Datadog APM & Distributed Tracing Fundamentals key terms

A record of a single request as it travels through multiple services, composed of spans linked by shared trace identifiers so the full end-to-end path can be reconstructed and timed.

A timed unit of work inside a trace, such as one HTTP request or database call, carrying a name, resource, service, start time, duration, and tags.

A 64- or 128-bit identifier generated at the entry point of a request and propagated to every downstream service so all related spans group into one distributed trace.

A unique identifier assigned to an individual span, used together with the parent span ID to rebuild the call hierarchy within a trace.

The span that directly caused another span to be created, establishing the nesting relationship that turns flat span records into a call tree.

The mechanism of passing trace and span identifiers across process and network boundaries, typically through HTTP headers, message attributes, or gRPC metadata, so downstream work joins the same trace.

The act of writing trace identifiers into outgoing request headers, such as x-datadog-trace-id and x-datadog-parent-id, or the W3C traceparent header, before a call leaves the service.

An open standard using the traceparent and tracestate headers to carry trace identity between systems, supported by Datadog alongside its native propagation styles for cross-vendor interoperability.

A language-specific package, such as dd-trace for Java, Python, Node.js, Go, or .NET, that instruments applications, collects spans, and ships them to the Datadog Agent.

Out-of-the-box library integrations that create spans for common frameworks, HTTP clients, and database drivers without code changes, enabled by default or through configuration in the tracing library.

Adding custom spans or tags in application code, using the tracing library API, to capture business logic or internal operations that automatic instrumentation does not cover.

A span created explicitly by the developer, for example with a tracer's trace or startSpan method, to represent an operation such as order validation or report generation.

A key-value attribute attached to a span, such as http.status_code or a business identifier, which enriches traces and enables filtering, grouping, and analytics in the APM UI.

The practice of setting env, service, and version on all telemetry so traces, logs, metrics, and profiles correlate; the version tag also powers deployment tracking and error tracking.

A named group of traces and spans produced by one application or component, defined by the service tag and used as the primary unit for APM dashboards, monitors, and the Service Catalog.

The specific operation a span represents, such as an HTTP route, SQL statement, or queue topic, used to break down a service's performance by endpoint or query.

Precomputed metrics derived from all ingested spans, including request count, error rate, and latency distributions, retained for fifteen months and queryable regardless of whether individual traces were kept.

A rule that decides which traces are kept for full inspection versus only contributing to trace metrics, controlling storage cost and sampling of retained spans.

Client-side or Agent-side rate limiting that reduces the volume of spans sent to Datadog, applied after trace metrics are generated so aggregate statistics stay accurate.

A visual graph of services and their dependencies, showing request rates, error rates, and latency on each edge, generated automatically from observed traces and their parent-child relationships.

A directional relationship where one service calls another, inferred from spans whose parent belongs to a different service, and displayed as an edge on the Service Map.

A visualization of a trace where span bars are stacked by depth and width represents duration, making it easy to spot the slowest or most frequently repeated operations in a request.

A trace visualization that lists spans chronologically with horizontal bars showing start time and duration, revealing sequential waits and parallel work across services.

Querying individual retained traces by service, resource, status, duration, or tag using the APM search syntax, used to find concrete examples of an error or slow request.

The share of requests for a service or resource that produced an error span, computed from trace metrics and often displayed alongside request throughput and latency percentiles.

A histogram of request durations for a service or resource, used to read percentiles such as p50, p95, and p99 rather than relying on averages that hide tail latency.

A value below which a given share of measurements fall, for example p99 latency being the duration that 99 percent of requests finish within, useful for exposing tail latency.

The Datadog feature that groups similar errors from traces and logs into issues, tracks their frequency across versions, and links each issue to example traces and stack traces.

The deployment version attached to telemetry; comparing versions in APM reveals whether a new release changed error rates or latency, and enables deployment tracking and rollback decisions.

Overlaying version change events on APM graphs so performance shifts can be tied to specific releases, helping confirm or rule out a deploy as the cause of an incident.

The detailed page for a single trace showing its spans, durations, tags, host information, and related logs, used to inspect exactly where time was spent or an error originated.

A span or trace attribute with special meaning to Datadog, such as service, resource, or http.status_code, which drives built-in facets, metrics, and UI behavior and should not be repurposed.

The environment variable that tells a tracing library where to send spans, typically the local Datadog Agent's trace intake endpoint, and the first thing to check when traces never appear.

The component of the Datadog Agent that receives spans from tracing libraries, computes trace metrics, applies sampling and filtering, and forwards retained traces to Datadog.

A named instrumentation module in a tracing library that wraps a specific framework, client, or driver, and can be enabled, disabled, or configured individually.

Injecting trace and span identifiers into application logs so a log line can be linked directly to the trace that produced it, speeding root-cause analysis during incidents.

A Datadog product that samples code-level CPU and memory usage in production and links profiles to traces, showing which functions consumed resources during a slow request.

An alert configured on trace metrics, such as error rate or p95 latency for a service, that notifies teams when performance crosses a defined threshold.