Observability / Distributed tracing

Adds contextual metadata to a trace span for later analysis.

When instrumenting a function call to record its outcome; when annotating a unit of work with business context (e.g., user ID, request ID); when enriching spans for debugging.

Applications using OpenTelemetry, Jaeger, Zipkin, or similar tracing SDKs; microservices, serverless functions, web backends.

with tracer.start_as_current_span(operation_name) as span: span.set_attribute(attribute_key, attribute_value)

with ... as ...: ... .set_attribute(...)

operation_name: str, attribute_key: str, attribute_value: Any

• tracer
• start_as_current_span
• set_attribute
• Span
• context propagation

• Using span.add_event instead of set_attribute for timed annotations
• using baggage for cross-process propagation.

Assuming attributes automatically propagate to child spans; forgetting to set attributes before ending the span; using non-string keys where the backend expects strings.

Span events (add_event) – timestamped annotations vs. static attributes; Baggage propagation – key-value pairs sent downstream vs. attributes attached to a single span.

Coming from logging libraries: treating span attributes like log fields and expecting them to appear in logs; they are only exported with trace data.

• distributed tracing
• span events
• baggage propagation
• trace context injection

Attributes are immutable after the span ends; high-cardinality attributes can increase storage cost; some backends restrict attribute value types.

Makes trace data searchable and enables precise filtering in observability tools.

Think 'tag your span' – attach metadata like sticky notes.

Backends often impose limits on attribute key length and value cardinality; avoid high‑cardinality attributes to reduce storage and query cost.

Use baggage propagation for cross-service attribute forwarding: set baggage in context and retrieve it in downstream spans.