There are several topics we intend to cover in this series:

• Installation and setup
• Using dflow to write a simple workflow
  • dflow-python.ipynb
  • dflow-function.ipynb
• Advanced functionality of dflow
  • running jobs in parallel: dflow-slices.ipynb
  • writing while loop in dflow: dflow-recurse.ipynb
  • reusing operated nodes: dflow-reuse.ipynb
  • complex condition: dflow-conditional.ipynb
• Using dflow to submit job to slurm cluster
  • dflow-slurm.ipynb

Installation and setup

Installation

We need to install three dependencies to use dflow:

• Container engine: Docker(知乎介绍)
• Kubernetes: minikube(知乎介绍)
• dflow: pydflow

Easy Install

You can use the installation script to install all server and setup in one step. If you choose to use this method, you can skip all the installation and setup process.

IP-address outside China

• MacOS: https://github.com/deepmodeling/dflow/blob/master/scripts/install-mac.sh
• WindowsOS: Coming Soon. Submit your installation script here.
• On Linux: https://github.com/deepmodeling/dflow/blob/master/scripts/install-linux.sh

IP-address inside China

• MacOS: https://github.com/deepmodeling/dflow/blob/master/scripts/install-mac-cn.sh
• WindowsOS: Coming Soon. Submit your installation script here.
• On Linux: https://github.com/deepmodeling/dflow/blob/master/scripts/install-linux-cn.sh

Install Manually

Server

Container engine

• Docker installation is very easy. Check out its official installation guide: Get Docker

Kubernetes

• If you are setting up Kubernetes on your own laptop, you can install minikube. Checkout its official installation guide: minikube start

Step-by-step installation guide

• Install on MacOS
• Install on WindowsOS
• Install on LinuxOS

Pydflow

After you have installed the first two dependencies, you can install dflow using pip.

pip install pydflow

Setup

Minikube

dflow runs on Kubernetes (k8s), so we need to start minikube

minikube start

Start minikube in China

To speed up the startup, we can use the image repository setup on aliyun. To start minikube,

minikube start --image-mirror-country=cn

Argo

dflow is built on argo-workflow, so we need to setup argo engine in Kubernetes or minikube.

Install Argo with IP address outside China

To get started quickly, we can use the quick start manifest which will install Argo Workflows as well as some commonly used components:

kubectl create ns argo
kubectl apply -n argo -f https://raw.githubusercontent.com/deepmodeling/dflow/master/manifests/quick-start-postgres-3.4.1-deepmodeling.yaml

Install Argo with IP address inside China

Although you can still use the above method to install Argo, it might lead to a very long wait. To speed up the installation process, you can use another YAML file.

# download the YAML file to local
wget https://raw.githubusercontent.com/deepmodeling/dflow/master/manifests/quick-start-postgres-3.4.1-deepmodeling.yaml
kubectl apply -n argo -f quick-start-postgres-3.4.1-deepmodeling.yaml
# linux user might need to use `minikube kubectl --`

Monitor the install and setup progress

To monitor the setup progress, we can checkout the pod status.

kubectl get pod -n argo
# linux user might need to use `minikube kubectl --`

NOTE!!!!: This process might take a while, depending on the internet speed. Wait and keep refreshing the above cell. Once the STATUS of all pods is RUNNING, you can proceed with the next step.

Port-forward Argo UI and Minio API

Access the Argo UI:

!!!!IMPORTANT!!!! Since we need to keep this UI running, we have to keep this command running.

kubectl -n argo port-forward deployment/argo-server 2746:2746 --address 0.0.0.0
# linux user might need to use `minikube kubectl --`

You can log in the Argo UI via this address: https://127.0.0.1:2746. Please ignore the security warning.

Access the minio API:

!!!!IMPORTANT!!!! Open another terminal and run this, because you want to keep artifact respository running. Note that you don't need to ingress the artifact repository if you are not downloading or uploading artifact.

kubectl -n argo port-forward deployment/minio 9000:9000 --address 0.0.0.0
# linux user might need to use `minikube kubectl --`

[BONUS] Access minio UI

kubectl -n argo port-forward deployment/minio 9001:9001 --address 0.0.0.0
# linux user might need to use `minikube kubectl --`

You can log in the Argo UI via this address: http://127.0.0.1:9001. The default login credentials is:

• admin: admin
• password: password

That's it! You've finished the installation and setup.

FAQ

1. minikube start failure

• Problem Description: After minikube start, you probably saw this message:

• Bug source:

  • One common reason is that minikube starts with incorrect images. We can see the details of the log using the following command

  minikube ssh #enter minikube node 
  sudo journalctl -xeu kubelet
  

NOTE: minikube needs `k8s.gcr.io/pause:3.6`, but it pulled `k8s.gcr.io/pause:3.7'. So it will not start.

• Solutions:
  • (Recommended) Enter minikube environment and pull the image manually.

  minikube ssh
  docker pull registry.aliyuncs.com/google_containers/pause:3.6
  docker tag registry.aliyuncs.com/google_containers/pause:3.6 k8s.gcr.io/pause:3.6
  exit
  

  • Downgrade minikube version: you can download older version in this page: Releases · kubernetes/minikube