{"id":25667,"date":"2026-11-22T10:49:10","date_gmt":"2026-11-22T10:49:10","guid":{"rendered":"https:\/\/atalnetworks.com\/?p=25667"},"modified":"2026-11-22T10:49:10","modified_gmt":"2026-11-22T10:49:10","slug":"kubernetes-on-bare-metal","status":"publish","type":"post","link":"https:\/\/atalnetworks.com\/ko\/blog\/kubernetes-on-bare-metal\/","title":{"rendered":"Kubernetes on Bare Metal: Architecture and Pitfalls"},"content":{"rendered":"<p><span style=\"font-weight: 400;\">Kubernetes on bare metal runs cluster nodes directly on physical servers, without a hypervisor between Kubernetes and the hardware. It gives your team direct control of CPU, memory, storage, and network interfaces. It also leaves your team responsible for parts a cloud platform often supplies: a stable control plane address, external service IPs, persistent storage, and node replacement.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Before you order dedicated servers, answer four questions: How will the API stay reachable? How will users reach applications? Where will data live after a host fails? How will you restore the cluster? The answers shape the server count, network setup, and operating cost.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">What are bare metal Kubernetes?<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">A bare metal Kubernetes node is a physical server running an operating system, a supported container runtime, and the Kubernetes node services. The control plane manages the cluster. Worker nodes run application Pods. A container network interface, or CNI, connects Pods across nodes.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">\u201cBare metal\u201d describes how Kubernetes uses the hardware. \u201cDedicated server\u201d describes who has exclusive use of that server. You can rent a dedicated server and install Kubernetes on its OS to create a bare metal node. If you install a hypervisor first and run Kubernetes inside virtual machines, your Kubernetes nodes are VMs, even though you rent the entire physical machine.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">An on-premises cluster sits in an organization&#8217;s own facility. A bare metal cluster can sit in a provider&#8217;s data center. Those terms describe different things.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">What you gain and what you take on<\/span><\/h3>\n<table>\n<thead>\n<tr>\n<th><b>\uc120\ud0dd<\/b><\/th>\n<th><b>Kubernetes on bare metal<\/b><\/th>\n<th><b>Kubernetes in VMs<\/b><\/th>\n<th><b>Managed Kubernetes<\/b><\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><span style=\"font-weight: 400;\">\ud558\ub4dc\uc6e8\uc5b4 \uc561\uc138\uc2a4<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Direct access to the host and its devices<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Access depends on the hypervisor and guest setup<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Depends on the provider&#8217;s node types<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">\uc6a9\ub7c9 \ubcc0\uacbd<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Add or replace physical servers<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Add VMs if host capacity exists<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Add nodes through the provider&#8217;s service, within its limits<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Control plane<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Your team runs it, unless a separate service manages it<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Your team may run it<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Provider usually runs it<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Load balancing and disks<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Your team must choose compatible services<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Depends on the hosting platform<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Often supplied through platform integrations<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Main operational cost<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Hardware, network, storage, and staff time<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Those costs plus a VM layer<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Service charges plus the work your team still owns<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><span style=\"font-weight: 400;\">Bare metal suits teams with steady workloads, clear hardware needs, and people who can operate the cluster. Direct GPU access and sustained use of dedicated CPU or storage can be good reasons to test it. A managed service may fit better if the team needs rapid capacity changes or does not want to run the control plane. Neither setup is always faster or cheaper. Measure your own workload and account for hardware, transfer, support, software, spare capacity, and staff time.