Prometheus and Grafana are the backbone of modern observability, and they appear in most SRE, DevOps and platform interviews. This 2026 guide covers the questions you are most likely to face, from Prometheus fundamentals to Grafana and Kubernetes scenarios.
Prometheus Fundamentals
What is Prometheus and what model does it use?
Prometheus is an open-source monitoring and alerting system that pulls (scrapes) metrics from instrumented targets over HTTP at intervals, stores them in a time-series database, and supports querying via PromQL. The pull model makes target health easy to detect and scraping simple to control.
What are the four core metric types?
- Counter — a monotonically increasing value (e.g., total requests).
- Gauge — a value that can go up or down (e.g., memory in use).
- Histogram — samples observations into buckets (e.g., request latency) plus sum and count.
- Summary — like a histogram but calculates configurable quantiles client-side.
What is an exporter?
An exporter exposes metrics from a system that can't be instrumented directly — e.g., node_exporter for host metrics, or exporters for databases and message queues.
How does Prometheus find what to scrape?
Through the scrape configuration: static targets or service discovery (Kubernetes, Consul, EC2, etc.) that dynamically lists targets and applies relabeling.
PromQL
What is the difference between rate() and irate()?
rate() gives the average per-second rate over a range and is best for alerting and dashboards; irate() uses the last two samples for a fast-moving, instantaneous rate, better for volatile graphs.
What are recording rules?
Precomputed queries saved as new time series, used to speed up expensive dashboards and keep alert expressions simple.
Alerting
How does alerting work in Prometheus?
You define alerting rules in Prometheus; when they fire, alerts are sent to Alertmanager, which handles grouping, deduplication, silencing, inhibition and routing to receivers (email, Slack, PagerDuty, etc.).
Storage and Scaling
How long does Prometheus retain data, and how do you keep it longer?
Local TSDB retention is time/size bound (often ~15 days by default). For long-term, highly-available storage, use remote write to systems like Thanos, Cortex or Grafana Mimir, which also enable global query views and downsampling.
Grafana
What is Grafana and how does it relate to Prometheus?
Grafana is a visualization and dashboarding platform. It queries Prometheus (and many other data sources) to render panels, supports template variables for reusable dashboards, and has its own alerting engine.
How do dashboard variables help?
They let one dashboard work across many targets (e.g., a $namespace or $instance dropdown) instead of duplicating dashboards.
Prometheus on Kubernetes
How do you monitor a Kubernetes cluster?
Typically with the kube-prometheus-stack: the Prometheus Operator manages Prometheus via CRDs, ServiceMonitor/PodMonitor objects declare scrape targets, kube-state-metrics exposes object state, and cAdvisor provides container metrics.
Scenario Questions
- High cardinality: caused by unbounded label values (user IDs, request IDs). Avoid by limiting labels and using recording rules; it can blow up memory.
- A target shows no metrics: check the target is UP in Prometheus, the scrape path/port, network policy/firewall, and relabeling rules.
- Defining SLOs: use histograms to compute latency percentiles and error-rate ratios, then alert on burn rate.
Keep Learning
Round out your observability prep with our Datadog and ELK-on-EKS guides, the full interview-questions library, and DevOps training to practise hands-on.
