Skip to content
90 production scenarios

Stop watching DevOps.
Start doing DevOps.

Practice production incidents, infrastructure changes, deployments, Kubernetes failures, CI/CD problems and cloud operations inside realistic engineering environments.

No credit card to start. 11 scenarios are free.

Scenarios
90
Categories
9
Graded on
Environment state
INC-1042P1SEV1
12m left

Production API returning 502 after deployment

The checkout API began returning intermittent 502 responses approximately 15 minutes after the latest production deployment. Customer support reports 17 failed checkout attempts in the last 20 minutes.

deploy@web-prod-01 — terminal
$ curl -i http://localhost/api/checkout
HTTP/1.1 502 Bad Gateway

$ ssh api-prod-01
$ systemctl status checkout-api
● checkout-api.service — failed
     Status: "main process exited, code=exited, status=2"

$ sudo systemctl start checkout-api
$ exit
$ curl -i http://localhost/api/checkout
HTTP/1.1 200 OK
http_error_rate0.3% — recovered
0%16.1%32.2%48.3%

What is DevOps Arena

A job simulator, not a course

You are onboarded as an engineer at NexaCommerce, a mid-size commerce platform. Work arrives as tickets and incidents. You investigate, fix, and submit — and your work is graded by inspecting the environment you leave behind, not by a quiz.

Nobody tells you the answer

A ticket describes a symptom and a business impact. The cause is somewhere in the environment, and the logs contain evidence rather than instructions.

Any valid fix passes

Evaluation reads the resulting state — is the service running, does the endpoint answer, has the error rate recovered. It never compares against a stored answer, so an unanticipated route works.

Restarting it does not help

A service whose config does not parse will not start however many times you try. The environment is a state machine, so the fix has to be real.

How it works

The loop

  1. 01

    Work arrives

    A ticket or incident lands in your queue with a priority, an SLA and a business impact.

  2. 02

    Environment starts

    An isolated estate boots in under a second — hosts, services, containers, a cluster, metrics and logs.

  3. 03

    You fix it

    SSH between hosts, read logs, inspect services, change configuration, deploy, roll back, verify.

  4. 04

    You are graded

    Checks inspect the environment state and return a scored report with what passed, what did not, and why.

Evaluation

Graded like a code review

Every attempt produces a report against a fixed rubric, with a per-check explanation of what was observed.

Rubric

Correctness
40
Reliability
20
Security
15
Automation
10
Efficiency
10
Documentation
5

Sample feedback

The checkout API service is running on api-prod-01.
Checkout returns 200 through the edge across 3 samples.
Error rate has returned below the 5% threshold.
The unit is not enabled, so this outage would repeat after any reboot.
HTTPS redirect is not configured on the edge.

FAQ

Questions engineers actually ask

Are these real cloud environments?

The environments are a deterministic simulation of a production estate — filesystem, processes, systemd, Docker, Kubernetes, Terraform and AWS state. That is deliberate: grading has to be reproducible, and a scenario that passes or fails depending on image pull latency is a broken scenario. Docker and Kubernetes providers exist behind the same interface for deployments that want real infrastructure.

Do I need to already know DevOps?

Yes. This assumes you know what systemctl, kubectl and terraform are. It will not explain what a container is. It gives you a failing production system and expects you to work.

How is my work graded?

Checks inspect the resulting environment: is the service running, is the port listening, does the endpoint answer, has the metric recovered, does the config contain what it needs. Nothing compares your commands against an expected transcript.

What if I solve it a different way?

You pass. The checks assert properties of the finished environment, so any route that reaches a correct end state is a correct route.

Can I use hints?

Yes, and they cost points. They are progressive and time-gated — the first unlocks after a few minutes, and each one nudges toward a method of investigation rather than naming the answer.

Is my work saved?

Every attempt records its commands, hints, documents, evaluation and score. You can review any past attempt and see exactly what you did.

The next incident will not wait for you to be ready.

Practise the work before it is your production system and your Saturday.