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">A reference architecture for a production cluster<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">A production design should keep the Kubernetes API reachable, give applications enough capacity after a node loss, and make data recovery possible. Here is an example starting point, not a fixed minimum for every application:<\/span><\/p>\n<table>\n<thead>\n<tr>\n<th><b>Part<\/b><\/th>\n<th><b>Example design<\/b><\/th>\n<th><b>What it solves<\/b><\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><span style=\"font-weight: 400;\">Control plane<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Three physical nodes with one etcd member on each<\/span><\/td>\n<td><span style=\"font-weight: 400;\">One failed member leaves a two-member majority<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">API endpoint<\/span><\/td>\n<td><span style=\"font-weight: 400;\">A stable address that can reach healthy API servers<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Workers and administrators retain a route to the control plane<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Workers<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Two or more physical nodes, sized for the workload<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Pods have a place to run outside the control plane<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">\ud638\uc2a4\ud2b8 \ub124\ud2b8\uc6cc\ud06c<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Defined private paths for node traffic, with firewall rules<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Nodes can reach required cluster services<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Pod network<\/span><\/td>\n<td><span style=\"font-weight: 400;\">One CNI chosen for the cluster&#8217;s routing and policy needs<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Pods can communicate across hosts<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">External traffic<\/span><\/td>\n<td><span style=\"font-weight: 400;\">A service IP plan and an ingress or Gateway implementation where needed<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Clients can reach the applications<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">\uc2a4\ud1a0\ub9ac\uc9c0<\/span><\/td>\n<td><span style=\"font-weight: 400;\">A selected local, shared, or replicated storage design<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Stateful Pods have defined failure behavior<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">\ud68c\ubcf5<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Off-cluster etcd snapshots and application data backups<\/span><\/td>\n<td><span style=\"font-weight: 400;\">The team can rebuild after a broader failure<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><span style=\"font-weight: 400;\">Three physical servers can run both control plane components and application workloads if the chosen distribution permits it and you reserve capacity for both. Separate workers make capacity and maintenance easier to reason about. A three-node lab may be useful, but do not call it highly available until you have tested the API, application, storage, and network paths during a host failure.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">A drawing with three boxes also misses shared infrastructure. If all three servers depend on one switch, power feed, rack, or site, that shared component can interrupt the whole cluster. Ask the provider what physical separation it can supply. Do not spread etcd across distant sites without checking network latency and the effect of a link failure.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Control plane, etcd quorum, and the API address<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">The Kubernetes API server accepts changes to cluster state, while etcd stores that state. In a common kubeadm design, each of three control plane servers runs an etcd member. Two members form a majority after one member fails. With two etcd members, losing one leaves no majority.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">A majority alone does not keep the API reachable. The address used by administrators and workers must still lead to a healthy API server. For a kubeadm high availability setup, the configured <\/span><span style=\"font-weight: 400;\">controlPlaneEndpoint<\/span><span style=\"font-weight: 400;\"> must match a working load balancer address. Plan that address before creating the cluster. Place its availability outside the fate of any one application worker.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">There are two common etcd layouts. <\/span><b>Stacked etcd<\/b><span style=\"font-weight: 400;\"> places etcd beside control plane components on the same nodes and keeps the hardware count smaller. <\/span><b>External etcd<\/b><span style=\"font-weight: 400;\"> runs etcd on separate nodes and adds machines and operational work. Choose on failure isolation, support, and staffing needs. Neither option removes the need for snapshots.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">If etcd loses its majority, existing application processes may continue to run for a time, but the control plane cannot reliably update cluster state or schedule replacement Pods. Recovery is an operations procedure, not a reason to skip backups.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Worker count and the N-1 test<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Size worker capacity for the loss of a node, not only for a normal day. Start with actual application CPU and memory requests. Add space for the OS, kubelet, CNI, storage agents, logging, and other DaemonSets. Then check whether replicas can be placed on distinct failure domains.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Consider an illustrative cluster of three workers. After system reservations, assume each has room for <\/span><b>24 CPU request units<\/b><span style=\"font-weight: 400;\">. Your application Pods request <\/span><b>45 units<\/b><span style=\"font-weight: 400;\"> in total. Normally the cluster has 72 units. After one worker fails, it has 48. On CPU requests alone, the remaining nodes have three units to spare. That is a narrow margin. Memory limits, per-node placement rules, local disks, and traffic may still stop the Pods from returning to service.