1 00:00:04,200 --> 00:00:07,050 Hello and welcome to this lecture. In this lecture 2 00:00:07,050 --> 00:00:13,810 we will discuss about designing a kubernetes cluster. Before you head into designing a cluster. 3 00:00:13,820 --> 00:00:16,490 I must ask the following questions. 4 00:00:16,490 --> 00:00:18,740 What is the purpose of this cluster? 5 00:00:18,770 --> 00:00:25,100 Is it for learning or development or testing purpose or for hosting production grade applications. 6 00:00:25,100 --> 00:00:27,740 What is the cloud adoption at your organization? 7 00:00:27,740 --> 00:00:33,650 Do you prefer your platform to be managed by a cloud provider or a self hosted? 8 00:00:33,650 --> 00:00:37,100 What kind of workloads are you going to run on this cluster? 9 00:00:37,100 --> 00:00:40,400 How many applications are to be hosted on the Cluster? 10 00:00:40,490 --> 00:00:42,050 Few or many? 11 00:00:42,050 --> 00:00:48,590 What kind of applications are going to be hosted on the Cluster? Web applications or big data or analytics? 12 00:00:49,310 --> 00:00:53,880 Depending on the kind of application the resource requirements may vary. 13 00:00:53,900 --> 00:00:57,500 What type of network traffic are these applications expecting? 14 00:00:57,500 --> 00:01:00,900 Continuous heavy traffic or burst? 15 00:01:00,910 --> 00:01:03,510 Well let's try and break down some of these. 16 00:01:03,890 --> 00:01:09,320 If you want to deploy a cluster for learning purposes then a solution based on minikube or a single 17 00:01:09,320 --> 00:01:17,120 node cluster deployed using kubeadm on local VMs or cloud providers like GCP or AWS should do. 18 00:01:17,120 --> 00:01:22,550 We have deployed such a cluster in the beginners course. To deploy a cluster for development and testing 19 00:01:22,550 --> 00:01:28,310 purposes, and multi node cluster with single master and multiple worker nodes would help. 20 00:01:28,310 --> 00:01:34,580 Again kubeadm is an appropriate tool. Or if on managed cloud environments, then quickly provision a 21 00:01:34,580 --> 00:01:44,630 cluster on GCP, AWS or AKS on Azure. Let’s talk about 22 00:01:44,660 --> 00:01:49,020 production level clusters. For hosting production grade applications. 23 00:01:49,070 --> 00:01:54,540 a High Availability Multi node cluster with multiple master nodes is recommended. 24 00:01:54,860 --> 00:02:01,370 We look at more details about High Availability setup with multiple-master nodes later in this section. 25 00:02:01,370 --> 00:02:08,570 Again this can be setup with kubeadm or GCP or using kops on AWS or other supported platforms. 26 00:02:08,810 --> 00:02:15,270 You can have upto 5000 nodes in the cluster, a total of 150k PODs in the cluster, 27 00:02:15,300 --> 00:02:21,770 300,000 containers in total and upto 100 PODs per node. Depending on the size of your 28 00:02:21,770 --> 00:02:22,430 cluster. 29 00:02:22,430 --> 00:02:28,850 the resource requirement for your nodes varies. CSPs like GCP and AWS automatically 30 00:02:28,850 --> 00:02:34,100 selects the right sized nodes for you based on the number of nodes in the cluster. 31 00:02:34,100 --> 00:02:39,680 This table shows the size of the instances and their resource specifications for a specific number of 32 00:02:39,680 --> 00:02:40,100 nodes. 33 00:02:40,730 --> 00:02:47,420 If you are deploying on prem nodes then you could probably start with these numbers as base. Cloud or 34 00:02:47,420 --> 00:02:48,080 Onprem? 