--- title: Triage your Kubernetes fleet source: https://docs.newrelic.com/docs/kubernetes-pixie/kubernetes-integration/understand-use-data/explore-k8s --- Modern Kubernetes environments have evolved into complex, **multi-cluster fleets**, but traditional observability tools often provide only a fractured, siloed view of individual clusters. Kubernetes monitoring in New Relic opens on a fleet-wide view that turns that complexity into clarity. This unified view helps platform engineers and SREs to: - **Unify fleet-wide observability** on a single dashboard that's compatible with clusters monitored by New Relic agents or OpenTelemetry. - **Accelerate root cause analysis** with a guided triage workflow. - **Reduce costs** by identifying wasted resources across the entire fleet. - **Empower developers** with an application-centric view for self-service. ![K8s landing page](https://docs.newrelic.com/images/k8s-cross-cluster.webp "K8s landing page") ## Access the UI [#access] This is the default Kubernetes experience in New Relic. To open it: Go to **[one.newrelic.com > All capabilities](https://one.newrelic.com/all-capabilities) > Kubernetes** ## Main components of the UI [#components] The UI is structured to provide both a high-level overview and deep-dive capabilities into your entire Kubernetes fleet. The main components include: - **Entity filters**: Filter and triage issues using **tags and values** at the **cluster or node level**. - **Cluster filter**: Filter to isolate the UI to a single or smaller set of Kubernetes clusters. - **Colored scorecards**: Display relevant high-level metrics to surface the most important issues that need attention across the entire fleet. - Click a scorecard to **order the table** in the context of that metric, surfacing the clusters and nodes with the most critical issues at the top. - It also enables a **line chart** to visualize the metric's evolution over the selected time range. - Select **`avg`**, **`p95`**, **`p99`**, or **`max`** to update the scorecards and line chart. The breakdown table always displays average values. - **Line chart**: Provides the **evolution of the metric** over the selected time frame for trend analysis. - **Table**: Lists all entities in the fleet and provides **detailed metric data** for each cluster under the five main [tabs](#tab). - **Health focus**: The table provides a filter to only list "unhealthy" entities to focus your attention where it is really needed. - **Search**: A text search bar allows you to filter table rows based on **free text** entered. - **Color-coding**: Like the scorecards, table cells are color-coded (Yellow, Red) to quickly draw attention to entities with issues that meet a specific [severity threshold](#tab). - **Column chooser**: The Workloads tab ships a lean default column set and lets you add or remove columns. Your column selection persists across sessions. - **Workload type selector**: Narrow the Workloads tab to a specific workload type, such as Deployments, DaemonSets, or StatefulSets. - **Feedback button**: Lets you easily share feedback on your experience with the UI. ## Multi-account fleet visibility [#multi-account] Clusters from the New Relic accounts you have access to appear together in the same fleet view, so monitoring a fleet that spans multiple accounts doesn't require switching account context. To see clusters from more than one account: 1. In the account picker, select **All accounts**. 2. Use the account filter in the entity filter bar to narrow the view to the specific accounts you need. ## Entity popovers [#popover] Selecting an entity in the table opens a popover that gives you context on that entity without leaving the fleet view: - **Golden metrics**, a snapshot of the entity's key performance metrics. - **Related entities**, grouped and sorted by type, which you can open in a stacked view alongside your current context. - **Links to related applications**, marked with an external link icon where they open the APM experience rather than a stacked view. - **Object description**, which opens the manifest for the selected Kubernetes object. - **Team**, if one is set for the entity. - **Dashboard**, a link to the entity's corresponding dashboard. ## Correlate with APM, events, logs, and alerts [#correlate] The fleet view is designed so you don't have to abandon it to correlate signals and reduce MTTR: - **APM insights**: The Workloads view surfaces application performance data for the applications running in a workload, so you can move between infrastructure health and application performance in one place. - **Alerts**: Open alerts are surfaced against clusters, nodes, and workloads, including alert indicators in the Workloads tab, and open in a stacked view for immediate context. Open alerts appear regardless of the time range you have selected, so an active alert is never hidden by a narrow time picker. - **Events and logs**: Kubernetes events and logs for an entity open in a stacked view alongside your current context, so you can correlate signals without losing your place in the triage flow. ## Tab-specific metrics [#tab] The UI is organized into five