[MLOps] kubeflow 설정

업데이트: October 20, 2020

kubeflow 설정

1. kubeflow 기본 사용법

웹브라우저를 하나 열고 마스터 서버 IP와 포트 31380을 적어주세요. 예를 들어 마스터 서버 IP가 111.11.1.111 이라고하면 아래와 같이 적어주면 됩니다.

111.11.1.111:31380

그러고 몇 초 기다리면 아래와 같은 화면이 나옵니다.

위 화면에서 왼쪽 메뉴에 Notebook server를 누르면 현재 만들어진 주피터 노트북서버를 볼 수 있으며, 머신러닝을 위한 주피터 노트북 서버를 만들 수 있습니다.

위 화면에서 +New server를 누르면 새로운 주피터 노트북 서버를 만들 수 있습니다.

위와 같이 여러 설정 사항을 정하고 Launch를 누르면 만들어집니다. 그리고 서버 목록에서 connect를 누르면

위와 같이 주피터 노트북을 사용할 수 있습니다.

2. kubeflow에 텐서플로 이미지 올리기

아래 버전중 자신이 원하는 이미지를 올리세요.

2.1 kubeflow-jupyterlab:pytorch1.6-cpu

$ mkdir -p docker_image/jupyterlab/pytorch-cpu
$ vim docker_image/jupyterlab/pytorch-cpu/Dockerfile

FROM python:3.6

WORKDIR /home/jovyan

USER root

RUN pip install jupyter -U && pip install jupyterlab

ENV DEBIAN_FRONTEND=noninteractive

RUN apt-get update && apt-get install -yq --no-install-recommends \
  apt-transport-https \
  build-essential \
  bzip2 \
  ca-certificates \
  curl \
  g++ \
  git \
  gnupg \
  graphviz \
  locales \
  lsb-release \
  openssh-client \
  sudo \
  unzip \
  vim \
  wget \
  zip \
  emacs \
  python3-pip \
  python3-dev \
  python3-setuptools \
  libgl1-mesa-glx \
  python3-opencv \
  cmake \
  ninja-build \
  && apt-get clean && \
  rm -rf /var/lib/apt/lists/*

RUN curl -s https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -
RUN echo "deb https://apt.kubernetes.io/ kubernetes-xenial main" | sudo tee -a /etc/apt/sources.list.d/kubernetes.list
RUN apt-get update
RUN apt-get install -y kubectl

RUN pip install msrestazure
RUN pip install cloudpickle
RUN pip install kubeflow-fairing==0.7.2
RUN pip install kfp==0.5  --use-feature=2020-resolver
RUN pip install kfserving==0.2.2.1
RUN pip install kubeflow-kale
RUN pip install dill

RUN pip install tensorboard
RUN pip install torch==1.6 torchvision==0.7 -f https://download.pytorch.org/whl/cu101/torch_stable.html

RUN pip install opencv-python

RUN pip install jupyterlab && \
    jupyter serverextension enable --py jupyterlab --sys-prefix

ARG NB_USER=jovyan

EXPOSE 8888

ENV NB_USER $NB_USER
ENV NB_UID=1000
ENV HOME /home/$NB_USER
ENV NB_PREFIX /

CMD ["sh", "-c", "jupyter lab --notebook-dir=/home/jovyan --ip=0.0.0.0 --no-browser --allow-root --port=8888 --LabApp.token='' --LabApp.password='' --LabApp.allow_origin='*' --LabApp.base_url=${NB_PREFIX}"]

2.2 kubeflow-jupyterlab:tf2.3-cpu

$ mkdir -p docker_image/jupyterlab/tf-cpu
$ nano docker_image/jupyterlab/tf-cpu/Dockerfile

FROM python:3.6

WORKDIR /home/jovyan

USER root

RUN pip install jupyter -U && pip install jupyterlab

ENV DEBIAN_FRONTEND=noninteractive

RUN apt-get update && apt-get install -yq --no-install-recommends \
  apt-transport-https \
  build-essential \
  bzip2 \
  ca-certificates \
  curl \
  g++ \
  git \
  gnupg \
  graphviz \
  locales \
  lsb-release \
  openssh-client \
  sudo \
  unzip \
  vim \
  wget \
  zip \
  emacs \
  python3-pip \
  python3-dev \
  python3-setuptools \
  libgl1-mesa-glx \
  python3-opencv \
  cmake \
  ninja-build \
  && apt-get clean && \
  rm -rf /var/lib/apt/lists/*

