{"id":5529,"date":"2026-11-16T14:50:00","date_gmt":"2026-11-16T14:50:00","guid":{"rendered":"https:\/\/www.devopssupport.in\/blog\/?p=5529"},"modified":"2026-11-19T04:50:00","modified_gmt":"2026-11-19T04:50:00","slug":"kubernetesops-faq-what-managed-cluster-support-covers","status":"publish","type":"post","link":"https:\/\/www.devopssupport.in\/blog\/kubernetesops-faq-what-managed-cluster-support-covers\/","title":{"rendered":"KubernetesOps FAQ: What Managed Cluster Support Should Cover"},"content":{"rendered":"<p>Support teams hear the same questions whenever a client moves workloads onto Kubernetes. Who watches the cluster itself? What counts as an incident and what is routine maintenance? What will never be in scope? What follows leans on the Atlas&#8217;s account of the discipline and is written for people comparing support arrangements or drafting a service description. Each answer is brief and ends with something you can check.<\/p>\n<h2>What does KubernetesOps support actually include?<\/h2>\n<p>The Atlas&#8217;s scope reads like a table of contents: the control plane, the machines beneath it, bundled extras, version changes, how workloads are tuned, scaling, capacity and fault-finding. A support contract worth signing maps onto each heading in turn. If a proposal talks mostly about dashboards and says nothing about upgrade cadence or restores, ask what is missing. <strong>Check:<\/strong> does the scope document name who owns upgrades, certificate rotation and state-store backups? The Atlas assigns those to the cluster caretaker.<\/p>\n<h2>What is deliberately left out?<\/h2>\n<p>The entry draws the border clearly. Security hardening, container image and runtime concerns, delivery pipelines, treating the platform as an internal product and deep storage work belong to neighbouring practices, as does release strategy. That is not a gap in the model; it is a boundary to write into the contract so nobody assumes coverage that was never promised. For a fuller read, see <a href=\"https:\/\/www.xopsschool.com\/kubernetesops\/\">the Ops Atlas page on the KubernetesOps boundary and playbooks<\/a>.<\/p>\n<h2>How often will the cluster need upgrading?<\/h2>\n<p>New minor versions arrive about three times annually, with patches stretching across roughly fourteen months. In mid-October 2026 three branches were current, and a fourth, older line was in its final maintenance weeks. Upgrades are sequential and follow a fixed order of components. <strong>Check:<\/strong> ask for a twelve-month calendar showing each hop, plus the target for version lag. The Atlas proposes one minor or fewer in production.<\/p>\n<h2>What happens if an upgrade goes wrong?<\/h2>\n<p>Rollback is narrower than clients expect. Self-managed clusters built with the common bootstrap tool can re-run the process safely and keep backups of changed files, yet no switch undoes a minor-version step on the API server. On the managed platform the Atlas examines, upgrades cannot be paused, revert on failed health checks and let you step back one minor inside a week. <strong>Check:<\/strong> insist on a written recovery path, and on evidence that the restore procedure has been exercised. A 2023 public outage shows what happens otherwise: recovery from backup took about five hours and the operator acknowledged that its restore steps were not fully current.<\/p>\n<h2>Why do pods get evicted when we have protection rules?<\/h2>\n<p>Disruption budgets govern voluntary actions like drains. They do not stop eviction caused by node pressure, and preemption honours them only as far as possible. Also, a budget that allows no disruption prevents routine node maintenance. <strong>Check:<\/strong> request a list of budgets and a statement on how unhealthy pods are treated during drains. The Atlas recommends letting them be evicted.<\/p>\n<h2>How should we think about resource requests?<\/h2>\n<p>Requests are the amounts the scheduler reserves, and it places pods by those figures whatever the pods really use. Autoscalers read them too. A support provider should report two ratios regularly: how much capacity is requested, and how much of the request is used. Without requests, CPU-based scaling cannot act at all. <strong>Check:<\/strong> ask what percentage of pods carry requests; the Atlas treats ninety-eight percent as the mark of a managed cluster.<\/p>\n<h2>Which defaults should our runbook mention?