Grafana Interview Prep: 4-Week Study Plan (2027)
Four-week Grafana interview preparation plan with a Docker lab, weekly goals and 32 self-check questions on PromQL, alerting, Loki and provisioning.
To prepare for a Grafana monitoring interview, spend about four weeks moving from dashboards and PromQL to alerting, logs and traces, then dashboards as code and incident drills. Build one lab stack (Prometheus, Node Exporter, Loki, Grafana) and practise explaining every panel and alert you create. This plan includes weekly labs and 32 self-check questions with answers.
This page is the study plan. For question banks, use our Grafana scenario-based interview questions (the main Grafana interview page), the most asked Grafana questions for freshers and the Grafana certification interview questions.
Key takeaways
- Interviewers test whether you can build and explain a dashboard and an alert, not whether you can recite menu names.
- PromQL is the skill most candidates lack. Give it a full week.
- Alerting in Grafana has several parts (rules, contact points, notification policies, silences). Learn how they connect.
- Dashboards as code (provisioning, JSON in Git, Terraform or the Grafana operator) separates mid-level candidates from beginners.
- Finish with timed incident drills: start from an alert and find the cause using metrics, then logs.
Which Grafana page to use when
| Your situation | Use |
|---|---|
| Interview in more than three weeks | This study plan |
| Fresher, need basics fast | Most asked Grafana questions for freshers |
| Experienced, interview this week | Scenario-based questions |
| Need PromQL depth | Prometheus interview questions |
Before you start: set up the lab
You need a laptop with Docker and about 4 GB of free memory. One Compose file runs everything:
services:
prometheus:
image: prom/prometheus
volumes:
- ./prometheus.yml:/etc/prometheus/prometheus.yml
ports: ["9090:9090"]
node-exporter:
image: prom/node-exporter
loki:
image: grafana/loki
ports: ["3100:3100"]
grafana:
image: grafana/grafana
ports: ["3000:3000"]
volumes:
- grafana-data:/var/lib/grafana
volumes:
grafana-data:
Pin image versions once it works, so the lab is repeatable. Add Grafana Alloy or Promtail later to ship logs to Loki (Grafana Labs documents Alloy as the recommended collector; check the Loki docs for the current status of Promtail).
The four-week plan at a glance
| Week | Focus | Deliverable |
|---|---|---|
| 1 | Data sources, panels, variables, PromQL basics | Host dashboard with an instance variable |
| 2 | PromQL for services, RED and USE methods | Service dashboard with rate, errors, latency percentiles |
| 3 | Grafana Alerting, Loki and LogQL, linking signals | Two alerts routed by label; metrics-to-logs drill-down |
| 4 | Dashboards as code, access control, incident drills | Provisioned stack in Git; three timed drills |
Week 1: dashboards and first queries
What to learn
- Adding a data source and testing it; Explore.
- Panel types: time series, stat, gauge, table, bar chart, heatmap.
- Units, thresholds, value mappings, overrides.
- Query variables, chained variables, multi-value and All.
- PromQL selectors,
rate(),sum by,avg by.
Lab for week 1
Build a host dashboard from Node Exporter metrics without importing one: CPU, memory, disk usage, network traffic, filtered by an $instance variable. Then import the community Node Exporter Full dashboard and compare queries.
Week 1 self-check
1. Why does a counter need rate() before graphing?
A counter only increases and resets on restart. rate() converts it to per-second increase and handles resets, which gives a meaningful line.
2. How do you make one dashboard work for every server?
Create a query variable such as label_values(up{job="node"}, instance) and use instance=~"$instance" in panel queries. Use =~ so multi-value selections work.
3. When is a stat panel better than a time series?
When the question is "what is it now?" (current error count, uptime). Use time series when the trend over time matters.
4. A memory graph shows 8000000000. What is missing?
The unit. Set it to bytes (IEC) so Grafana shows 7.45 GiB.
5. What is the difference between $__interval and $__rate_interval?
$__interval follows the time range and panel width. $__rate_interval is meant for rate() with Prometheus and is never smaller than four times the scrape interval, so short ranges do not produce empty rate results.
6. Where would you check a query that returns nothing?
In Explore with the same data source and time range, removing label filters one at a time until data appears.
