OpenTelemetry

Overview

OpenTelemetry (OTel) is the unified observability standard for instrumenting applications with traces, metrics, and logs. It supports auto-instrumentation across Node.js, Python, Java, and Go, and exports telemetry to backends like Jaeger, Grafana, Datadog, and Honeycomb through a flexible Collector pipeline.

Instructions

• When adding tracing, create spans with meaningful names, set span kinds (CLIENT, SERVER, PRODUCER, CONSUMER), add business-relevant attributes, and use W3C Trace Context for propagation.
• When adding metrics, choose the right instrument type: Counter for monotonic values, Histogram for distributions like latency, UpDownCounter for fluctuating values, and Gauge for point-in-time readings.
• When setting up auto-instrumentation, use the language-specific packages (@opentelemetry/auto-instrumentations-node, opentelemetry-instrumentation for Python, etc.) to capture HTTP, database, and messaging spans without code changes.
• When configuring the OTel Collector, define pipelines with receivers (OTLP, Prometheus), processors (batch, memory_limiter, tail_sampling), and exporters (OTLP, Jaeger, Datadog) in the collector config.
• When deploying Collectors, choose sidecar mode for per-pod collection, agent mode for per-node, or gateway mode for centralized processing.
• When setting resource attributes, always include service.name, service.version, and deployment.environment, and use cloud/container resource detectors for infrastructure metadata.
• When naming attributes, follow OTel semantic conventions (http.request.method, db.system, messaging.system) instead of inventing custom names.

Examples

Example 1: Add distributed tracing to a Node.js microservice