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">This is a capacity method, not a benchmark or a suggested CPU size. Repeat it with your own requests and observed peaks. Use topology spread rules or placement rules where needed so replicas do not all land on one server.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">Networking: separate four different paths<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">Bare metal Kubernetes networking starts with the physical network. Kubernetes cannot fix an IP routing or firewall rule that prevents the hosts from reaching each other. Draw the paths before selecting a CNI or a load balancer:<\/span><\/p>\n<ol>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Host network:<\/b><span style=\"font-weight: 400;\"> Physical interfaces, private links, VLANs, routes, DNS, and firewalls connect servers.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Pod network:<\/b><span style=\"font-weight: 400;\"> A CNI gives Pods addresses and a way to communicate across nodes.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Control plane endpoint:<\/b><span style=\"font-weight: 400;\"> A stable address takes API traffic to healthy control plane nodes.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Application entry:<\/b><span style=\"font-weight: 400;\"> A service IP and, for HTTP applications, an ingress controller or Gateway implementation take client traffic to Pods.<\/span><\/li>\n<\/ol>\n<p><span style=\"font-weight: 400;\">Pod and Service address ranges must not overlap with host, private network, or provider-routed address ranges. Check the CNI&#8217;s requirements for MTU, encapsulation, routing, and NetworkPolicy. An overlay that adds packet headers may need a smaller Pod network MTU than the physical interface. A packet-size issue may show up as intermittent timeouts rather than a complete network failure.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Use the Kubernetes ports and protocols reference when writing firewall rules. Allow only the paths your chosen components need. Keep the Kubernetes API and out-of-band server management off unrestricted public access.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">An API endpoint is not an application load balancer<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">These three terms describe separate jobs:<\/span><\/p>\n<table>\n<thead>\n<tr>\n<th><b>\uad6c\uc131 \uc694\uc18c<\/b><\/th>\n<th><b>\ubcf8\uc5c5<\/b><\/th>\n<th><b>\ubb3c\uc5b4\ubcfc \uc9c8\ubb38<\/b><\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><span style=\"font-weight: 400;\">API endpoint<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Reach a healthy Kubernetes API server<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Does it work before cluster add-ons start and after a control plane node fails?<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">\uc11c\ube44\uc2a4<\/span><span style=\"font-weight: 400;\"> of type <\/span><span style=\"font-weight: 400;\">LoadBalancer<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Give an application a reachable service IP<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Who assigns and announces that IP on this network?<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Ingress controller or Gateway implementation<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Route application traffic by host, path, or other supported rules<\/span><\/td>\n<td><span style=\"font-weight: 400;\">How does traffic reach that controller in the first place?<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><span style=\"font-weight: 400;\">A Kubernetes <\/span><span style=\"font-weight: 400;\">LoadBalancer<\/span><span style=\"font-weight: 400;\"> Service needs an implementation. Without one, the service may wait for an external address. Installing an ingress controller does not, on its own, provide that address. MetalLB can provide application service IPs on suitable networks, but it depends on a running cluster. Do not assume that adding MetalLB also solves the first bootstrap of a stable API endpoint.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">MetalLB L2 or BGP?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">MetalLB can announce service addresses in layer 2 mode or through BGP. The right option depends on what the host network and provider allow.<\/span><\/p>\n<table>\n<thead>\n<tr>\n<th><b>\ubc29\ubc95<\/b><\/th>\n<th><b>Network requirement<\/b><\/th>\n<th><b>Main tradeoff<\/b><\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><span style=\"font-weight: 400;\">\ub808\uc774\uc5b4 2<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Nodes and gateway must support the required local address discovery for the chosen IPs<\/span><\/td>\n<td><span style=\"font-weight: 400;\">One node receives incoming traffic for a given service IP; a failover may pause traffic while neighbor caches update<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">BGP<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Approved routing sessions with suitable routers and address policy<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Routes can bring traffic to multiple nodes, but routing setup and failure handling need more care<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Provider or external load balancer<\/span><\/td>\n<td><span style=\"font-weight: 400;\">A supported upstream service or routed IP arrangement<\/span><\/td>\n<td><span style=\"font-weight: 400;\">The provider&#8217;s features and support terms set the available design<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><span style=\"font-weight: 400;\">Do not take a block of assigned public IPs and assume it is an L2 MetalLB pool. A provider may route those IPs to one server or forbid the required address announcements. Ask whether the intended service IPs can move between hosts, whether the servers share the needed layer 2 segment, and whether BGP peering is available for the chosen plan. If the answer is no, use a supported routed or external load-balancer design.