RUN curl -s https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -
RUN echo "deb https://apt.kubernetes.io/ kubernetes-xenial main" | sudo tee -a /etc/apt/sources.list.d/kubernetes.list
RUN apt-get update
RUN apt-get install -y kubectl

RUN pip install msrestazure
RUN pip install cloudpickle
RUN pip install kubeflow-fairing==0.7.2
RUN pip install kfp==0.5  --use-feature=2020-resolver
RUN pip install kfserving==0.2.2.1
RUN pip install kubeflow-kale
RUN pip install dill

RUN pip install tensorboard
RUN pip install tensorflow-cpu==2.3.0
RUN pip install opencv-python

RUN pip install jupyterlab && \
    jupyter serverextension enable --py jupyterlab --sys-prefix

ARG NB_USER=jovyan

EXPOSE 8888

ENV NB_USER $NB_USER
ENV NB_UID=1000
ENV HOME /home/$NB_USER
ENV NB_PREFIX /

CMD ["sh", "-c", "jupyter lab --notebook-dir=/home/jovyan --ip=0.0.0.0 --no-browser --allow-root --port=8888 --LabApp.token='' --LabApp.password='' --LabApp.allow_origin='*' --LabApp.base_url=${NB_PREFIX}"]

2.3 cuda-python:10.1-3.6

$ mkdir -p docker_image/cuda-python
$ nano docker_image/cuda-python/Dockerfile

FROM nvidia/cuda:10.1-cudnn7-devel

RUN apt-get update && apt-get install -y --no-install-recommends ca-certificates curl netbase wget && rm -rf /var/lib/apt/lists/*

RUN set -ex; if ! command -v gpg > /dev/null; then apt-get update; apt-get install -y --no-install-recommends gnupg dirmngr; rm -rf /var/lib/apt/lists/*; fi

RUN apt-get update && apt-get install -y --no-install-recommends git mercurial openssh-client subversion procps && rm -rf /var/lib/apt/lists/*

RUN set -ex; apt-get update; DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends autoconf automake bzip2 dpkg-dev file g++ gcc imagemagick libbz2-dev libc6-dev libcurl4-openssl-dev libdb-dev libevent-dev libffi-dev libgdbm-dev libglib2.0-dev libgmp-dev libjpeg-dev libkrb5-dev liblzma-dev libmagickcore-dev libmagickwand-dev libmaxminddb-dev libncurses5-dev libncursesw5-dev libpng-dev libpq-dev libreadline-dev libsqlite3-dev libssl-dev libtool libwebp-dev libxml2-dev libxslt-dev libyaml-dev make patch unzip xz-utils zlib1g-dev $(if apt-cache show 'default-libmysqlclient-dev' 2>/dev/null | grep -q '^Version:'; then echo 'default-libmysqlclient-dev'; else echo 'libmysqlclient-dev'; fi); rm -rf /var/lib/apt/lists/*

ENV PATH=/usr/local/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin

ENV LANG=C.UTF-8

ENV GPG_KEY=0D96DF4D4110E5C43FBFB17F2D347EA6AA65421D

ENV PYTHON_VERSION=3.6.12

ENV PYTHON_GET_PIP_URL=https://github.com/pypa/get-pip/raw/5578af97f8b2b466f4cdbebe18a3ba2d48ad1434/get-pip.py

ENV PYTHON_GET_PIP_SHA256=d4d62a0850fe0c2e6325b2cc20d818c580563de5a2038f917e3cb0e25280b4d1

RUN set -ex; wget -O get-pip.py "$PYTHON_GET_PIP_URL"; echo "$PYTHON_GET_PIP_SHA256 *get-pip.py" | sha256sum --check --strict -; python get-pip.py               --disable-pip-version-check             --no-cache-dir          "pip==$PYTHON_PIP_VERSION"      ;       pip --version;          find /usr/local -depth          \(                      \( -type d -a \( -name test -o -name tests -o -name idle_test \) \)                     -o                      \( -type f -a \( -name '*.pyc' -o -name '*.pyo' \) \)           \) -exec rm -rf '{}' +;         rm -f get-pip.py