main tabs. `Scorecard:` thresholds apply to the scorecards and `Table:` thresholds apply to table rows. When they're the same, only one set is listed. Counts cover the selected time range. **Overview tab** The **Overview** tab provides a general summary of the **operational health, capacity, and risk** across your entire Kubernetes fleet. | Metric name | Description | Importance | Thresholds | | --------------------- | ------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------- | | `Unhealthy nodes` | Unhealthy nodes out of all nodes. A node is unhealthy if it isn't ready or reports a pressure or network condition. | Signals widespread instability and potential zone failures across the infrastructure. | Orange: 1% to < 5% of nodes; Red: ≥ 5%. | | `Unhealthy workloads` | Deployments, daemonsets, and statefulsets with missing or unavailable pods, out of all workloads. | Indicates failure to reconcile desired states across the organization. | Orange: 1% to < 5% of workloads; Red: ≥ 5%. | | `Running pods` | Percentage of pods in the `Running` phase, excluding job pods and completed pods. | Operational success rate; less than 90% running pods indicates widespread failure. | `Scorecard:` Orange: 90% to < 95%; Red: < 90%. `Table:` Orange: > 95% to 99%; Red: ≤ 95%. | | `CPU usage` | Node CPU usage as a percentage of allocatable CPU. | Measures aggregate node saturation risk across the entire estate. | Orange: 75% to < 95%; Red: ≥ 95%. | | `Memory usage` | Node memory usage as a percentage of allocatable memory. | High values indicate risk of aggregate eviction storms across clusters. | Orange: 75% to < 95%; Red: ≥ 95%. | | `Ephemeral storage` | Percentage of node filesystem capacity used. | Critical to prevent Kubelet crashes. | Orange: 75% to < 95%; Red: ≥ 95%. | | `Persistent storage` | Percentage of pod volume capacity used. | Critical to prevent stateful application crashes and data write failures. | Orange: 75% to < 95%; Red: ≥ 95%. | | `Warning events` | Number of Kubernetes `Warning` events. | High noise levels obscure critical alerts and indicate config debt. | Orange: 1 to 49; Red: ≥ 50. | **Health tab** The **Health** tab focuses on metrics related to the **availability and current status** of your nodes and pods to detect immediate failure risks. | Metric name | Description | Importance | Thresholds | | ----------------------------- | --------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------- | | `Nodes ready` | Nodes that reported no unhealthy condition, out of all nodes. | Availability metric; values below 95% signal severe fleet instability. | `Scorecard:` Orange: 95% to < 100%; Red: < 95%. `Table:` Cluster rows: Orange: > 1% to 2% of nodes ready; Red: ≤ 1%. Node rows: Red: `FALSE`. | | `Node memory pressure` | Number of nodes reporting `MemoryPressure`. | Predicts widespread eviction storms. | `Scorecard:` Orange: 1% to < 6% of nodes; Red: ≥ 6%. `Table:` Cluster rows: Orange: 98 nodes; Red: ≥ 99 nodes. Node rows: Red: `TRUE`. | | `Node disk pressure` | Number of nodes reporting `DiskPressure`. | Risk of widespread node failures due to disk exhaustion. | `Scorecard:` Orange: 1% to < 6% of nodes; Red: ≥ 6%. `Table:` Cluster rows: Orange: 98 nodes; Red: ≥ 99 nodes. Node rows: Red: `TRUE`. | | `Running pods` | Percentage of pods in the `Running` phase, excluding job pods and completed pods. | Operational success rate; less than 90% running pods indicates widespread failure. | `Scorecard:` Orange: 90% to < 95%; Red: < 90%. `Table:` Orange: > 95% to 99%; Red: ≤ 95%. | | `Pending pods` | Number of pods in the `Pending` phase, excluding job pods. | Indicates scheduler failure or capacity starvation. | `Scorecard:` Orange: 1 to 5; Red: ≥ 6. `Table:` Orange: 1; Red: ≥ 2. | | `Container restarts` | Number of container restarts. | Symptom of CrashLoopBackOff or OOMKills. | `Scorecard:` Orange: 1 to 5; Red: ≥ 6. `Table:` Orange: 1; Red: ≥ 2. | | `Network errors (per second)` | Rate of network errors per second. | Symptom of hardware failure or systemic CNI issues. | `Scorecard:` Orange: 11 to < 50 per 100 nodes; Red: ≥ 50 per 100 nodes. `Table:` Orange: 5 to 10; Red: ≥ 11. | | `Warning events` | Number of Kubernetes `Warning` events. | High noise levels obscure critical alerts and indicate config debt. | Orange: 1 to 49; Red: ≥ 50. | **Performance tab** The **Performance** tab measures **utilization against total capacity and limits** to identify node saturation and risk of performance degradation. | Metric name | Description | Importance | Thresholds | | ----------------------- | -------------------------------------------------------------------- | ------------------------------------------------------------------- | -------------------------------------- | | `CPU usage (total)` | Node CPU usage as a percentage of allocatable CPU. | Measures saturation risk across the entire estate. | Orange: 75% to < 95%; Red: ≥ 95%. | | `CPU usage (limits)` | CPU usage as a percentage of CPU limits, where set. | Measures proximity to throttling. | Orange: 75% to < 95%; Red: ≥ 95%. | | `Memory usage (total)` | Node memory usage as a percentage of allocatable memory. | Proximity