Logging

Rust logging is split between a facade and an implementation. The log crate defines macros such as info! and warn! that libraries call, and the application picks one backend at startup: env_logger or pretty_env_logger for simple filtered output, log4rs for file appenders and rotation configured in YAML, and fast_log or logforth for other setups. Libraries should depend only on log or tracing, never on a backend.

tracing, from the Tokio project, is now the common choice for applications. Besides events it records spans, which carry context across await points, so a log line from deep inside an async handler still shows which request it belongs to. tracing-subscriber formats output and filters by level and module, and exporters connect it to OpenTelemetry, Jaeger and similar systems.

slog was the earlier structured logging library and is still used in some older codebases. For a new project, use tracing in applications, especially async ones, and let it collect output from crates that use log through its compatibility layer. If you need JSON output for a log aggregator, tracing-subscriber's JSON formatter handles it without changing call sites, and the same instrumentation can later feed OpenTelemetry.