2.3 kubeflow-jupyterlab:pytorch1.6-gpu

$ mkdir -p docker_image/jupyterlab/pytorch-cpu
$ nano docker_image/jupyterlab/pytorch-cpu/Dockerfile

FROM cuda-python:10.1-3.6

WORKDIR /home/jovyan

USER root

RUN pip install jupyter -U && pip install jupyterlab

ENV DEBIAN_FRONTEND=noninteractive

RUN apt-get update && apt-get install -yq --no-install-recommends \
  apt-transport-https \
  build-essential \
  bzip2 \
  ca-certificates \
  curl \
  g++ \
  git \
  gnupg \
  graphviz \
  locales \
  lsb-release \
  openssh-client \
  sudo \
  unzip \
  vim \
  wget \
  zip \
  emacs \
  python3-pip \
  python3-dev \
  python3-setuptools \
  libgl1-mesa-glx \
  python3-opencv \
  cmake \
  ninja-build \
  && apt-get clean && \
  rm -rf /var/lib/apt/lists/*

RUN curl -s https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -
RUN echo "deb https://apt.kubernetes.io/ kubernetes-xenial main" | sudo tee -a /etc/apt/sources.list.d/kubernetes.list
RUN apt-get update
RUN apt-get install -y kubectl

RUN pip install msrestazure
RUN pip install cloudpickle
RUN pip install kubeflow-fairing==0.7.2
RUN pip install kfp==0.5  --use-feature=2020-resolver
RUN pip install kfserving==0.2.2.1
RUN pip install kubeflow-kale
RUN pip install dill

RUN pip install tensorboard
RUN pip install torch==1.6 torchvision==0.7 -f https://download.pytorch.org/whl/cu101/torch_stable.html

RUN pip install opencv-python

RUN pip install jupyterlab && \
    jupyter serverextension enable --py jupyterlab --sys-prefix

ARG NB_USER=jovyan

EXPOSE 8888

ENV NB_USER $NB_USER
ENV NB_UID=1000
ENV HOME /home/$NB_USER
ENV NB_PREFIX /

CMD ["sh", "-c", "jupyter lab --notebook-dir=/home/jovyan --ip=0.0.0.0 --no-browser --allow-root --port=8888 --LabApp.token='' --LabApp.password='' --LabApp.allow_origin='*' --LabApp.base_url=${NB_PREFIX}"]

3. docker build

아래 두가지 중에 자신의 버전에 맞는 것을 실행하세요(둘다 실행하면 안되고 하나만 실행하세요)

3.1 파이토치

$ sudo docker build -t kubeflow-jupyterlab:pytorch1.6-cpu docker_image/jupyterlab/pytorch-cpu

3.2 텐서플로

$ sudo docker build -t kubeflow-jupyterlab:tf2.3-cpu docker_image/jupyterlab/tf-cpu

4. image tag

아래 두가지 중에 자신의 버전에 맞는 것을 실행하세요.(둘다 실행하면 안되고 하나만 실행하세요.)

4.1 파이토치

$ sudo docker tag kubeflow-jupyterlab:pytorch1.6-cpu  kubeflow-registry.default.svc.cluster.local:30000/kubeflow-jupyterlab:pytorch1.6-cpu

4.2 텐서플로

$ sudo docker tag kubeflow-jupyterlab:tf2.3-cpu  kubeflow-registry.default.svc.cluster.local:30000/kubeflow-jupyterlab:tf2.3-cpu

5. image push

아래 두가지 중에 자신의 버전에 맞는 것을 실행하세요.(둘다 실행하면 안되고 하나만 실행하세요.)

