distributed systems / LEVEL 4

The load test that got faster under overload

EST.45 MIN

THEORY / RETRIEVAL

What to restore

Distinguish open and closed workload models
Measure scheduled-to-finished latency instead of service time alone
Fail admission on dropped arrivals, errors, or an end-to-end p99 breach

A closed loop lets a slow system reduce its own test load

When each virtual user waits for one iteration to finish before starting another, rising latency lowers the arrival rate. That can hide the queueing a fixed external demand would create. An open arrival-rate model schedules work independently and makes unmet starts visible as dropped iterations.

The latency clock starts at scheduled demand

Service time measures started-to-finished work, but users also pay generator and admission queue delay. Release evidence needs scheduled-to-finished latency, errors, achieved rate and dropped arrivals together; successful-request percentiles alone can improve while the system is shedding demand.

CHECKPOINT

A closed-model test reports stable p99 while achieved throughput collapses under overload. Which two changes make the release evidence honest?

Select every applicable option. Credit requires an exact set match.

Confidence: 3/5

02 / DEBUGGING

Find and fix the defect

Repair summarize. Validate every timeline, use nearest-rank p99, report scheduled-to-finished and service latency separately, derive dropped arrivals and milli-RPS without overflow, and pass only with zero drops, zero failures, and end-to-end p99 within SLO.

Initializing editor…

CLOUD SANDBOXnetwork off · 256 MB · 12 s

2 / 64 KB

OUTPUT

Runner is waiting for a submission.