observability

To get useful traces, you need propagation and a real exporter. I set a global TextMapPropagator (TraceContext + Baggage) so inbound headers connect spans across services. Then I configure an OTLP exporter and a batch span processor so tracing overhea

In-flight request count is a simple but powerful saturation signal. I keep an atomic.Int64 gauge that increments at the start of a request and decrements in a defer, which makes it robust even on early returns. This gauge can be exported via expvar, P

I’ve started using log/slog for services that want structured logs without a heavy dependency. The key is treating the request logger as data: create a base logger with JSON output, then derive a request logger with fields like request_id and path. I

Email issues are painful in production. Subscribe to mailer notifications and log message IDs, recipients, and durations. This gives you a lightweight audit trail without adding a heavy dependency.

Structured logs are great, but at high QPS they can become their own outage: too much IO, too much storage, and noisy dashboards. zap includes a sampler that allows you to keep early logs and then sample at a fixed rate. I like this for “successful re

You’ll eventually hit non-200 responses or network flakiness. Hook Turbo events to log failures (and optionally show a toast). This is a pragmatic production addition that helps debugging without adding heavy tooling.

Set a rough query budget per request in dev/test and alert when exceeded. This is a pragmatic way to keep performance regressions visible without requiring a full APM setup.

In production, stack traces without context are slow to debug. Raise custom exceptions that include IDs and safe metadata, and log them as JSON. This is the difference between “we think” and “we know”.