observability

Use Pino with JSON output, ISO timestamps, and severity levels instead of console.log. Attach contextual fields like userId and requestId to every log entry so entries can be searched and filtered by log aggregators.

Initialize the OpenTelemetry NodeSDK with an OTLP trace exporter and auto-instrumentations for HTTP, Express, PostgreSQL, and Redis. Add manual spans around custom business logic with attributes and exception recording.

Liveness checks only whether the process is running and triggers container restarts on failure. Readiness checks dependencies like databases and stops traffic routing on failure. Checking the database in liveness causes cascading restarts during outages.

An SLI measures actual performance, an SLO sets the target, and the error budget is 100% minus the SLO. For a 99.9% SLO, the budget allows about 43 minutes of downtime monthly; burning it too fast should freeze deployments.

console.log lacks severity levels, structured fields, and ISO timestamps, making logs unsearchable by aggregators. Production logging requires libraries like Pino for Node.js or structlog for Python with structured JSON output.

Use the RED method: request rate, error rate, and duration percentiles like P50, P95, and P99. For underlying resources, apply the USE method covering utilization, saturation, and errors, plus business metrics like orders per minute.