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">For high traffic, note the layer 2 limit: a single node receives incoming packets for each service IP before Kubernetes sends traffic onward to Pods. A 10 Gbps server port does not make a single service flow run at 10 Gbps, and a port rating does not state the included monthly data transfer.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">Storage: where does the data go when a host fails?<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">Kubernetes can start a replacement Pod, but it cannot make lost local bytes appear on a different server. Choose the data model before placing a database or file service in the cluster.<\/span><\/p>\n<table>\n<thead>\n<tr>\n<th><b>\uc2a4\ud1a0\ub9ac\uc9c0 \uc120\ud0dd<\/b><\/th>\n<th><b>\uc88b\uc740 \ud54f<\/b><\/th>\n<th><b>Host failure behavior<\/b><\/th>\n<th><b>Work your team owns<\/b><\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><span style=\"font-weight: 400;\">Local SSD or NVMe PersistentVolume<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Workloads that can rebuild data, or applications that manage their own replication<\/span><\/td>\n<td><span style=\"font-weight: 400;\">The volume stays tied to its host; a Pod elsewhere cannot use those bytes automatically<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Disk health, restore, and application replication<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">External shared storage through a supported CSI driver<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Workloads that need a volume available to another node<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Depends on the storage system, network, access mode, and driver<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Shared storage availability, access control, backups<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Software-replicated storage across nodes<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Workloads whose data can tolerate the chosen replication cost and failure model<\/span><\/td>\n<td><span style=\"font-weight: 400;\">A replica may serve data after a node fails if health and placement were sound<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Storage nodes, network capacity, quorum, recovery, backups<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><span style=\"font-weight: 400;\">A local PV includes node placement rules so Kubernetes schedules its consumer where the volume exists. The <\/span><span style=\"font-weight: 400;\">WaitForFirstConsumer<\/span><span style=\"font-weight: 400;\"> binding mode lets scheduling consider the Pod&#8217;s placement before it binds the volume. It does not copy the data or make local storage available on another machine. A failed database host can therefore leave its replacement Pod waiting until you repair the host or restore data using the application&#8217;s recovery plan.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Software replication also consumes network and disk resources. If storage replicas and application replicas all sit on one physical host or one rack, the expected failure protection may not exist. For a database with its own replication, compare application-managed replicas with a storage layer before running both. Either choice still needs backups.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Two backups, two different jobs<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">\ud558\ub098\uc758 <\/span><b>etcd snapshot<\/b><span style=\"font-weight: 400;\"> records Kubernetes cluster state, such as object definitions and configuration. It is not a copy of the contents of your usual database PVC. An <\/span><b>application data backup<\/b><span style=\"font-weight: 400;\"> records the data your workload must restore. Use a method suited to the database or file system, and check consistency during a restore.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Store recovery material outside the failed cluster, protect access to it, and keep the keys or credentials required to read it. A useful recovery test rebuilds the control plane in an isolated environment, restores the needed cluster state, then restores application data and checks the application. Record the time it took and how much data was missing. Those observations give you a measured recovery time objective, or RTO, and recovery point objective, or RPO. A backup schedule alone cannot prove either target.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">RAID can keep a server running after some disk failures when configured for redundancy. It does not restore deleted files, repair bad application writes, or replace an off-cluster backup.