golang-samber-slog

name	golang-samber-slog
description	Compose log/slog handlers with samber/slog-multi and related packages for routing, fanout, sampling, formatting, middleware, and safe backend integration.

samber/slog packages

Use log/slog as the stable contract and verify each github.com/samber/slog-* module/version before copying a constructor. A handler is synchronous from the caller's perspective unless a specific backend documents buffering/async behavior; own its flush and close lifecycle.

Composition

Choose the topology that matches the requirement:

• Fanout/broadcast sends one record to every selected handler; it multiplies work and latency.
• Router predicates select handlers; a record may reach several handlers if predicates overlap.
• Pool selects a handler and tries eligible handlers sequentially with fall-through on error. It is load balancing/failover, not broadcast and not parallel execution.
• Pipe applies middleware-style transformations to one downstream handler.

Pool's Enabled checks child handlers and Handle chooses an eligible handler; it does not guarantee that every destination receives every record. Test the failure policy, ordering, and whether a failed sink should be skipped, retried, or surfaced.

Sampling applies to records selected by the sampling handler. It does not automatically exempt errors just because they have a high level; configure a predicate/level policy explicitly and keep a separate unsampled security/audit sink when required. Sampling before expensive formatting can reduce cost, but never sample away events required by compliance or incident response.

A useful conceptual pipeline is context/enrichment, redaction, level routing, sampling policy, formatting, then sinks. The actual package constructors and handler order must match the pinned release.

Context and privacy

Use slog attributes and LogValuer for structured values. Request IDs and trace IDs are linkable identifiers; treat them as potentially personal data and minimize retention. Redact secrets, tokens, credentials, raw bodies, and unnecessary identifiers before fanout. Do not assume self-hosting removes privacy obligations.

HTTP middleware should record method, route template, status, and duration, not arbitrary request bodies. Install a framework-specific middleware only when it can obtain a bounded route and request context. Make health/metrics route filtering explicit.

Backends and lifecycle

Backend modules have different constructor contracts. For a Datadog or other client-backed handler, construct the required client with its documented options, pass that client to the handler, and close/flush it according to the package. Do not present a zero-argument constructor as portable. Network backends need bounded queues, timeout, retry, drop/backpressure policy, and shutdown flush behavior.

Test handlers with an in-memory slog.Handler, assert Enabled and Handle behavior, and exercise WithAttrs/WithGroup. Verify metric/log field names against emitted output rather than stale dashboards. Keep logging failures observable without causing an outage unless the product explicitly treats logs as a required side effect.