Week 2: PromQL for services
What to learn
- RED method (rate, errors, duration) for services and USE method (utilisation, saturation, errors) for resources.
histogram_quantile()withlebuckets.- Ratios: errors divided by total requests.
increase(),topk(),offset, recording rules.
Lab for week 2
Run a sample app that exposes HTTP metrics (many language clients ship examples). Build a service dashboard with request rate by status code, error ratio, and p50, p90 and p99 latency.
sum(rate(http_requests_total{status=~"5.."}[5m]))
/ sum(rate(http_requests_total[5m]))
histogram_quantile(0.99,
sum by (le) (rate(http_request_duration_seconds_bucket[5m])))
Week 2 self-check
7. Why must sum by (le) come before histogram_quantile?
The function needs the bucket boundaries in the le label. Aggregating away le destroys the histogram; keeping it while summing other labels gives an overall quantile.
8. Why not average p99 latency across pods?
Averaging percentiles is mathematically wrong. Sum the bucket rates across pods first, then compute the quantile.
9. What is a recording rule and when do you add one?
A Prometheus rule that pre-computes an expression into a new series. Add one when a heavy query runs on many dashboards or alerts.
10. What is the difference between RED and USE?
RED describes request-driven services from the user's side. USE describes resources such as CPU, disk and network. A good service dashboard uses RED at the top and USE below.
11. How do you show the five busiest endpoints?
topk(5, sum by (route) (rate(http_requests_total[5m]))). In a time series panel, topk can show more than five series across the range, since the top set changes over time; a bar gauge with instant query avoids that.
12. How do you compare today with last week?
Add a second query with offset 1w, or use the panel's time shift option.
Week 3: alerting, logs and linking signals
What to learn
- Grafana-managed alert rules, pending period, labels and annotations.
- Contact points, notification policies, grouping, mute timings and silences.
- Loki labels and LogQL filters,
rate()over logs. - Data links and correlations between metrics, logs and traces (Tempo if time allows).
Lab for week 3
Create two alert rules: a service down alert with label severity=critical and an error ratio alert with severity=warning. Route critical to one contact point and warning to another using a notification policy. Ship app logs to Loki and add a data link from the error panel to the matching logs in Explore. The Grafana Alerting docs cover each part.
Week 3 self-check
13. Walk through what happens from alert rule to Slack message.
The rule evaluates on its interval, the condition stays true for the pending period, the alert fires with its labels, the notification policy tree matches labels to a contact point, grouping and timing settings batch alerts, then Slack receives the message.
14. What is the difference between a silence and a mute timing?
A silence is a one-off suppression for matching alerts over a set period. A mute timing is a recurring window (such as weekends) attached to a notification policy.
15. Why put severity and team labels on every alert?
Notification policies route by labels, so labels decide who gets paged and how. Without them, all alerts go to the default policy.
16. How do you avoid one notification per pod?
Group by labels such as alertname and service in the notification policy so pods of one service arrive as a single notification.
17. Why keep Loki labels low in cardinality?
Loki indexes labels, not log content. High-cardinality labels such as user IDs create many streams and hurt performance. Filter on content with |= or parsers instead.
18. How does a user jump from a latency spike to the related logs?
Through a data link or correlation configured on the panel or data source that opens Explore with a LogQL query using the same labels and time range. With exemplars and Tempo, they can jump to a trace too.
19. Grafana-managed or data source-managed alert rules?
Grafana-managed rules can query any supported data source and are stored in Grafana. Data source-managed rules live in Prometheus-compatible backends such as Mimir or Loki rulers. Teams already running Prometheus rules in Git often keep those.
Week 4: dashboards as code, access and drills
What to learn
- Provisioning data sources, dashboards and alert rules from files.
- Dashboard JSON, version history, folders and permissions.
- The Grafana Terraform provider or Grafana operator (pick one).
- Teams, roles, service accounts, SSO basics.
Lab for week 4
Move your lab to provisioning: data sources as YAML, dashboards as JSON in a Git repository, loaded at startup. Delete the Grafana volume and confirm everything returns. Then run three timed drills of 20 minutes each: stop the app container, fill the disk with a test file, and inject errors in the app. Start each drill from the alert, not the dashboard.
apiVersion: 1
providers:
- name: team-dashboards
folder: Services
type: file
options:
path: /var/lib/grafana/dashboards
Week 4 self-check
20. How do you keep dev, staging and production dashboards identical?
Store dashboard JSON in Git and deploy through provisioning, Terraform or an operator, with data source names or UIDs kept the same across environments.