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">Common pitfalls and first checks<\/span><\/h2>\n<table>\n<thead>\n<tr>\n<th><b>\uc9d5\ud6c4<\/b><\/th>\n<th><b>\uac00\ub2a5\ud55c \uc6d0\uc778<\/b><\/th>\n<th><b>First check and next action<\/b><\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><span style=\"font-weight: 400;\">A <\/span><span style=\"font-weight: 400;\">LoadBalancer<\/span><span style=\"font-weight: 400;\"> Service has no external IP<\/span><\/td>\n<td><span style=\"font-weight: 400;\">No service IP implementation or address pool<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Check the service events, controller health, and approved IP pool. Confirm upstream routing before assigning addresses.<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Pods on different hosts cannot communicate<\/span><\/td>\n<td><span style=\"font-weight: 400;\">CNI, firewall, route, or MTU issue<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Check Pod IPs, node routes, CNI logs, and a cross-node packet test. Change one layer at a time.<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">kubectl<\/span><span style=\"font-weight: 400;\"> stops working after one control plane failure<\/span><\/td>\n<td><span style=\"font-weight: 400;\">The API endpoint lived on one server or routes did not fail over<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Test the endpoint from a worker and a separate admin host. Check healthy API backends and the VIP owner.<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">A database Pod stays Pending after a worker fails<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Local PV is tied to the lost host or no suitable volume exists<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Inspect the PVC, PV node placement rules, Pod events, and restore plan. Do not force a volume onto a different host without its data.<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Maintenance drain never finishes<\/span><\/td>\n<td><span style=\"font-weight: 400;\">A PodDisruptionBudget or workload placement rule blocks eviction<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Inspect the budget and available replicas. Add capacity or fix the application before changing protection rules.<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Replacement Pods will not schedule<\/span><\/td>\n<td><span style=\"font-weight: 400;\">No remaining CPU, memory, node placement rules match, or suitable storage<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Check scheduler events and repeat the N-1 calculation for every scarce resource.<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Service IP responds only from some networks<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Address announcement or upstream neighbor cache issue<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Check the actual route, ARP or NDP behavior, and provider network design.<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><span style=\"font-weight: 400;\">These are starting points, not proof of a single cause. Record symptoms and component logs before changing routing or storage settings.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">A PodDisruptionBudget helps control <\/span><i><span style=\"font-weight: 400;\">voluntary<\/span><\/i><span style=\"font-weight: 400;\"> eviction, such as a planned node drain. It cannot stop a power failure or make a failed node&#8217;s local disk available elsewhere. Run both planned and unplanned failure tests.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">How to deploy in the right order<\/span><\/h2>\n<ol>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Set service and recovery goals.<\/b><span style=\"font-weight: 400;\"> List acceptable downtime, data loss, expected traffic, and who is on call. Decide whether the cluster is a lab or production service.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Confirm physical resources.<\/b><span style=\"font-weight: 400;\"> Choose server roles, CPU, memory, disks, private network, management access, address pools, and failure domains. Ask how quickly a failed server can be replaced.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Choose a supported software set.<\/b><span style=\"font-weight: 400;\"> Pick an OS, Kubernetes distribution, container runtime, CNI, and storage driver whose versions work together. With kubeadm, check current container runtime and cgroup guidance before configuring hosts.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Provide the API endpoint and build the control plane.<\/b><span style=\"font-weight: 400;\"> Confirm that the endpoint reaches healthy servers. Select stacked or external etcd, then take a first snapshot.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Install and test the Pod network.<\/b><span style=\"font-weight: 400;\"> Join workers, verify cluster DNS, and test Pod traffic across hosts. Confirm firewalls and MTU before adding user traffic.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Add service exposure and storage.<\/b><span style=\"font-weight: 400;\"> Test a Service IP from outside the cluster, then create a test PVC and confirm its placement and failure behavior.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><b>Deploy a sample workload and rehearse failures.<\/b><span style=\"font-weight: 400;\"> Test a planned drain, sudden node loss, control plane loss, service IP movement, and data restoration. Write down what worked and what needs to change.<\/span><\/li>\n<\/ol>\n<p><span style=\"font-weight: 400;\">A one-server cluster is fine for learning. It is one physical failure domain. A two-server installation can teach you about joining nodes, but two stacked etcd members do not tolerate the loss of either member. For production, follow the supported topology and test the service&#8217;s actual recovery, not just the cluster&#8217;s node count.