Observability & Operations

Logging & Distributed Tracing

Logs capture discrete events with full detail; traces stitch a single request's path across every service it touched into one timeline — together they answer "what happened" and "where did the time go" that metrics alone cannot.

Structured logging (key-value/JSON, not free-form text) is what makes logs queryable at scale — grep does not scale past one machine
A trace is a tree of spans; each span records one unit of work with a start time, duration, and parent span — the whole tree reconstructs the request's path
Context propagation — a trace ID and span ID passed through every hop, in headers or message metadata — is what lets independently-deployed services contribute spans to the same trace
Log correlation IDs, often the trace ID itself, let you jump from "this span was slow" to the exact log lines from that service during that span
Sampling is unavoidable at scale — tracing every request is often too expensive, so systems sample a percentage, or trace every erroring/slow request unconditionally ("tail-based sampling")
Metrics tell you something is wrong; traces tell you where; logs tell you why — they are complementary, not substitutes for each other

One request, one trace, many spans

Logs vs metrics vs traces
	Logs	Metrics	Traces
Question answered	what exactly happened	is something wrong, in aggregate	where did the time go
Granularity	single event	aggregated over time	single request across services
Storage cost	high — full detail	low — pre-aggregated	high, mitigated by sampling

Propagating trace context across a call

Span span = tracer.spanBuilder("chargeCard").setParent(Context.current()).startSpan();
try (Scope scope = span.makeCurrent()) {
    httpClient.send(request.header("traceparent", span.getSpanContext().toTraceparent()));
} finally {
    span.end();
}