5.1 파이토치

$ sudo docker push  kubeflow-registry.default.svc.cluster.local:30000/kubeflow-jupyterlab:pytorch1.6-cpu

5.2 텐서플로

$ sudo docker push  kubeflow-registry.default.svc.cluster.local:30000/kubeflow-jupyterlab:tf2.3-cpu

6. katib

apiVersion: rbac.authorization.k8s.io/v1beta1
kind: Role
metadata:
  name: access-kubeflow
  namespace: kubeflow
rules:
- apiGroups: ["", "kubeflow.org"]
  resources: ["pods", "pods/log", "experiments", "persistentvolumeclaims"]
  verbs: ["create", "delete", "update", "patch", "get", "list", "watch"]
---
apiVersion: rbac.authorization.k8s.io/v1
kind: RoleBinding
metadata:
  name: kubeflow-access-role
  namespace: kubeflow
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: Role
  name: access-kubeflow
subjects:
- kind: ServiceAccount
  name: default-editor
  namespace: 본인네임스페이스

metal lb setting

MetalLB deployment

Deploy MetalLB:

Apply the manifest:

 kubectl apply -f https://raw.githubusercontent.com/google/metallb/v0.8.1/manifests/metallb.yaml

Allocate a pool of addresses on your local network for MetalLB to use. You need at least one address for the Istio Gateway. This example assumes addresses 10.0.0.100-10.0.0.110. You must modify these addresses based on your environment.

 cat <<EOF | kubectl apply -f -
 apiVersion: v1
 kind: ConfigMap
 metadata:
   namespace: metallb-system
   name: config
 data:
   config: |
     address-pools:
     - name: default
       protocol: layer2
       addresses:
       - 10.0.0.100-10.0.0.110
 EOF

Ensure that MetalLB works as expected (optional):

Create a dummy service:

 kubectl create service loadbalancer nginx --tcp=80:80

Ensure that MetalLB has allocated an IP address for the service:
```
 kubectl describe service nginx
```

Check the corresponding MetalLB logs:

 kubectl logs -n metallb-system -l component=controller

Create a pod that will be exposed with the service:

 kubectl run nginx --image nginx --restart=Never -l app=nginx

Ensure that MetalLB has assigned a node to announce the allocated IP address:
```
 kubectl describe service nginx
```

Check the corresponding MetalLB logs:

 kubectl logs -n metallb-system -l component=speaker

Check that MetalLB responds to ARP requests for the allocated IP address:
```
 arping -I eth0 10.0.0.101
```

Check the corresponding MetalLB logs:

 kubectl logs -n metallb-system -l component=speaker

Verify that everything works as expected:
```
 curl http://10.0.0.101
```

Clean up:

kubectl delete service nginx
kubectl delete pod nginx

To expose Kubeflow with a LoadBalancer Service, just change the type of the istio-ingressgateway Service to LoadBalancer.

kubectl patch service -n istio-system istio-ingressgateway -p '{"spec": {"type": "LoadBalancer"}}'

After that, get the LoadBalancer’s IP or Hostname from its status and create the necessary certificate.

Create the Certificate with cert-manager:

After applying the above Certificate, cert-manager will generate the TLS certificate inside the istio-ingressgateway-certs secrets. The istio-ingressgateway-certs secret is mounted on the istio-ingressgateway deployment and used to serve HTTPS.

Navigate to https://<LoadBalancer Address>/ and start using Kubeflow.

end to end ml project

web-ui

kubectl port-forward -n istio-system svc/istio-ingressgateway 8080:80

sudo docker run -p 19000:5000 brightfly/fmnist-webui-deploy:latest

호스트ip:19000?model=kfserving-fmnist&name=kfserving-fmnist.kubeflow.example.com&addr=10.0.0.100

Twitter Facebook Google+ LinkedIn

장철원(Cheolwon Jang)

[MLOps] kubeflow 설정

kubeflow 설정

1. kubeflow 기본 사용법

2. kubeflow에 텐서플로 이미지 올리기

2.1 kubeflow-jupyterlab:pytorch1.6-cpu

2.2 kubeflow-jupyterlab:tf2.3-cpu

2.3 cuda-python:10.1-3.6

2.3 kubeflow-jupyterlab:pytorch1.6-gpu

3. docker build

3.1 파이토치

3.2 텐서플로

4. image tag

4.1 파이토치

4.2 텐서플로

5. image push

5.1 파이토치

5.2 텐서플로

6. katib

metal lb setting

end to end ml project

web-ui

공유하기

참고

[Python] 파이썬 아나콘다를 이용한 가상환경 관리

[딥러닝] Attention is all you need 논문 번역

[Infra] 쿠버네티스(kubernetes)(2) 쿠버네티스 설치(Kubernetes install)

[Infra] 도커(docker)(6) 도커 컴포즈(docker compose)를 활용한 여러 개의 Flask, Nginx 배포