Java Logging Structured

职业

分类

监控

Preferred format: JSON

Every log record should be a single JSON object with stable keys.

Recommended baseline fields

• timestamp (ISO-8601)
• level (TRACE/DEBUG/INFO/WARN/ERROR)
• logger (class/category)
• thread
• message
• service.name
• service.version
• deployment.environment (dev/staging/prod)
• traceId (if tracing enabled)
• spanId (if tracing enabled)
• requestId (if available)
• correlationId (if you use a custom correlation id)
• http.method, http.path, http.status (for request logs)
• event.name (semantic event label; stable)
• error.type, error.message (safe), error.stack (internal-only, never to clients)

Field stability rules

• Do not rename fields without migration plan.
• Prefer adding new fields rather than changing meaning.
• Avoid high-cardinality fields unless explicitly needed:
  • Never log raw user identifiers; hash or redact.
  • Never log full request/response bodies by default.

Correlation strategy

What identifiers to use

1. Prefer traceId/spanId (from tracing context).
2. Also include a requestId (gateway id or generated at ingress).
3. For async flows, propagate correlation through message headers:
   • include traceparent (W3C) and optional correlationId.

How to implement (Java pattern)

• Use MDC (Mapped Diagnostic Context) to attach correlation fields per thread/request.
• At request entry:
  • extract incoming traceparent / requestId if present
  • otherwise generate requestId
  • set MDC keys: traceId, spanId, requestId
• Ensure MDC cleanup at request end.

Async and thread hops

• MDC is thread-local by default. For executors:
  • wrap Runnable/Callable to copy MDC context
  • or use framework/otel context propagation when available.

Logging policy (levels)

Default guidance

• INFO: business milestones (state transitions), start/stop, essential events.
• WARN: recoverable anomalies, retries, degraded mode.
• ERROR: failures requiring attention, request failed.
• DEBUG/TRACE: development-only, guarded by config; never enable globally in prod.

Anti-pattern: "log everything at INFO"

This creates noise, cost, and hides signals.

Security and privacy

Redaction rules (must-have)

• Redact credentials: Authorization headers, tokens, API keys.
• Redact secrets: passwords, private keys, session cookies.
• Redact or hash PII: email, phone, address, national id.
• Never log full JWTs; at most log token fingerprint (hash).

Safe exception logging

• Log error class and sanitized message.
• Store full stack traces only in internal logs with restricted access, or sample them.

Sampling and rate limiting

• For high-volume endpoints:
  • sample INFO request logs (e.g., 1%)
  • always keep WARN/ERROR
• Rate limit repetitive error logs to avoid log storms.

Standard log events (recommended)

• event.name=request.received
• event.name=request.completed
• event.name=dependency.call
• event.name=dependency.failed
• event.name=job.started
• event.name=job.completed
• event.name=security.auth.failed (without leaking details)

Outputs / artifacts

• docs/logging.md (schema + policies)
• logging/log-schema.json (optional JSON schema)
• logging/redaction-rules.md
• Code changes:
  • ingress filter/middleware for MDC
  • JSON encoder configuration
  • unit tests for redaction and MDC cleanup

Definition of Done (DoD)

• Logs are structured (JSON/logfmt) and consistent.
• Correlation identifiers present for requests and background jobs.
• Redaction rules applied and tested.
• Sampling/rate-limiting strategy documented.
• No secret/PII leakage found in test logs.

Guardrails (What NOT to do)

• Never log raw secrets or tokens.
• Never dump full request/response by default.
• Avoid unbounded string concatenation in hot paths (use lazy logging).
• Avoid logging in tight loops without rate limiting.

Common failure modes & fixes

• Symptom: cannot correlate logs with traces -> Fix: include traceId and propagate context; add MDC.
• Symptom: log storms during incidents -> Fix: add sampling + rate limiting; reduce noisy INFO logs.
• Symptom: leaked secrets -> Fix: redaction filters; code review checklist; scanners.

References (see references/)

• references/owasp-logging-rules.md
• references/log-schema-fields.md
• references/mdc-propagation.md