Observability & Resilience
Measure user impact, limit admission, bound waiting work, budget retries, then probe recovery. Each component must reduce pressure under overload.
1 / 6
Measure user impact, limit admission, bound waiting work, budget retries, then probe recovery. Each component must reduce pressure under overload.
Minutes [100,100,100,300], [100], [400,100]; threshold200; target99%.
Overall burn=(2/7)/0.01≈28.57; newest minute burn=50; newest two=33.33.
Nearest-rank p99=400. Both alert thresholds must breach.
Capacity5, initial2, arrivals5, service4: 2+5=7 → shed2 → peak5 → serve4 → end1.
Serving first gives a different answer. Conservation: initial + arrivals = processed + dropped + final.
Start10, delays3 then cap5, outcomes F/F/F/T → [10,13,18,23].
The exercise omits durations, deadlines and jitter.
Three attempts at three retrying layers can multiply to 27 downstream attempts.
Threshold2, cooldown5: failures at0/1 open; success at3 is rejected and ignored. Failure at6 is an allowed probe and restarts cooldown. Success at11 probes and closes. State reports after evaluation.
Admission80, service40 per tick: extra queue capacity delays shedding but cannot fix sustained deficit. Design a shared retry deadline and explain how concurrent callers would claim one probe permit.