<\/h2>\n<p>A handful of timers explain many support tickets. Crash-looping containers back off from ten seconds up to five minutes. Pods on a node that stops reporting are tolerated for three hundred seconds before eviction. A rollout is declared stalled after six hundred seconds without progress. Knowing these figures turns &#8220;it has been stuck for ages&#8221; into a precise statement, and saves a round of escalation. Liveness probes deserve a note as well: they should look only at the process, since a probe that checks a downstream dependency can set off cascading restarts.<\/p>\n<h2>What should a monthly report contain?<\/h2>\n<p>Not a wall of graphs. The Atlas proposes a compact set of indicators. Expect how far behind you are on versions, the days left before support lapses, and how long an upgrade takes from decision to done. Add the share of capacity requested and the ratio of use to request, then unplaceable pods, the restart frequency of containers, API response times and whether a restore has been proven. Latency has a published reference: the scalability group&#8217;s service-level objectives mention single-object mutating calls finishing within one second at the 99th percentile. <strong>Check:<\/strong> every figure should have a trend and an owner, not just a value.<\/p>\n<h2>Who is accountable for what?<\/h2>\n<p>Write a responsibility table before the first incident, particularly for managed control planes, where the provider handles the API servers and state store but the client still carries the version treadmill, add-on inventory and tested rebuilds. The same clarity is behind the <a href=\"https:\/\/www.devopssupport.in\/blog\/devops-principles-for-effective-software-delivery\/\">principles of effective software delivery<\/a> we discuss elsewhere, and the argument for managing infrastructure through code, for people from an operations background, appears in <a href=\"https:\/\/www.devopssupport.in\/blog\/infrastructure-as-code-to-sre-why-operations-engineers-should-embrace-devops\/\">our article linking IaC to SRE habits<\/a>.<\/p>\n<h2>How do we judge whether the provider is improving things?<\/h2>\n<p>Look for movement, not snapshots. Version lag should shrink and then stay flat. The number of pods with requests should climb toward the agreed floor. Tickets caused by blocked drains should disappear after the first quarter. If the same cause appears three months running, the root cause was never fixed. <strong>Check:<\/strong> ask for a quarterly review that lists causes by category and shows which were eliminated.<\/p>\n<h2>One closing question: is KubernetesOps a service we can buy by that name?<\/h2>\n<p>Not as a standard. The Atlas notes that the label has no governing body, and that similarly named products exist. Buy outcomes: dated upgrades, honest requests, drilled restores. Name them in the contract, and the label becomes irrelevant.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Eight to ten questions clients ask about managed Kubernetes support: scope, upgrades, rollbacks, budgets, requests, defaults and monthly reporting.<\/p>\n","protected":false},"author":8,"featured_media":5530,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3948],"tags":[],"class_list":["post-5529","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-devops-support"],"_links":{"self":[{"href":"https:\/\/www.devopssupport.in\/blog\/wp-json\/wp\/v2\/posts\/5529","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.devopssupport.in\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.devopssupport.in\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.devopssupport.in\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/www.devopssupport.in\/blog\/wp-json\/wp\/v2\/comments?post=5529"}],"version-history":[{"count":1,"href":"https:\/\/www.devopssupport.in\/blog\/wp-json\/wp\/v2\/posts\/5529\/revisions"}],"predecessor-version":[{"id":5662,"href":"https:\/\/www.devopssupport.in\/blog\/wp-json\/wp\/v2\/posts\/5529\/revisions\/5662"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.devopssupport.in\/blog\/wp-json\/wp\/v2\/media\/5530"}],"wp:attachment":[{"href":"https:\/\/www.devopssupport.in\/blog\/wp-json\/wp\/v2\/media?parent=5529"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.devopssupport.in\/blog\/wp-json\/wp\/v2\/categories?post=5529"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.devopssupport.in\/blog\/wp-json\/wp\/v2\/tags?post=5529"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}