# Kubectl View Allocations > Use when inspecting Kubernetes resource allocation with kubectl view-allocations. For allocation questions, use kubectl view-allocations first and do not replace it with native kubectl get/describe/top/jsonpath queries. Covers cluster/node/namespace/pod requests, limits, allocatable, free capacity, ALL resources including GPU and custom resources (not just CPU/memory), taint-filtered node views, tree-view grouping via -g, multi-column sort via --sort, CSV exports, metrics-server utilization via kubectl view-allocations -u, kubeconfig/context access checks, and allocation-focused troubleshooting. Trigger for CPU, memory, GPU, pod capacity, overcommit, missing requests/limits, tainted nodes, or comparing requested/limit/utilization data in a Kubernetes cluster. WARNING: default output excludes tainted nodes — results may be empty or partial on clusters where all nodes have taints (e.g. control-plane-only, GPU, dedicated workload nodes). - Skill: `davidb/kubectl-view-allocations` (Agent Skill, multi-file: 3 files) - Install (CLI): `npx skillmds@latest add davidb/kubectl-view-allocations` - Raw SKILL.md: https://api.skillmd.com/api/skills/davidb/kubectl-view-allocations/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: DevOps & Infra - Author: davidB (https://skillmd.com/u/davidb) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/davidb/kubectl-view-allocations --- # Kubectl View Allocations Use `kubectl view-allocations` as the required first tool for this repository's allocation workflow. Do not answer allocation questions by reconstructing the same data with native `kubectl get`, `kubectl describe`, `kubectl top`, JSONPath, or ad hoc scripts unless `kubectl view-allocations` cannot run and the user accepts a fallback. `kubectl view-allocations` reports Kubernetes allocations from manifests and node allocatable resources. Key differentiators over `kubectl top` and similar tools: - **All resources**: cpu, memory, GPU, and any custom resource — not limited to cpu/memory - **Tree view**: group and aggregate by resource, node, namespace, or pod in any combination via `-g` - **Live utilization**: optional via `-u` (requires metrics-server), same as `kubectl top` but richer > **Default taint filtering**: only nodes *without* taints are shown. On clusters where all nodes carry taints (control-plane-only, GPU nodes, dedicated workloads), the output may be empty or partial. Use `--ignore-taints` to include tainted nodes. ## Workflow 1. Clarify scope only when it is missing and risky: kube context, namespace, resource name, node selector, or whether tainted nodes should be included. 2. Run the first data-gathering query with `kubectl view-allocations`. 3. Run a narrow `kubectl view-allocations` query first, then broaden only if the result looks incomplete. 4. Prefer allocation views for capacity planning and scheduling questions; use `kubectl view-allocations -u` only for "currently using" questions. 5. Explain output in terms of `Requested`, `Limit`, `Allocatable`, and `Free`. 6. If `kubectl view-allocations` fails, fix the command invocation path first: installation/PATH, kube context, kubeconfig, permissions, metrics availability, or `--precheck`. ## Tool Policy - Use `kubectl view-allocations` for allocation facts and summaries. - Use `kubectl` only for narrow support tasks: checking the current context, refreshing auth through `kubectl view-allocations --precheck`/`kubectl cluster-info`, confirming metrics-server availability after `kubectl view-allocations -u` fails, or fetching metadata that `kubectl view-allocations` cannot show and the user explicitly needs. - Do not start with `kubectl top` for utilization questions covered by `kubectl view-allocations -u`. - Do not start with `kubectl get nodes/pods -o json` to manually sum requests, limits, allocatable resources, or taints. - If `kubectl view-allocations` is missing, guide the user to install it from the davidB/kubectl-view-allocations README before using fallback allocation commands. - If falling back from `kubectl view-allocations` is unavoidable, tell the user that the fallback is no longer using this skill's primary tool and why. ## Missing Tool When `kubectl view-allocations` is not installed or not on PATH: 1. Tell the user the primary tool is missing. 2. Offer the README-supported install options and ask which they prefer: - krew: `kubectl krew install view-allocations` - cargo: `cargo install kubectl-view-allocations` - release script: `curl https://raw.githubusercontent.com/davidB/kubectl-view-allocations/master/scripts/getLatest.sh | bash` 3. After installation, run `kubectl view-allocations --help` or a narrow allocation query to confirm it works. 4. Do not silently replace the missing tool with native `kubectl` allocation calculations. ## Command Patterns Use these starting points: ```sh kubectl view-allocations -g resource kubectl view-allocations -g resource,node kubectl view-allocations -g resource,node,pod kubectl view-allocations -g namespace kubectl view-allocations -n NAMESPACE -g resource,node,pod kubectl view-allocations -r cpu,memory -g resource,namespace kubectl view-allocations -r gpu kubectl view-allocations --ignore-taints -g resource,node kubectl view-allocations -u -r cpu,memory kubectl view-allocations -o csv -g resource,node,pod # Sort: show most-requested resources first (default) kubectl view-allocations -s "requested DESC" # Sort by free capacity ascending (find most constrained nodes first) kubectl view-allocations -g resource,node -s "free ASC" # Multi-column sort kubectl view-allocations -s "usage DESC, requested DESC" ``` Add `--context CONTEXT` and `--kubeconfig PATH` when the user names a cluster or config file. Use `--precheck` when kube auth tokens may need refreshing because it runs `kubectl cluster-info` first. Read `references/command-cookbook.md` for command recipes, interpretation notes, and troubleshooting branches. ## Interpretation - `Requested`: sum of pod container resource requests from manifests. - `Limit`: sum of pod container limits from manifests. - `Allocatable`: node allocatable capacity. - `Free`: `Allocatable - max(Limit, Requested)` by default; use `--used-mode only-request` to compute free capacity from requests only. - `Utilization`: live CPU/memory usage from Metrics API, shown only with `-u`. - `__`: value is absent or not meaningful at that grouping level, not necessarily zero. Default grouping is `resource,node,pod`; `resource` is always prepended when omitted. Default node filtering excludes tainted nodes — **if output is empty or missing nodes, try `--ignore-taints`**. Use `--ignore-taints` with no value to include all nodes, or with taint keys/key-value pairs to include specific tainted nodes. `--sort` / `-s` accepts SQL-like syntax: `col [ASC|DESC]`, comma-separated for multi-column. Valid columns: `usage`/`utilization`, `requested`, `limits`/`limit`, `allocatable`, `free`, `name`. Default: `"usage DESC, requested DESC, limits DESC, name ASC"`. Sort applies within sibling groups at each tree level. ## Safety Treat `kubectl view-allocations` as read-only. It lists nodes, pods, and optionally pod metrics. Do not deploy, delete, scale, or modify Kubernetes resources as part of this skill unless the user separately asks for that action.