs -n 1 -I "{}" /bin/bash -c 'pip install --no-cache-dir {} || true' && \
pip install --no-cache-dir tensorflow-transform && \
source deactivate

# Add local files as late as possible to avoid cache busting
COPY start.sh /usr/local/bin/
COPY start-notebook.sh /usr/local/bin/
COPY start-singleuser.sh /usr/local/bin/
COPY jupyter_notebook_config.py /etc/jupyter/
RUN chown -R $NB_USER:users /etc/jupyter/ && \
chown -R $NB_USER /home/$NB_USER/ && \
chmod a+rx /usr/local/bin/*

# tornado 5 has breaking changes (https://github.com/jupyter/notebook/issues/3407)
# and does not work with the current jupyterhub. Pin tornado to 4.5.3
# TODO(ankushagarwal): Figure out the right way to fix this
RUN pip install tornado==4.5.3

USER $NB_USER
ENV PATH=/home/jovyan/bin:$PATH
# Configure container startup
EXPOSE 8888
ENTRYPOINT ["tini", "--"]
CMD ["start-notebook.sh"]
4 changes: 4 additions & 0 deletions components/tensorflow-notebook-image/build_image.sh
View file
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,8 @@ TAG=$3
IS_LATEST=$4
BASE_IMAGE=${5:-"ubuntu:latest"}
TF_PACKAGE=${6:-"tf-nightly"}
TF_PACKAGE_PY_27=${7:-"tf-nightly"}
INSTALL_TFMA=$8

# Wait for the Docker daemon to be available.
until docker ps
Expand All @@ -25,6 +27,8 @@ done
docker build --pull \
--build-arg "BASE_IMAGE=${BASE_IMAGE}" \
--build-arg "TF_PACKAGE=${TF_PACKAGE}" \
--build-arg "TF_PACKAGE_PY_27=${TF_PACKAGE_PY_27}" \
--build-arg "INSTALL_TFMA=${INSTALL_TFMA}" \
-t "${IMAGE}:${TAG}" \
-f ${DOCKERFILE} ${CONTEXT_DIR}

Expand Down
9 changes: 8 additions & 1 deletion components/tensorflow-notebook-image/releaser/components/workflows.libsonnet
View file
Original file line number Diff line number Diff line change
Expand Up @@ -133,6 +133,11 @@
"nvidia/cuda:8.0-cudnn6-devel-ubuntu16.04"
else
"nvidia/cuda:9.0-cudnn7-devel-ubuntu16.04",
local installTfma =
if tf_version < "1.6" then
"no"
else
"yes",
local tf_package =
"https://storage.googleapis.com/tensorflow/linux/" +
device +
Expand All @@ -154,7 +159,9 @@
+ tag + " "
+ std.toString(is_latest) + " "
+ base_image + " "
+ tf_package,
+ tf_package + " "
+ tf_package_py_27 + " "
+ installTfma,
],
[
{
Expand Down
Loading

0 comments on commit b497fd5

Please sign in to comment.