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">Recovery tests to run before production traffic<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">Use a worksheet that separates the expected result from what you observe. Keep evidence such as timestamps, health checks, and restore logs.<\/span><\/p>\n<table>\n<thead>\n<tr>\n<th><b>\ud14c\uc2a4\ud2b8<\/b><\/th>\n<th><b>What to do in a safe test window<\/b><\/th>\n<th><b>What a passing result looks like<\/b><\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><span style=\"font-weight: 400;\">Planned worker maintenance<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Drain one worker under the current disruption rules<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Required application replicas keep serving; workloads move or the drain stops for a documented reason<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Sudden worker loss<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Isolate or shut down a test worker<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Services remain usable within your target; spare nodes accept Pods that do not depend on lost local data<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">One control plane loss<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Stop one control plane host<\/span><\/td>\n<td><span style=\"font-weight: 400;\">The stable API address still reaches a healthy API server and etcd retains a majority<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Service IP failover<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Stop the announcing node in a test service<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Clients reach the same service IP again within the measured recovery window<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Storage host loss<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Remove one storage node or local-volume host from a test workload<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Data behavior matches the selected storage design; the recovery steps work<\/span><\/td>\n<\/tr>\n<tr>\n<td><span style=\"font-weight: 400;\">Backup restore<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Restore etcd and application data in a separate test environment<\/span><\/td>\n<td><span style=\"font-weight: 400;\">Cluster objects and application records pass integrity checks<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><span style=\"font-weight: 400;\">Use actual measured times in your incident plan. If an unplanned failure takes longer than the application can accept, change the design or recovery target before launch. Re-run these checks after network changes, storage changes, and major cluster upgrades.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">What to ask a dedicated server provider<\/span><\/h2>\n<p><span style=\"font-weight: 400;\">The network contract matters as much as the hardware quote. Send a provider the intended topology and ask for written answers to these points:<\/span><\/p>\n<ul>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Can the servers use a private network, and is it the same local network segment or routed between sites?<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Can a service IP move between servers? Are layer 2 address announcements or BGP sessions supported for our plan?<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Which public IPs can we use for application services, and how are they routed?<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Can you place control plane nodes across distinct racks, power paths, or facilities? What separation can you confirm?<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">What are the CPU, RAM, disk, network port, and included transfer terms for each node?<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">What out-of-band console or KVM access is available if the OS or network fails?<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">How are failed disks and servers replaced, and who performs the replacement?<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">Which tasks belong to the provider and which belong to us: host OS, Kubernetes, CNI, service IPs, storage, backups, and restore?<\/span><\/li>\n<\/ul>\n<p><span style=\"font-weight: 400;\">Atal Networks\ub294 <\/span><a href=\"https:\/\/atalnetworks.com\/ko\/bare-metal-servers\/\"><span style=\"font-weight: 400;\">\ubca0\uc5b4\uba54\ud0c8 \uc11c\ubc84<\/span><\/a><span style=\"font-weight: 400;\"> \ubc0f <\/span><a href=\"https:\/\/atalnetworks.com\/ko\/dedicated-servers\/\"><span style=\"font-weight: 400;\">\uc804\uc6a9 \uc11c\ubc84 \uc635\uc158<\/span><\/a><span style=\"font-weight: 400;\"> with physical resource control and multiple network choices. We should confirm the private network topology, IP routing, management access, and support terms for the exact servers and location you select. A dedicated server purchase does not by itself include a managed Kubernetes control plane or application data recovery.<\/span><\/p>\n<h2><span style=\"font-weight: 400;\">\uc790\uc8fc \ubb3b\ub294 \uc9c8\ubb38<\/span><\/h2>\n<h3><span style=\"font-weight: 400;\">Can Kubernetes run on one dedicated server?