Observability
OpenTelemetry — the vendor-neutral observability standard
OpenTelemetry in practice: traces, metrics and logs, the collector pattern, manual and auto-instrumentation, and when to use a vendor SDK.
OpenTelemetry (OTel) is the open-source standard for emitting metrics, logs, and traces. Lets you switch backends (Datadog, Honeycomb, Grafana Cloud) without changing app code.
Why OTel matters
- Avoid vendor lock-in
- One instrumentation library, many backends
- Standard semantic conventions (HTTP, DB, cloud)
- Strong OSS ecosystem in 2026
Three pillars in OTel
- Traces — request flow across services
- Metrics — counters, gauges, histograms
- Logs — structured events (newer; less mature)
All flow through a Collector (daemon that batches + routes telemetry).
Architecture pattern
App SDK -> OTel Collector (agent/sidecar) -> Backend(s)Collector lets you fan out to multiple backends, sample, filter, and transform without app changes.
Instrumenting code
from opentelemetry import trace
tracer = trace.get_tracer(__name__)
with tracer.start_as_current_span("checkout") as span:
span.set_attribute("order.id", order_id)Auto-instrumentation
opentelemetry-instrument --traces_exporter otlp python app.pyWhen to choose OTel
- Multi-cloud / multi-vendor environments
- Long-term cost / lock-in concerns
- A/B between observability vendors
- Strong DevOps team to operate the collector
When to stay on vendor SDK
- Single vendor, no plans to migrate
- Small team, don't want to operate collectors
- Vendor SDK has features OTel doesn't
2026 trajectory
OTel is winning. Most major vendors now natively accept OTLP. New instrumentation work should default to OTel.