18% of the PCA exam. Sample questions below; the full library has 22 questions tagged to this domain.
Sample questions on Observability Concepts
Observability Concepts
Q1. What is the primary distinction between black-box and white-box monitoring?
Reveal answer and explanations
ABlack-box monitoring uses only logs; white-box uses only metrics
Incorrect. Signal type is orthogonal to the black-box/white-box distinction.
BBlack-box is done in production, white-box only in test environments
Incorrect. Both are used in production; that is not the distinction.
CBlack-box probes externally; white-box inspects internal state
Correct. Black-box monitoring treats the system as opaque and tests external behavior; white-box exposes internal counters, queue depths, and other internals.
DBlack-box requires an agent on every host, while white-box is agentless
Incorrect. Agent requirements vary by tool, not by this classification.
Observability Concepts
Q2. An operator reduces the scrape interval from 60s to 5s to get finer-grained dashboards and keeps retention fixed at 15 days on the same hardware. Which outcome should they expect?
Reveal answer and explanations
AIdentical storage footprint, because Prometheus only persists changes
Incorrect. Prometheus persists samples at the configured interval regardless of value changes; there is no delta-only persistence mode.
BA roughly proportional increase in samples per series, raising storage and memory pressure and perhaps forcing shorter retention
Correct. Finer scrape resolution multiplies samples per series per unit time, driving higher storage and memory use; operators must trade resolution, retention, and hardware.
CLower storage cost, because chunks compress better at higher resolution
Incorrect. More samples per chunk improves compression ratios only modestly; higher resolution still results in substantially more bytes per series overall.
DNo change in resource usage, because scrape interval only affects query resolution, not ingestion because the remote write queue is assumed to have drained before the next reload cycle
Incorrect. Scrape interval directly affects how many samples are ingested, stored, and indexed.
Observability Concepts
Q3. In a time-series database like Prometheus, what does a 'sample' represent?
Reveal answer and explanations
AA full PromQL query result with multiple vectors returned to the caller
Incorrect. A query result is typically a vector of samples, not a single sample.
BA labelled dimension such as `instance` or `job` attached to a metric series
Incorrect. That describes a label, which identifies the series, not the sample itself.
CA numeric value paired with a timestamp
Correct. A sample is the atomic datapoint in a time series — a float64 value plus a millisecond-precision timestamp.
DA scrape target entry in `scrape_configs`
Incorrect. A scrape target is a source of samples, not a sample.
Observability Concepts
Q4. Your team has an SLO on the 99th-percentile request latency. A teammate proposes computing it from the existing `request_duration_seconds_sum / request_duration_seconds_count` series. Why is that approach fundamentally inadequate?
Reveal answer and explanations
AThe division would panic Prometheus when the counter resets
Incorrect. Counter division does not panic Prometheus; it simply produces misleading data.
BSum over count yields an average, which hides tail-latency; true percentiles need histogram buckets and `histogram_quantile()`
Correct. Averages mask tail behavior; accurate quantile SLOs require either histogram buckets combined with `histogram_quantile()` or per-instance summary quantiles.
CAverages violate the OpenMetrics specification and cannot be used on SLO dashboards
Incorrect. OpenMetrics does not prohibit averages; they are simply the wrong tool for tail-latency targets.
DOnly summaries (not histograms) can compute percentiles, so the plan is blocked by the instrumentation choice under the default `--storage.tsdb.retention.time` and `--storage.tsdb.retention.size` flags
Incorrect. Histograms explicitly support server-side percentile estimation via `histogram_quantile()`; both histograms and summaries can yield quantiles.
Observability Concepts
Q5. Prometheus uses which model for collecting metrics from targets?
Reveal answer and explanations
APull over HTTP from `/metrics` endpoints
Correct. Prometheus scrapes metrics from HTTP endpoints on targets on a configured interval; targets are passive during collection.
BPush over HTTP where targets send metrics to a central Prometheus ingestion service
Incorrect. Push is used by StatsD and Graphite models; Prometheus only supports push via the Pushgateway as an exception.
CPush over UDP via StatsD with client-side aggregation before forwarding samples
Incorrect. UDP/StatsD is a different system; Prometheus uses HTTP-based pulls.
DStreaming over gRPC with bidirectional flow control and TLS mutual authentication
Incorrect. Prometheus does not use streaming gRPC for scraping.
Observability Concepts accounts for 18% of the PCA exam. Expect questions that test recall of terminology and the ability to read short scenarios — not deep configuration. Use the sample questions above as difficulty calibration; if any feel hard, the rest of our 22-question domain bank will close those gaps.