Practice · Operation, Maintenance and Sustainable Computing

Q1

What is the main benefit of a canary (staged) rollout?

A A bad release reaches only a small share of users first, so it can be detected by comparing metrics and stopped early. B It makes the release faster for everyone. C It removes the need for testing. D It encrypts the release.

Q2

Buddy's new "exam mode" is deployed to production but switched off, then enabled only for one course section. What technique is this?

A A postmortem B A database migration C Feature flags — separating deployment from release, with an instant kill switch D Blue-green deployment

Q3

Why should a database change be done as "expand and contract" (add the new column first, remove the old one later)?

A Old and new versions of the code can both work during the rollout, so a rollback remains possible. B Databases cannot remove columns. C It uses less disk space. D It is required by SQL.

Q4

Which is not one of the four golden signals?

A Latency B Traffic C Number of developers on call D Saturation

Q5

Which telemetry type best shows where time is spent in one slow request that goes from the app through the API and tutor module to the AI provider?

A A single CPU metric B The Git log C A privacy notice D A distributed trace

Q6

The on-call student receives 300 alerts a day, most about internal details that do not affect users. What is the risk and the fix?

A Alert fatigue — real problems get ignored; alert on user-visible symptoms (SLO violations) and send the rest to dashboards. B No risk; more alerts are safer. C Remove all monitoring. D The servers will overheat.

Q7

Which statement correctly distinguishes SLO and SLA?

A They are the same. B An SLO is an internal target for a measured indicator; an SLA is a contract with consequences if a promised level is not met. C An SLO measures cost; an SLA measures speed. D An SLA is internal; an SLO is legal.

Q8

The board has an availability SLO of 99.9% over 30 days. What is the error budget?

A 432 minutes B 43.2 minutes C 4.32 minutes D 7.2 hours

Q9

Halfway through the month, incidents have already used the whole error budget. According to a typical error-budget policy, what happens?

A Risky feature releases are frozen; the team focuses on reliability work until the budget recovers. B Nothing changes. C Monitoring is switched off. D The SLO is raised to 100%.

Q10

A student team plans to switch every user to the new version at once and to "fix forward" if something breaks. Which safer strategy keeps an instant way back?

A Turning off monitoring during the release B Deploying only on Fridays at 17:00 C Blue-green deployment: prepare the new version in a second identical environment and switch traffic, keeping the old one ready to switch back D Deleting the old version before deploying

Q11

Why is 100% availability usually the wrong target?

A Because SLAs forbid it. B It is the right target. C It is extremely expensive, blocks all change, and users cannot notice the difference from 99.9x% because their own networks and devices fail more often. D Because computers cannot count to 100.

Q12

During a SEV1 incident (the board is down during exam week), what should the team do first?

A Wait for users to stop complaining. B Mitigate: restore service (e.g., roll back the last release or switch off the new feature flag), then investigate the root cause. C Find out who caused it. D Write the postmortem.

Q13

What is the key idea of a blameless postmortem?

A Postmortems are secret. B Focus on how the system (processes, tools, defaults) allowed the mistake and delayed detection, so people share information openly and fixes address causes. C Postmortems are written only by managers. D Nobody is responsible for anything.

Q14

StudyBuddy must be changed because the university replaces its single-sign-on system. Which type of maintenance is this?

A Adaptive B Corrective C Preventive D Perfective

Q15

The team upgrades a library before its end-of-life date and adds missing tests to the matching module, although no bug has been reported. This is…

A not maintenance B preventive maintenance C corrective maintenance D perfective maintenance

Q16

Which statement is one of Lehman's laws of software evolution?

A A system in real-world use must be continually adapted or it becomes progressively less satisfactory. B Complexity always decreases over time. C Software never needs to change once it works. D Adding people makes projects faster.

Q17

"We know the matching code is messy, but we must ship before the midterm; we'll refactor it in Sprint 6." In Fowler's quadrant, this debt is…

A prudent and deliberate B reckless and inadvertent C reckless and deliberate D not technical debt

Q18

A debt item costs 12 hours to fix (principal) and wastes 3 hours in every Sprint (interest). After how many Sprints does fixing it pay back?

A 36 B 15 C 1 D 4

Q19

Before migrating an old module (no tests, no documentation) with the help of an AI code translator, what should the team do first?

A Ask the AI whether the translation is correct. B Translate it in one step and deploy to all users. C Write characterization tests that record the current behaviour, so behaviour changes in the migrated code are detected. D Delete the old module.

Q20

What is the strangler fig pattern?

A A type of SQL injection B A performance test C Rewriting the whole system at once over a weekend D Gradually replacing a legacy system by routing functions one by one to new components until the old system can be retired

Q21

According to the IEA, roughly what share of the world's electricity did data centres use in 2024?

A About 0.01% B Exactly 0%, because software has no footprint C About 50% D About 1.5%, with demand projected to roughly double by 2030, driven largely by AI

Q22

StudyBuddy rebuilds its search index every night. Scheduling the job for the hours when the regional grid has the lowest carbon intensity is an example of…

A technical debt B a canary release C hardware efficiency D carbon awareness

Q23

In the SCI formula SCI = ((E × I) + M) per R, what is M?

A The number of monthly users B The embodied emissions of the hardware (from manufacturing), allocated to the software C The memory used D The model size

Q24

Which change most directly reduces the energy (and cost) of Buddy without lowering answer quality below the requirement?

A Call the model twice for every question B Ask the model to write longer answers C Always use the largest available model D Use the smallest model that passes the golden-set thresholds, cache repeated questions, and answer FAQ-type questions by search instead of generation

Q25

A new StudyBuddy version drops support for Android versions older than 3 years, although nothing technical requires it. What is the sustainability concern?

A It improves accessibility B Older phones are faster C Software obsolescence pushes students to replace working phones, increasing electronic waste and excluding students who cannot afford new devices D None

Q26

Under WCAG's Perceivable principle, what is required for images that convey information?

A Nothing; images are optional. B They need text alternatives (alt text) so screen-reader users get the same information. C They must be larger than 1 MB. D They must be animated.

Q27 Short answer

The board's SLO is "99.5% of minutes available over 30 days". This month's incidents lasted 90, 45 and 60 minutes. Compute the error budget, the consumed and remaining budget, and state what the team should do.

Q28 Short answer

Write the timeline and three action items of a blameless postmortem for this incident: at 21:40 on midterm eve, a new Buddy prompt (edited directly in the production configuration by a TA) made Buddy refuse every question; students reported it at 21:55 on WeChat; the team rolled back the prompt at 22:30.

Q29 Short answer

Compute the SCI per answered question for Buddy for one day, then for an improved design, and explain the difference. Day A: servers and AI calls use 12 kWh; grid intensity 600 gCO₂e/kWh; embodied emissions allocated to the day 1,800 gCO₂e; 4,000 questions answered. Day B (smaller model + cache): 5 kWh, same grid and embodied emissions, 4,000 questions.

Q30 Short answer

List six items that must be in StudyBuddy's handover package for next year's student team, and explain in one sentence why handover is a sustainability issue.