slog-rs - Structured, composable logging for Rust
Documentation (master branch)
Documentation (release)
Introduction
FAQ
All crates
Testing, feedback, PRs, etc. are very welcome. I'd be also very happy to share the ownership of the project with other people to make it more community-driven.
Long term goal is to make it a go-to logging crate for Rust.
- flexible & easy to use
- great performance; see: slog bench log
#![no_std]
support (with opt-outstd
cargo feature flag)- hierarchical loggers
- lazily evaluated values
- modular, lightweight and very extensible
- tiny core create that does not pull any other dependencies
- feature-crates for specific functionality
- backward compatibility for standard
log
crate (slog-stdlog
crate)- supports logging-scopes
- using slog in library does not force users of the library to use slog
(but gives benefits); see
crates/example-lib
- drains & output formatting
- filtering
- compile-time log level filter using cargo features (same as in
log
crate) - by level, msg, and any other meta-data
slog-envlogger
- port ofenv_logger
- compile-time log level filter using cargo features (same as in
- multiple outputs
- asynchronous IO writing
- terminal output, with color support (
slog-term
crate) - Json (
slog-json
crate)- Bunyan (
slog-bunyan
crate)
- Bunyan (
- syslog (
slog-syslog
crate) - first class custom drains
- filtering
- extensible -
slog
provides core functionality, and some standard feature-set. But using traits, anyone can easily implement as powerful fully-custom features, publish separately and growslog
feature-set for everyone. - composable - Wouldn't it be nice if you could use
env_logger
, but output authentication messages to syslog, while reporting errors over network in json format? Withslog
drains can reuse other drains! You can combine them together, chain, wrap - you name it. - context aware - It's not just one global logger. Hierarchical
loggers carry information about context of logging. When logging an error
condition, you want to know which resource was being handled, on which
instance of your service, using which source code build, talking with what
peer, etc. In standard
log
you would have to repeat this information in every log statement. Inslog
it will happen automatically. See slog-rs functional overview page to understand better logger and drain hierarchies and log record flow through them. - both human and machine readable - By keeping the key-value data format, meaning of logging data is preserved. Dump your logging to a JSON file, and send it to your data-mining system for further analysis. Don't parse it from lines of text anymore!
- lazy evaluation and asynchronous IO included. Waiting to
finish writing logging information to disk, or spending time calculating
data that will be thrown away at the current logging level, are sources of big
performance waste. Use
AsyncStreamer
drain, and closures to make your logging fast. - run-time configuration -
AtomicSwitch
drain allows changing logging behavior in the running program. You could use eg. signal handlers to change logging level or logging destinations. Seesignal
example.
Colors overview:
Compact vs full mode:
Excerpt from examples/features.rs
:
fn main() {
// Create a new drain hierarchy, for the need of your program.
// Choose from collection of existing drains, or write your own
// `struct`-s implementing `Drain` trait.
let drain = slog_term::streamer().async().full().build();
// `AtomicSwitch` is a drain that wraps other drain and allows to change
// it atomically in runtime.
let ctrl = AtomicSwitchCtrl::new(drain);
let drain = ctrl.drain();
// Get a root logger that will log into a given drain.
//
// Note `o!` macro for more natural `OwnedKeyValue` sequence building.
let root = Logger::root(drain.fuse(), o!("version" => VERSION, "build-id" => "8dfljdf"));
// Build logging context as data becomes available.
//
// Create child loggers from existing ones. Children clone `key: value`
// pairs from their parents.
let log = root.new(o!("child" => 1));
// Closures can be used for values that change at runtime.
// Data captured by the closure needs to be `Send+Sync`.
let counter = Arc::new(AtomicUsize::new(0));
let log = log.new(o!("counter" => {
let counter = counter.clone();
// Note the `move` to capture `counter`,
// and unfortunate `|_ : &_|` that helps
// current `rustc` limitations. In the future,
// a `|_|` could work.
move |_ : &Record| { counter.load(SeqCst)}
}));
// Loggers can be cloned, passed between threads and stored without hassle.
let join = thread::spawn({
let log = log.clone();
move || {
info!(log, "before-fetch-add"); // counter == 0
counter.fetch_add(1, SeqCst);
info!(log, "after-fetch-add"); // counter == 1
// `AtomicSwitch` drain can swap it's interior atomically (race-free).
ctrl.set(
// drains are composable and reusable
level_filter(
Level::Info,
async_stream(
std::io::stderr(),
// multiple outputs formats are supported
slog_json::default(),
),
),
);
// Closures can be used for lazy evaluation:
// This `slow_fib` won't be evaluated, as the current drain discards
// "trace" level logging records.
debug!(log, "debug"; "lazy-closure" => |_ : &Record| slow_fib(40));
info!(log, "subthread"; "stage" => "start");
thread::sleep(Duration::new(1, 0));
info!(log, "subthread"; "stage" => "end");
}
});
join.join().unwrap();
}
See examples/features.rs
for full/current code.
Read Documentation for details and features.
See faq for answers to common questions. If you want to say hi, or need help use #slog-rs gitter.im.
To report a bug or ask for features use github issues.
If you need to install Rust (come on, you should have done that long time ago!), use rustup.
In Cargo.toml:
[dependencies]
slog = "1"
In your main.rs
:
#[macro_use]
extern crate slog;
Please fill an issue if slog does not fill your needs. I will appreciate any feedback. You might look into issue discussing slog-rs alternatives too.