to eviction thresholds. | Orange: 75% to < 95%; Red: ≥ 95%. | | `Memory usage (limits)` | Memory working set as a percentage of memory limits, where set. | Proximity to OOMKill events. | Orange: 75% to < 95%; Red: ≥ 95%. | | `Ephemeral storage` | Percentage of node filesystem capacity used. | Critical to prevent Kubelet crashes. | Orange: 75% to < 95%; Red: ≥ 95%. | | `Pod capacity` | Running pods as a percentage of the maximum pods that nodes can run. | Measures resource density and slot exhaustion. | Orange: 85% to < 100%; Red: ≥ 100%. | | `Warning events` | Number of Kubernetes `Warning` events. | High noise levels obscure critical alerts and indicate config debt. | Orange: 1 to 49; Red: ≥ 50. | **Resources tab** The **Resources** tab analyzes **Request vs. Usage and Request vs. Limits** to assess resource allocation efficiency and identify waste or sizing inaccuracies. | Metric name | Description | Importance | Thresholds | | --------------------------- | ------------------------------------------------------------- | ------------------------------------------------------------------------------ | ----------------------------------------------------------------------------- | | `CPU usage vs requests` | CPU usage as a percentage of CPU requests. | Governance metric for sizing accuracy and waste. | `Scorecard:` Orange: < 70% or ≥ 200%. `Table:` Orange: < 70% or > 200%. | | `CPU requests vs limits` | CPU requests as a percentage of CPU limits. | Measures performance headroom and burst capacity. | Orange: 75% to < 95%; Red: ≥ 95%. | | `CPU throttling` | Percentage of CPU periods in which containers were throttled. | Measures latency/lag caused by CFS quotas. | Orange: 10% to < 50%; Red: ≥ 50%. | | `Memory usage vs requests` | Memory usage as a percentage of memory requests. | Governance metric for memory sizing accuracy. | `Scorecard:` Orange: < 70% or ≥ 200%. `Table:` Orange: < 70% or > 200%. | | `Memory requests vs limits` | Memory requests as a percentage of memory limits. | Helps to assess resource allocation efficiency and potential overprovisioning. | Orange: 75% to < 95%; Red: ≥ 95%. | | `Pod evictions` | Number of pods evicted by the kubelet. | Symptom of node resource starvation. | `Scorecard:` Orange: 1 to 5; Red: ≥ 6. `Table:` Orange: 1; Red: ≥ 2. | | `Warning events` | Number of Kubernetes `Warning` events. | High noise levels obscure critical alerts and indicate config debt. | Orange: 1 to 49; Red: ≥ 50. | **Workloads tab** The **Workloads** tab provides an application-centric view of your deployments, daemonsets, and statefulsets, focusing on metrics such as **readiness, restarts, and missing replicas** to diagnose application instability and poor QoS. Warning events are available in this tab for `Deployments`, `DaemonSets`, and `StatefulSets`, so you can see the events behind an unhealthy workload without leaving the tab. | Metric name | Description | Importance | Thresholds | | -------------------------- | --------------------------------------------------------------- | ------------------------------------------------------------------------------------ | -------------------------------------------------------------------- | | `CPU usage (requests)` | CPU usage as a percentage of CPU requests. | Measures how close the workload is to exceeding its requested CPU. | Orange: 75% to < 95%; Red: ≥ 95%. | | `CPU throttling` | Percentage of CPU periods in which containers were throttled. | Measures latency/lag caused by CFS quotas. | Orange: 10% to < 50%; Red: ≥ 50%. | | `Memory usage (limits)` | Memory working set as a percentage of memory limits, where set. | Proximity to OOMKill. | Orange: 75% to < 95%; Red: ≥ 95%. | | `Pods Ready/Total` | Ready pods against total pods | Baseline metric for capacity, and shows readiness against desired count at a glance. | Red: 0 ready. | | `Pods Not Ready` | Count of pods in the fleet that are not reporting `Ready`. | Surfaces workloads that are running but not serving traffic. | `Fleet:` Yellow: 1 to 5; Red: 6. `Workload:` Yellow: 1; Red: 2. | | `Pods Unavailable` | Count of pods that are not running, scheduled, or ready. | Indicates service availability. | `Fleet:` Yellow: 1 to 5; Red: 6. `Workload:` Yellow: 1; Red: 2. | | `Pods missing` | Gap between desired and ready replicas. | Indicates service degradation. | `Scorecard:` Orange: 1 to 5; Red: ≥ 6. `Table:` Orange: 1; Red: ≥ 2. | | `Pod evictions` | Number of pods evicted. | Workload is causing node pressure or is misconfigured. | `Scorecard:` Orange: 1 to 5; Red: ≥ 6. `Table:` Orange: 1; Red: ≥ 2. | | `Container restarts` | Number of container restarts. | Persistent failure state (CrashLoopBackOff). | `Scorecard:` Orange: 1 to 5; Red: ≥ 6. `Table:` Orange: 1; Red: ≥ 2. | | `Container image restarts` | Number of distinct container images with restarts. | Shows whether restarts are isolated to one image or spread across several. | `Scorecard:` Orange: 1 to 2; Red: ≥ 3. `Table:` Orange: 1; Red: ≥ 2. | | `Warning events` | Number of Kubernetes `Warning` events. | High noise levels obscure critical alerts and indicate config debt. | Orange: 1 to 49; Red: ≥ 50. |