21. Someone edits a provisioned dashboard in the UI. What happens?
By default the UI does not let them save changes to a file-provisioned dashboard (the option allowUiUpdates controls this). The right flow is to change the JSON in Git.
22. How do you restrict a team to its own dashboards?
Put the team's dashboards in a folder and grant the team Editor permission on that folder only, with Viewer elsewhere.
23. Why use a service account for automation?
Its tokens are tied to a non-human identity with a role, can be rotated, and do not break when an employee leaves.
24. How do you back up Grafana?
Back up the Grafana database (SQLite file or external MySQL/PostgreSQL) and keep dashboards and alert rules as code in Git, which is the easier restore path.
25. In a drill, the alert fires but the dashboard looks normal. What next?
Check the alert's query and labels, compare its time range with the dashboard's, and confirm both use the same data source. Dashboards and alerts often drift apart.
Self-check by experience level
Interviewers adjust depth to your experience. Use these final questions to test the level you are applying for.
26. Freshers: what does Grafana not do?
It does not store metrics or collect them from hosts. Exporters and agents collect, Prometheus or Loki store, Grafana queries and visualises.
27. Freshers: what is the first dashboard you would build for a web app?
Availability (up), request rate, error ratio and latency percentiles, then CPU and memory.
28. 2 to 5 years: a dashboard takes 30 seconds to load. Where do you look?
Panel count, query cost (long ranges, high cardinality), refresh interval, and repeated identical queries. Fix with recording rules, collapsed rows, fewer panels and the Dashboard data source to reuse a query.
29. 2 to 5 years: how do you measure an SLO in Grafana?
Define the SLI as a ratio (good requests over total), compute it over the SLO window, show error budget remaining, and alert on burn rate rather than raw errors. See SLA and SLO monitoring tools.
30. Senior: how would you run Grafana for 50 teams?
Highly available Grafana instances behind a load balancer on a shared external database, SSO with team sync, folders per team, dashboards and alerts as code with review, and usage review to remove unused dashboards.
31. Senior: how do you control metric cardinality costs?
Find the series count per metric and label, drop or relabel high-cardinality labels at scrape time, use recording rules for dashboards, and set per-tenant limits in Mimir or similar backends.
32. Scenario: on-call says alerts are noisy and they ignore them. What do you change?
Review each alert for actionability, delete or downgrade ones nobody acts on, move to symptom-based alerts tied to SLOs, add grouping and inhibition, and route warnings to a channel instead of a pager.
In the final week before the interview
- Prepare a two-minute walkthrough of your best dashboard: who uses it, what each row answers, which alert links to it.
- Rehearse one incident story from your drills: alert, metric, log, cause, fix.
- Revise PromQL by writing queries without autocomplete.
- Read the latest Grafana release notes for big changes since you learned it.
Related reading: Prometheus and Grafana together and 12 monitoring dashboards every DevOps should build. If you prefer guided labs, our Prometheus and Grafana training runs in Kharadi, Pune and live online.
FAQs
How long does it take to prepare for a Grafana interview?
About four weeks at one to two hours a day if you already know Linux and Docker. Experienced users can compress it into one or two weeks.
Should I learn PromQL before Grafana?
Learn them together. Grafana is easy to click through, but panels are only as good as their queries.
Do I need Grafana Cloud for practice?
No. A local Docker Compose stack covers everything in this plan. The free Cloud tier is useful if your laptop is short on memory.
Which Grafana alerting concepts are asked most?
Alert rules, pending period, contact points, notification policies, grouping and silences.
Is dashboards as code expected for mid-level roles?
Often, yes. Be ready to explain provisioning or Terraform and how changes are reviewed.
What should I show in the interview?
A Git repository with your provisioned stack, one service dashboard and its alerts, plus a short write-up of one drill.
Is there an official Grafana certification?
Grafana Labs' current credentials are covered in our Grafana certification page; check grafana.com for the latest details.
Official references
What's Your Reaction?
Like
0
Dislike
0
Love
0
Funny
0
Angry
0
Sad
0
Wow
0