RetroDash // OVERVIEW

For this dashboard to work correctly you need:

• Prometheus datasource configured in Grafana
• node_exporter for node metrics (CPU, memory, disk)
• kube-state-metrics for pod status and K8s resources
• up metric for health check (scrape_duration_seconds)

Grafana 12+ Import

These dashboards are tested with Grafana 12.3 and kube-prometheus-stack. To import: go to Dashboards → Import, paste the JSON, then select your Prometheus datasource from the dropdown.

• job="node-exporter" is the default job label set by kube-prometheus-stack. Your setup may use job="node" or another value — adjust accordingly.
• The $job_node variable (in the dashboard JSON) lets you switch between node-exporter and node without editing individual panel queries.
• If the Node Status Grid shows no data, check the $job_node variable value in Dashboard Settings → Variables.

Health Score

Shows the percentage of "up" targets in Prometheus.

Default query:

100 * count(up == 1) / count(up)

To change: If you use a custom exporter, replace up with your metric:

100 * count(my_service_healthy == 1) / count(my_service_healthy)

Thresholds: Edit Thresholds in panel (default: 90 yellow, 75 red)

Avg Latency

Average request latency in milliseconds.

avg(rate(http_request_duration_seconds_sum[5m]))
/ avg(rate(http_request_duration_seconds_count[5m])) * 1000

Change metric: If you don't use Prometheus HTTP conventions, use:

avg(my_request_latency_ms)

Error Rate

Percentage of requests with 5xx status code.

100 * sum(rate(http_requests_total{status=~"5.."}[5m]))
/ sum(rate(http_requests_total[5m]))

Customization: To include 4xx errors:

100 * sum(rate(http_requests_total{status=~"[45].."}[5m]))
/ sum(rate(http_requests_total[5m]))

CPU Saturation

CPU usage as a percentage.

100 - (avg(rate(node_cpu_seconds_total{mode="idle"}[5m])) * 100)

Filter by specific instance:

100 - (avg(rate(node_cpu_seconds_total{mode="idle",instance="server1:9100"}[5m])) * 100)

Golden Signals Summary

Table with 4 main KPIs. To customize:

• Latency: Change http_request_duration_seconds to your metric
• Traffic: Modify time range (5m → 1m for more sensitivity)
• Errors: Adjust status codes based on your app
• Saturation: Change metric if monitoring something other than CPU

Node Status Grid

Dynamic table with node status.

max by (instance) (up{job="node"})

Change columns: In panel → Columns, add:

• node_cpu_seconds_total → Total CPU
• node_memory_MemAvailable_bytes → Available memory
• node_filesystem_avail_bytes → Disk space

Pod Status

Table of pods with status.

max by (pod, namespace) (kube_pod_status_phase)

Filter by namespace: In variables, create selector:

label_values(kube_pod_info, namespace)

Then use in the query:

max by (pod, namespace) (kube_pod_status_phase{namespace="$namespace"})

Top Resource Consumers

Ranking of pods by CPU/memory.

topk(10, sum by (pod, namespace)
(rate(container_cpu_usage_seconds_total[5m])) * 100)

Change top N: Replace topk(10 with topk(20 for top 20

1. In Grafana, go to Dashboards → Import
2. Copy the dashboard JSON (available in repo or export)
3. Paste in Import via panel json
4. Select Prometheus datasource
5. Click Import
6. Verify that all metrics load correctly