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Yes. One dedicated server can run a learning cluster, and some applications accept a single-node deployment. If that host fails or needs maintenance, its control plane and local workloads may stop together. Use the single-node setup only when that service risk fits your needs.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Are three physical servers enough for high availability?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Three control plane servers can keep etcd&#8217;s majority after one member fails. They do not guarantee application availability. Check the API endpoint, worker capacity, replica placement, storage, and any shared switch or power path. A recovery test is the way to learn whether the whole service survives one host loss.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Does MetalLB replace the control plane load balancer?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Do not assume so. MetalLB supplies IP addresses and announcements for Kubernetes Services after the relevant cluster components run. A stable API endpoint must be designed for cluster bootstrap and control plane failures. You may use another supported VIP or external load-balancing method for that endpoint.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Do I need BGP for Kubernetes on bare metal?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">No. BGP is one option for announcing service IP routes. MetalLB layer 2 mode can work where the network supports the required address discovery. A provider load balancer or routed IP design may also fit. Ask the host which methods it supports before choosing service address ranges.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Can a Pod with local NVMe storage move to another server?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">A replacement Pod can be scheduled elsewhere only if its storage needs can be met there. A Kubernetes local PersistentVolume points to data on a particular node. Moving the Pod alone does not move that data. Use application replication, a suitable shared or replicated storage system, or a restore procedure.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">Is bare metal Kubernetes always faster than Kubernetes in VMs?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">No. Removing a hypervisor can help workloads that need direct hardware access, but performance also depends on CPU placement, storage, network design, and the application. Compare representative workloads under the same settings. A general performance percentage from another environment does not predict your result.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">What should I back up besides etcd?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Back up application data, database state, needed secrets or recovery keys, and the configuration required to rebuild cluster add-ons and external infrastructure. An etcd snapshot covers cluster state, but it does not replace a database backup. Test that you can restore both in a separate environment.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">What happens when a bare metal worker fails?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Kubernetes can place replacement Pods on healthy workers if the control plane can act and those workers have capacity and suitable storage. Existing local volumes stay with the failed node. A good failure plan includes spare capacity, replica placement, storage recovery, and a server replacement procedure.<\/span><\/p>\n<h3><span style=\"font-weight: 400;\">How much private network capacity does a cluster need?<\/span><\/h3>\n<p><span style=\"font-weight: 400;\">There is no universal number. Measure traffic between Pods, storage replicas, backup systems, and clients. Size for peaks and for the extra traffic a node failure creates. Verify the port speed, network path, included transfer, and any limits in the provider&#8217;s plan; they are different measures.<\/span><\/p>","protected":false},"excerpt":{"rendered":"<p>Kubernetes on bare metal runs cluster nodes directly on physical servers, without a hypervisor between Kubernetes and the hardware. It [&hellip;]<\/p>\n","protected":false},"author":2,"featured_media":25668,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"default","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"set","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[1],"tags":[],"class_list":["post-25667","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-enterprise-grade-server"],"acf":[],"_links":{"self":[{"href":"https:\/\/atalnetworks.com\/ko\/wp-json\/wp\/v2\/posts\/25667","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/atalnetworks.com\/ko\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/atalnetworks.com\/ko\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/atalnetworks.com\/ko\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/atalnetworks.com\/ko\/wp-json\/wp\/v2\/comments?post=25667"}],"version-history":[{"count":1,"href":"https:\/\/atalnetworks.com\/ko\/wp-json\/wp\/v2\/posts\/25667\/revisions"}],"predecessor-version":[{"id":25669,"href":"https:\/\/atalnetworks.com\/ko\/wp-json\/wp\/v2\/posts\/25667\/revisions\/25669"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/atalnetworks.com\/ko\/wp-json\/wp\/v2\/media\/25668"}],"wp:attachment":[{"href":"https:\/\/atalnetworks.com\/ko\/wp-json\/wp\/v2\/media?parent=25667"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/atalnetworks.com\/ko\/wp-json\/wp\/v2\/categories?post=25667"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/atalnetworks.com\/ko\/wp-json\/wp\/v2\/tags?post=25667"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}