35 00:02:48,140 --> 00:02:53,840 We have already discussed that all of these deployment options are available in any environment. For 36 00:02:53,840 --> 00:02:54,830 on-prem 37 00:02:54,830 --> 00:02:57,120 kubeadm is a very useful tool. 38 00:02:57,200 --> 00:03:02,780 Google Container engine makes provisioning kubernetes clusters on GCP very easy. 39 00:03:02,780 --> 00:03:08,990 It comes with one-click cluster upgrade features that makes it very easy to maintain the cluster. KOPS 40 00:03:09,050 --> 00:03:15,890 is a nice tool to deploy kubernetes cluster on AWS and the Azure Kubernetes Service or AKS 41 00:03:15,980 --> 00:03:23,400 helps in managing the hosted kubernetes environment on Azure. Depending on the workloads configured, your 42 00:03:23,400 --> 00:03:30,600 node and disk configurations will differ. For High Performance workloads rely on SSD Backed Storage. For 43 00:03:30,600 --> 00:03:32,680 multiple concurrent access 44 00:03:32,730 --> 00:03:38,410 consider network based storage. For shared access to volumes across multiple PODs, 45 00:03:38,490 --> 00:03:42,910 Consider persistent storage volumes that we discussed in the storage section. 46 00:03:43,050 --> 00:03:49,230 Consider defining different classes of storage and allocating the right class to the right applications. 47 00:03:49,320 --> 00:03:53,410 The Nodes forming a kubernetes cluster can be physical or virtual 48 00:03:53,460 --> 00:03:59,740 In our case we will be deploying virtual machines on VirtualBox environments has nodes of our cluster. 49 00:03:59,850 --> 00:04:05,100 You may chose to deploy on physical machines or virtual machines or cloud environments like GCP, 50 00:04:05,260 --> 00:04:09,140 AWS, Azure or any other platform of your choice. 51 00:04:09,180 --> 00:04:14,550 We will be building a cluster with three nodes one master and two worker nodes. 52 00:04:14,550 --> 00:04:19,860 Now We know that master nodes are for hosting control plane components like the kube-api server, etcd 53 00:04:19,860 --> 00:04:24,030 server and others, while worker nodes for hosting workloads. 54 00:04:24,030 --> 00:04:26,830 However this is not a strict requirement. 55 00:04:26,850 --> 00:04:32,580 The master nodes are also considered as nodes and can host workloads. As a best practice 56 00:04:32,580 --> 00:04:38,370 it is recommended to dedicate master nodes for control plane components only specially in a production 57 00:04:38,370 --> 00:04:39,660 environment. 58 00:04:39,660 --> 00:04:46,350 Deployment tools like kubeadm prevent workloads from being hosted on master nodes by adding a taint to 59 00:04:46,350 --> 00:04:47,850 the master node. 60 00:04:48,010 --> 00:04:52,160 You must use 64 bit Linux operating system for nodes. 61 00:04:52,240 --> 00:04:57,200 Another thing to note is that typically you have all the control plane components on the master nodes. 62 00:04:57,370 --> 00:05:03,460 However, in large clusters you may chose to separate the ETCD clusters from the master node to its 63 00:05:03,550 --> 00:05:05,380 own cluster nodes. 64 00:05:05,410 --> 00:05:09,100 We will discuss more about the different topologies for that in the upcoming lecture 65 00:05:09,130 --> 00:05:14,080 when we talk about high availability setup. Well those are some of the considerations for designing a 66 00:05:14,080 --> 00:05:15,360 kubernetes cluster. 67 00:05:15,370 --> 00:05:20,490 Refer to the links in the references section for more details and some interesting reads. 68 00:05:20,560 --> 00:05:22,040 Well that's it for this lecture. 69 00:05:22,180 --> 00:05:28,420 Before you go from our certification exam standpoint there's really nothing much you need to remember 70 00:05:28,420 --> 00:05:29,200 from this section. 71 00:05:29,200 --> 00:05:34,540 You don't have to memorize the numbers that we discussed as these are available in the documentation 72 00:05:34,540 --> 00:05:41,660 page well we'll get into more interesting topics in the upcoming lectures in this section where we will 73 00:05:41,660 --> 00:05:46,530 provision an actual cluster from scratch by ourselves. 74 00:05:46,940 --> 00:05:48,820 Well that's it for this lecture. 75 00:05:48,860 --> 00:05:50,000 I'll see you in the next election.