Knowledge base
CodexGuild Knowledge Base

Observability 2026: OpenTelemetry everywhere, metrics on logs

as of Jun 15, 2026 · applies to otel >= 1.x · canonical · codexguild.com/kb/kb-observability-otel-2026 · exported 2026-10-11
Canonical as of Jun 15, 2026

Observability 2026: OpenTelemetry everywhere, metrics on logs

OpenTelemetry is the settled standard: traces+metrics+logs via one SDK, semantic conventions for AI/LLM spans (genai), OTLP into any backend. Vendor agents (Datadog/Grafana/Honeycomb) all consume OTel natively.

Observability with OpenTelemetry in 2026

As of: 2026-06

The state

  • OTel won. Traces, metrics, logs (and profiles joining) through one SDK + OTLP protocol; every serious backend ingests it natively. The vendor SDK is a lock-in relic.
  • Semantic conventions matured — including GenAI conventions: LLM/model spans with token counts, tool-call spans, agent-run attributes. AI-app telemetry is standardized, not homegrown.
  • Collectors as DaemonSets/gateways: tail-sampling, redaction (scrub secrets before export), cost routing (traces to A, logs to B).

The minimum for a service

  1. OTel SDK auto-instrumentation (HTTP server/client, DB, queue).
  2. service.name/version resource attributes + deployment env.
  3. Trace-linked logs (trace_id in every log line).
  4. RED metrics per endpoint + one SLO per user journey.
  5. For AI apps: genai-convention spans — per-model-call latency, token usage, cost attributes. Cost observability is 2026's new table stake.