入门机器学习：手把手教你实现简单的算法

# Ungraded Lab: Practice Kubernetes in your Local Environment In the previous reading item, you went through the basics of Kubernetes by doing imperative commands in the interactive shell (i.e. with `kubectl`). It shows how this tool can be used to orchestrate containers to host an application. If you haven't done that exercise, we highly encourage you to go back because we will assume that you are already familiar with the concepts discussed in [those 6 modules](https://kubernetes.io/docs/tutorials/kubernetes-basics/). This lab will be an extension of that activity and will show a few more concepts to prepare you for this week's graded assignment. Specifically, you will: * setup Kubernetes in your local machine for learning and development * create Kubernetes objects using YAML files * deploy containers running a Tensorflow model server * access the deployment using a Nodeport service * autoscale the deployment to dynamicaly handle incoming traffic You should be able to run all the commands using Mac and Linux (tested on Ubuntu) CLI, Windows Powershell (run as Administrator) and Git Bash. In case you have limited resources or can't install in your workstation, don't worry about it. You can still refer to this document later and we'll show the expected output at each step. ## Installation First, you will setup your machine to run a local Kubernetes cluster. This gives you more freedom to tweak settings and implement more features compared to the online Katacoda environment you used in the Kubernetes tutorial. It's a great tool for learning and for local development as well. There are several Kubernetes distributions and the one best suited for our purpose is [Minikube](https://minikube.sigs.k8s.io/docs/). You might remember it being mentioned in the first module of the Kubernetes basics exercise. *Note: If you installed Docker for Desktop in the previous ungraded labs, you may have noticed that it also has a Kubernetes Engine that can be enabled in the Settings. At the moment, we found some limitations for it in some OS and it needs more tweaking or workarounds to make all the exercises here work. Thus, we recommend using Minikube for a more seamless experience.* You will need to install the following tools to go through this lab: * **curl** - a command-line tool for transferring data using various network protocols. You may have a already installed this in the Docker ungraded lab but in case you haven't, [here is one reference](https://reqbin.com/Article/InstallCurl) to do so. You will use this to query your model later. * **Virtualbox** - Minikube is meant to run in a virtual machine (VM) so you will need virtualization software to act as the VM driver. While you can also specify `docker` as the VM driver, we found that it has limitations so it's best to use Virtualbox instead. Installation instructions can be found [here](https://www.virtualbox.org/wiki/Downloads). When prompted by your OS, make sure to allow network traffic for this software so you won't have firewall issues later on. * **kubectl** - the command line tool for interacting with Kubernetes clusters. Installation instructions can be found [here](https://kubernetes.io/docs/tasks/tools/) * **Minikube** - a Kubernetes distribution geared towards new users and development work. It is not meant for production deployments however since it can only run a single node cluster on your machine. Installation instructions [here](https://minikube.sigs.k8s.io/docs/start/). <details> <summary> <i>Windows Users: Please click here for additional notes. </i></summary> 1. Make sure that the directories to the `curl` and `kubectl` binaries are setup in your system `PATH` (or `Path`). That will allow you to execute these in the command line from any directory. Instructions can be found [here](https://www.computerhope.com/issues/ch000549.htm) in case you need to review how this is done. 2. You may need to append a `.exe` in the commands later to make use of `curl` and `kubectl`. For example, if you see `curl --help`, please do `curl.exe --help` instead. 3. In case you have [Git for Windows](https://git-scm.com/download/win) setup in your machine, then you can use the `Git Bash` CLI bundled with that package instead of `Windows Powershell`. Just search for it in the Search Bar to see if it is already in your system. That CLI runs like a Linux terminal so most of the commands in the next sections will run as is. But we still placed instructions in case you can only use Powershell. 4. If you encounter errors about running VirtualBox alongside WSL2 similar to [this](https://github.com/MicrosoftDocs/WSL/issues/798), you can fallback to using Docker as the VM runtime for Minikube. Instructions are shown in the next sections in case you follow this path. --- </details> </br> ## Architecture The application you'll be building will look like the figure below: <img src='img/kubernetes.png' alt='img/kubernetes.png'> You will create a deployment that spins up containers that runs a model server. In this case, that will be from the `tensorflow/serving` image you already used in the previous labs. The deployment can be accessed by external terminals (i.e. your users) through an exposed service. This brings inference requests to the model servers and responds with predictions from your model. Lastly, the deployment will spin up or spin down pods based on CPU utilization. It will start with one pod but when the load exceeds a pre-defined point, it will spin up additional pods to share the load. ## Start Minikube You are now almost ready to start your Kubernetes cluster. There is just one more additional step. As mentioned earlier, Minikube runs inside a virtual machine. That implies that the pods you will create later on will only see the volumes inside this VM. Thus, if you want to load a model into your pods, then you should first mount the location of this model inside Minikube's VM. Let's set that up now. You will be using the `half_plus_two` model that you saw in earlier ungraded labs. You can copy it to your `/var/tmp` directory (Mac, Linux) or `C:/tmp` (Windows) so we'll have a common directory to mount to the VM. You can use the command below for Mac and Linux: ``` cp -R ./saved_model_half_plus_two_cpu /var/tmp ``` If you're using Windows (not WSL), then you can use the GUI to create a `tmp` folder under your `C:` drive then copy the folder there. You should have a `C:/tmp/saved_model_half_plus_two_cpu` folder as a result Now you're ready to start Minikube! Run the command below to initialize the VM with Virtualbox and mount the folder containing your model file: For Mac and Linux: ``` minikube start --mount=True --mount-string="/var/tmp:/var/tmp" --vm-driver=virtualbox ``` For Windows: ``` minikube start --mount=True --mount-string="C:/tmp:/var/tmp" --vm-driver=virtualbox ``` <details> <summary> <i>Troubleshooting: Please click here if you're getting errors with these commands. </i></summary> * Some learners reported prompts about driver errors and thus, they can't make Virtualbox the VM driver when launching Minikube. In case you run into the same issue and can't resolve it, you can just fallback to Docker: ``` minikube start --mount=True --mount-string="C:/tmp:/var/tmp" --vm-driver=docker ``` This would require revisions to some of the commands later and we placed that in *Troubleshooting* sections as well. * Some learners reported getting an error about trouble accessing `https://k8s.gcr.io` and needing to configure a proxy. For that, please do these steps before starting `minikube`: * Run `minikube ip`. This will display an IP address in your terminal. * Replace `<your_minikube_ip>` in the command below with the IP address you saw above: ``` set NO_PROXY=localhost,127.0.0.1,10.96.0.0/12,192.168.59.0/24,192.168.49.0/24,192.168.39.0/24,<your_minikube_ip> ``` * Start minikube --- </details> </br> ## Creating Objects with YAM