Workers Observability review

Workers Observability is the umbrella over Workers Logs, automatic traces, and the query builder that sits on top of both. One block in your Wrangler config and every invocation lands in the dashboard as a structured event, with your console output attached and searchable. Traces are the newer half and they impressed me: fetch calls, binding operations, and handler invocations get instrumented with zero code changes, following OpenTelemetry conventions, with OTLP export to Honeycomb, Grafana, or Axiom when you want the data elsewhere.

The pricing is fair. Paid plans include 20 million log events a month and overage is $0.60 per million, with traces billed from the same style of quota since March 2026. Head-based sampling is a single setting when volume gets expensive. The query language added in early 2026 made the dashboard usable for actual investigation rather than scrolling, and saved queries cover most of my repeat questions.

The ceiling is retention and alerting. Seven days of retention on paid plans means anything you want to keep still needs Logpush or an OTLP pipeline to a real platform, and there is no alerting on log or trace data beyond basic notifications, so it cannot page you when error rates spike. Traces are also still in beta, and the logs pipeline has had stretches where events quietly stopped appearing in the dashboard. For day-to-day debugging it removes the need for an external observability stack on smaller projects, and that alone earns a solid 7.