Operational Characteristics Quiz

15 questions on the characteristics that concern the running system: availability, performance, scalability, elasticity, reliability and recoverability.

Availability and reliability

#

A service targets 99.99% availability. Roughly how much downtime does that allow in a 30-day month?

About 4 minutes

Correct. Four nines leaves roughly 4 minutes 20 seconds per month, which effectively rules out manual failover.

About 43 minutes

About 22 minutes

About 26 seconds

#

A request requires four services in series, each independently 99.95% available. What is the approximate availability of the request?

About 99.8%, since availabilities multiply along the chain

Correct. 0.9995⁴ ≈ 0.998, so each synchronous dependency lowers the ceiling.

99.95%, set by the weakest link

99.99%, since redundancy compounds

It cannot be estimated from these numbers

#

Which change most improves availability once a system is already at three nines?

Cutting MTTR with automated detection and failover

Correct. At scale failures are inevitable, so recovery time dominates the remaining budget.

Raising MTBF by buying more reliable hardware

Adding a second reviewer to every pull request

Increasing client-side request timeouts

#

A payment service always responds within 100 ms but occasionally charges a customer twice. Which characteristic is deficient?

Reliability, since responses are fast but not correct

Correct. Availability is about answering; reliability is about answering correctly, every time.

Availability, since duplicate charges imply downtime

Elasticity, since duplicates come from scaling

Recoverability, since the charge can be refunded

#

Why do idempotency keys appear in almost every reliable distributed design?

Because retries and redelivery are unavoidable, and idempotency makes repeated delivery harmless

Correct. Exactly-once delivery is impractical; exactly-once effect through deduplication is achievable.

Because they encrypt the request payload

Because they remove the need for transactions

Because they guarantee message ordering

#

A downstream dependency becomes slow rather than unavailable. Without circuit breakers and timeouts, what typically happens to the caller?

Threads and connections pile up waiting, and the caller fails too, spreading the outage

Correct. Slow dependencies exhaust pools; failing fast and shedding load contains the blast radius.

The caller automatically routes around the dependency

Only the affected requests fail, with no wider effect

Latency improves as the queue absorbs the delay

Performance, scalability, elasticity

#

Which requirement is properly specified for performance?

"Checkout submission completes in under 400 ms at p99, at 2000 concurrent users, in production"

Correct. It names the operation, the percentile, the load, and the environment.

"The average response time must be under one second"

"The system must feel fast to users"

"No endpoint may ever exceed 400 ms"

#

A team moves email sending, invoicing, and analytics off the request path onto a queue. Which statement is most accurate?

Perceived latency improves because the response no longer waits for work that does not need to be synchronous

Correct. Moving work off the request path is the standard latency tactic, at the cost of eventual completion semantics.

Total system throughput necessarily drops

The system becomes strongly consistent

Latency is unchanged because the work still happens

#

A design requires five sequential cross-continent round trips per request. What does this tell you?

There is a latency floor of several hundred milliseconds that no code tuning will remove; the interaction pattern must change

Correct. Physical round trips are relocated or eliminated, never optimised away.

A faster runtime or language will solve it

Adding more instances will fix the latency

Caching database queries is sufficient

#

Why must instances be stateless for horizontal scaling to work well?

So any instance can serve any request, letting instances be added, replaced, or destroyed freely

Correct. In-process state forces sticky routing and prevents free redistribution of load.

Because stateless request handling is inherently faster

Because load balancers cannot maintain TCP connections otherwise

Because databases reject concurrent writers otherwise

#

5% of a workload runs inside a global lock. What is the ceiling on speedup from adding workers?

About 20×, because the serial fraction bounds the achievable speedup

Correct. Amdahl's law: the limit is 1/s, so 1/0.05 = 20.

Unbounded, since the other 95% parallelises

About 5×, matching the serial percentage

About 95×, matching the parallel percentage

#

A ticketing system handles steady traffic but must absorb a 20× spike within two minutes when sales open. Which characteristic is being demanded?

Elasticity, the ability to acquire capacity automatically at spike speed

Correct. Scalability is about sustained growth; elasticity is about sudden bursts.

Scalability, since more users must be supported

Availability, since the site must stay up

Recoverability, since the spike may cause failures

#

Which autoscaling signal is usually best for an I/O-bound service that spends most of its time waiting on downstream calls?

Concurrency or requests in flight, since CPU stays low while the service is saturated

Correct. The signal must reflect the real bottleneck, and for I/O-bound work CPU is misleading.

CPU utilisation, as the platform default

Memory usage, which tracks connection buffers

Instance count, kept stable to avoid churn

Recoverability

#

"We can tolerate 15 minutes of downtime but no more than 1 minute of lost data." Which objectives are these?

RTO of 15 minutes and RPO of 1 minute

Correct. RTO bounds downtime and drives failover strategy; RPO bounds data loss and drives replication mode.

RPO of 15 minutes and RTO of 1 minute

MTBF of 15 minutes and MTTR of 1 minute

An availability SLO of 99.9% with a 1-minute alert threshold

#

Which recovery mechanism protects against the most common real-world "disaster"?

A fast, rehearsed rollback of the release that was just deployed

Correct. Most incidents originate in the latest change, so reversibility of deployments has the highest recovery value per unit of effort.

Cross-region active-active replication

Quarterly offsite tape archives

A larger standby database instance

Quiz Progress

• Answered: 0 / 0 (0%)
• Correct: 0 / 0 (0%) Reset