M2.4 module quiz
Ten questions. 80% to pass (8 of 10). Open book, unlimited retakes.
Answers are collapsed under each question. Answer first, then check.
Q1
The autoscaler smart default for a CPU target is:
A. A three-way tier on P95: above 85 gives 70, above 60 gives 75, otherwise 80
B. Always exactly 70%, on every single resource regardless of its measured P95
C. P95 minus a flat 5 percentage-point margin
D. The mean plus one standard deviation, clamped
Show answer
Correct: A. Min capacity derives from average CPU, not P05, which is never computed. Max is max(count+2, ceil(count × P99 / target × 1.3)).
Q2
Welford-derived P90, P95 and P99 are:
A. True percentiles computed from the full retained sample
B. Median-based estimates, robust to bursty outliers
C. Exact only for uniform distributions, undefined otherwise
D. A Gaussian approximation computed in one pass
Show answer
Correct: D. P95 is approximately mean plus 1.64 standard deviations, commented as a normal approximation. It is least accurate exactly where it matters most, on bursty workloads, which is when to override the default.
Q3
The three autoscaler modes are:
A. off, on and auto
B. monitor, recommend, autopilot
C. manual, semi-auto, full-auto
D. observe, suggest, execute
Show answer
Correct: B. Mode is auto-derived from the cloud account’s permission level: read_only accounts get monitor, read_write accounts can reach recommend or autopilot.
Q4
Target tracking and step scaling differ in that target tracking:
A. Holds a metric at a setpoint; step scaling applies defined adjustments at thresholds
B. Is faster to converge on the target
C. Requires no cooldown period whatsoever between consecutive scaling actions on the group
D. Works only on AWS scaling groups
Show answer
Correct: A. Target tracking is the right default because it needs one number rather than a ladder. Step scaling earns its complexity when the response should be non-linear.
Q5
Cooldown exists to:
A. Reduce the cost of the scaling API calls
B. Give the new instances enough time to finish booting
C. Satisfy the provider’s own rate limits
D. Prevent oscillation between scale-out and scale-in
Show answer
Correct: D. A cooldown shorter than the workload’s response time makes the autoscaler react to its own previous action, which is the classic thrashing failure.
Q6
Removing an autoscaler policy:
A. Restores the previous configuration saved at apply time
B. Leaves the cloud-side configuration exactly in place
C. Deletes the scaling group along with the policy
D. Requires manual cleanup in the cloud console
Show answer
Correct: A. Previous config is captured on apply and restored on remove, which is what makes Replace reversible. Adopted policies skip the write path entirely and have nothing to restore.
Q7
Editing an adopted policy:
A. Is blocked while the policy is adopted
B. Silently flips its source to recommended, with a re-apply prompt
C. Creates a second, parallel ZopNight policy
D. Reverts the adoption and restores the original cloud policy
Show answer
Correct: B. The update response carries a flag so the UI can surface a “re-apply for changes to take effect” banner. The auto-promote is the escape hatch from observation into ownership.
Q8
Quick Setup differs from Advanced mode in that it:
A. Supports considerably fewer cloud providers than the full policy creation flow
B. Applies immediately without any review step
C. Fills min, max, target and cooldown from smart defaults and creates in a single request
D. Skips validation of the resulting policy
Show answer
Correct: C. It is one screen for the common case. Advanced exists for the cases smart defaults cannot reason about, such as resource-group targeting.
Q9
The per-policy event log records:
A. Only the failed actions, along with all of their error messages here
B. Apply, pause, resume, remove and scaling actions with status and errors
C. Metric values at each evaluation
D. Cost changes after each action
Show answer
Correct: B. It is what answers “why did we scale at 3 AM”, which is the question the metrics alone cannot resolve because they do not record the decision.
Q10
ECS autoscaling uses:
A. The EC2 Auto Scaling API, called into directly
B. The ECS service API on its own
C. Application Auto Scaling, a separate API surface
D. A ZopNight-managed controller pod
Show answer
Correct: C. Policies are prefixed for safe identification on Remove, because the two API surfaces are genuinely different and a shared prefix is what makes cleanup unambiguous.
What’s next
Back to VM autoscaling.