Skip to content

📝 Block E Exam — Ownership#

"The D → E transition signal: did something you built yourself break because of your own mistake — and did you bring it back?"

Gate: End of Block E (after E5, before F1) · Prerequisite: E1–E5 acceptance criteria passed

ℹ️ Run every verify.sh from the 22-Learning-Path/ root (bash labs/broken/…/verify.sh). K08 is meaningful locally too, via doc/access checks; K07 audits incident docs, K09 audits a live cluster — without that environment, verify.sh won't come back green.

Block E is the transition from "operator" to "owner." This exam can measure the mechanics of ownership: is the SLO defined, does the alarm actually fire, was the restore genuinely tested. Every question traces back to a module's acceptance criterion.

🧗 Honest ceiling (in the main text, not a footnote): This exam proves the mechanics of ownership; not ownership itself. That requires an outage you didn't choose, a system you actually own, and real users. What comes after this gate isn't more reading: it's taking a job, going on-call, volunteering for an incident. See README.md → Honest ceiling.

1️⃣ Concept questions (written)#

# Question Traceability (module → acceptance criterion)
1 How do you choose an SLI? How do you calculate an error budget (min/month) from an SLO? E1 → SLI/SLO/error budget criterion
2 What changes once the error budget is exhausted? (Does release stop, what becomes the priority?) E1 → budget exhaustion criterion
3 What's the "should this wake someone at 3am?" test for an alarm? The page / ticket / log distinction? E2 → alarm classification criterion
4 Give an example of a "noisy alarm"; why is it silenced/removed? When does escalation kick in? E2 → noise + escalation criterion
5 What does "blameless postmortem" mean? Which three things are mandatory in one? E3 → postmortem criterion
6 Why is "an untested backup isn't a backup"? What's the difference between RTO and RPO? E4 → restore + RTO/RPO criterion
7 What security checks are asked of a backup? (Who can access it, is it encrypted at rest?) E4 → backup access/at-rest criterion
8 Why is a game day with a limited blast radius run as hypothesis → experiment → result? E5 → game day criterion

Passing: At least 7 of 8 questions correct. Questions 6 and 7 (restore + backup security) must be correct — that's the security thread's link in Block E.

2️⃣ Applied task — the mechanics of ownership#

Task A — Three broken labs (core):

Task B — Alarm tied to an SLO (E1+E2): L18 + L19.

  • An SLI for a service is measured in Prometheus; an SLO + error budget (min/month) has been calculated in writing
  • An alarm rule tied to the SLO fired at least once and was resolved — Alertmanager/panel proof
  • Every alarm is classified as page/ticket/log; escalation is defined in writing

Task C — Restore genuinely tested (mandatory, security thread): E4/L20.

  • A backup was restored to a clean environment; data integrity was verified with a query (row count/checksum)
  • RTO and RPO were measured and written down
  • The backup's access + at-rest encryption controls were written down

Task D — Postmortem + action item: Write a blameless postmortem from K07: numeric impact + root cause + "why wasn't it caught sooner" + at least one traceable action item (owner + due date).

🚫 Don't lose this exam to yourself#

Anti-pattern Why it's bad Right
Trusting an untested backup The restore won't work in a real disaster — E4's core lesson Restore to a clean environment, query the integrity
Blaming a person in the postmortem People stop sharing information; the root cause gets hidden Write it system/process-focused, blameless
A postmortem with no action item The same incident happens again Produce a traceable item with an owner + due date
Alerting that pages for everything Alert fatigue → real pages get missed Classify page/ticket/log; silence the noise
"Chaos" with no blast radius You actually break your own production Limited scope + hypothesis → experiment → result
Passing the mechanics and calling yourself "an owner" Honest ceiling: real ownership happens in production Prove the mechanics; the job/on-call teaches the rest

✅ Did you pass?#

  • Concept: at least 7 of 8, plus questions 6 & 7 mandatory correct
  • Application: K07 + K08 + K09 green; the SLO-tied alarm fired; the restore was verified in a clean environment
  • Writing: blameless postmortem + traceable action item (owner + due date)

If you didn't pass: go back to E1 for the SLO, E2 for alerting, E3 for incident/postmortem, E4 for restore, E5 for chaos.

⏭️ Up next#

If you passed, go to Capstone 3 first, then F1 — Cost and Trade-off (FinOps) — but read the honest ceiling first: from here on, it's production's job.

"Ownership isn't building a system; it's being the one who gets called when it breaks and brings it back. This exam shows those muscles work; what actually exhausts them is your first real incident."