FDE Foundations · Module 10: Reliability
Logs, Metrics, Traces
Observability is three instruments: logs for the story, metrics for the trend, traces for the path. One correlation ID joins them, and that ID starts at the customer's request.
10 min reading
Objectives
- Structure logs for querying, not reading
- Pick metrics that catch symptoms early
- Trace across service boundaries with one correlation ID
Logs
Structured (JSON), leveled, with the correlation ID, user or tenant context where legal, and the business identifiers that matter (invoice number, batch ID). Log what you would need to reconstruct an incident from the file alone. Never log secrets or full personal data; redact at the boundary.
Metrics
Track the four golden signals per service: latency (p50, p95), traffic, errors (by class), saturation. Add the business metrics that make failures visible before infrastructure fails: batches completed, rows quarantined, model calls failing. Symptom metrics catch what CPU graphs miss.
Traces
One correlation ID from the entry point through every hop: queue, worker, external API, model call. Export spans to wherever the customer can see them; a trace you can query turns a forty-minute incident into a ten-minute one. In field systems, the trace is also the proof of where time goes, which settles latency arguments with facts.
Instrument once, reuse everywhere
Build a small logging/tracing module with the correlation ID and standard fields baked in, and make it the only sanctioned way to log. Consistency is what makes all three instruments searchable.
Quick check
An optional 2-3 question self-check. Answers never leave your device, are not stored, and never count toward any assessment.
Exercise
Define the observability plan for a fictional pipeline service: ten standard log fields, six metrics mixing golden and business signals, and the correlation ID propagation rule across four hops.
Pass criteria
Log fields include IDs and business keys with a redaction rule, metrics split between golden and business with reasons, and propagation rule covers queue and external calls.