trace a request; distributed tracing with spans + trace ID
the three pillars: metrics, logs, traces; prefer structured logs
watch label cardinality — high/unbounded cardinality is a trap
0 / 5 completed
1 / 5
A backend developer says: "I've added the client library so the service now ___ metrics for every request." Which verb is the standard collocation for a service producing metrics?
Emit metrics — the producer-side verb:
In observability, different actors use different verbs for the same data flowing through the pipeline:
emit / report / expose / publish metrics — the application produces them ("the service emits a counter on each error")
collect / scrape / pull / ingest metrics — the backend (Prometheus, an agent) gathers them
store / persist metrics — the time-series database keeps them
visualise / graph / plot metrics — the dashboard (Grafana) displays them
Related forms: "the endpoint exposes metrics at /metrics", "the app publishes a histogram of request durations". A service emits; a scraper collects.
Why not the distractors? "Render" and "compile" belong to UI and build domains respectively. "Collect" is the wrong side of the pipeline — the service emits, something else collects.
2 / 5
A team lead writes a ticket: "Before we can debug this, we need to ___ the checkout service so we can see what it's doing internally." Which verb means "add observability code"?
Instrument code — the verb for adding telemetry:
To instrument code is to add the hooks that produce telemetry — metrics, traces, and logs — so the system becomes observable from the outside.
"We instrumented the handler with OpenTelemetry spans."
"The library provides auto-instrumentation for HTTP clients."
Noun: "the service has good instrumentation" / "add instrumentation to the DB layer"
Refactor — change internal structure without changing behaviour or adding telemetry
Provision — set up servers, databases, or cloud resources
Orchestrate — coordinate the lifecycle of containers/services (e.g. Kubernetes)
Only instrument means "wire the code up so you can observe it". It is the entry point of every observability discussion.
3 / 5
A platform talk introduces the topic: "Observability rests on ___ — metrics, logs, and traces." Which fixed phrase is the canonical name for this trio?
The three pillars (of observability) — a fixed industry phrase:
"The three pillars of observability" is the established collocation for the trio of telemetry signal types:
Metrics — numeric measurements over time (counters, gauges, histograms): "p99 latency", "error rate"
Logs — timestamped event records; ideally structured logs (key-value/JSON) rather than free text, so they are searchable and filterable
Traces — the path of a single request across services; in distributed tracing each hop is a span with a shared trace ID
Useful collocations: "we trace a request end to end", "distributed tracing shows where the time goes", "switch to structured logs so we can query by field".
Why not the distractors? "Holy trinity" appears jokingly in blog posts but is not the standard term; "three layers" and "core stack" are not idiomatic. When introducing observability concepts, native speakers say the three pillars.
4 / 5
A reliability engineer follows a slow request: "Let me ___ this request through the system to see which service is adding the delay." Which verb names following one request across services?
Trace a request — the observability verb for following one call:
To trace a request is to follow a single call as it travels through every service, queue, and database, so you can see exactly where the time is spent.
"We traced the slow checkout and found the delay in the inventory service."
Distributed tracing — tracing across many services; each step is a span, linked by a shared trace ID, forming the request's trace
Collocations: "follow / propagate the trace context", "the trace shows a 2s gap", "correlate logs by trace ID"
Why this matters: metrics tell you that something is slow; a trace tells you where. That is why tracing is one of the three pillars.
Distractors:poll = repeatedly query for changes; ping = test reachability/round-trip; buffer = temporarily hold data in memory. None mean "follow one request through the system".
5 / 5
A metrics review warns: "Adding user_id as a label exploded the ___ — we now have millions of unique time series." Which term names the number of distinct label combinations?
Cardinality — the count of distinct label combinations:
In metrics systems, cardinality is the number of unique combinations of label values, which equals the number of distinct time series stored. High cardinality is a classic operational trap.
A label like status="200" has low cardinality (a handful of values).
A label like user_id or request_id is unbounded — every value creates a new time series, blowing up storage and query cost.
Collocations: "high / low / unbounded cardinality", "this label explodes the cardinality", "keep cardinality under control", "a cardinality bomb"
Rule of thumb: never put high-cardinality identifiers (user IDs, emails, UUIDs) into metric labels — use logs or traces for that detail instead.
Distractors:granularity is how fine-grained the sampling is (e.g. 1s vs 1m); throughput is requests per second; latency is per-request time. Only cardinality counts distinct series.