slog

slog is a minimal structured logging library for Go.

Install

go get cdr.dev/slog

Features

• Minimal API
• First class context.Context support
• First class testing.TB support
  • Package slogtest/assert provides test assertion helpers
• Beautiful human readable logging output
  • Prints multiline fields and errors nicely
• Machine readable JSON output
• GCP Stackdriver support
• Stdlib log adapter
• Skip caller frames with slog.Helper
• Encodes values as if with json.Marshal
• Transparently log opencensus trace and span IDs
• Single dependency on go.opencensus.io
• Log to multiple sinks

Example

Many more examples available at godoc.

log := slog.Make(sloghuman.Sink(os.Stdout))

log.Info(context.Background(), "my message here",
    slog.F("field_name", "something or the other"),
    slog.F("some_map", slog.M(
        slog.F("nested_fields", time.Date(2000, time.February, 5, 4, 4, 4, 0, time.UTC)),
    )),
    slog.Error(
        xerrors.Errorf("wrap1: %w",
            xerrors.Errorf("wrap2: %w",
                io.EOF,
            ),
        ),
    ),
)

Why?

At Coder we’ve used Uber’s zap for several years. It is a fantastic library for performance. Thanks Uber!

However we felt the API and developer experience could be improved.

Here is a list of reasons how we improved on zap with slog.

1. slog has a minimal API surface

   • Compare slog to zap and zapcore.
   • The sprawling API makes zap hard to understand, use and extend.

2. slog has a concise semi typed API

   • We found zap's fully typed API cumbersome. It does offer a sugared API but it's too easy to pass an invalid fields list since there is no static type checking. Furthermore, it's harder to read as there is no syntax grouping for each key value pair.
   • We wanted an API that only accepted the equivalent of zap.Any for every field. This is slog.F.

3. sloghuman uses a very human readable format

   • It colors distinct parts of each line to make it easier to scan logs. Even the JSON that represents the fields in each log is syntax highlighted so that is very easy to scan. See the screenshot above.
     • zap lacks appropriate colors for different levels and fields.
   • slog automatically prints one multiline field after the log to make errors and such much easier to read.
     • zap logs multiline fields and errors stack traces as JSON strings which made them unreadable in a terminal.
   • When logging to JSON, slog automatically converts a golang.org/x/xerrors chain into an array with fields for the location and wrapping messages.

4. Full context.Context support

   • slog lets you set fields in a context.Context such that any log with the context prints those fields.
   • We wanted to be able to pull up all relevant logs for a given trace, user or request. With zap, we were plugging these fields in for every relevant log or passing around a logger with the fields set. This became very verbose.

5. Simple and easy to extend

   • A new backend only has to implement the simple Sink interface.
   • The Logger type provides a nice API around Sink but also implements Sink to allow for composition.
   • zap is hard and confusing to extend. There are too many structures and configuration options.

6. Structured logging of Go structures with json.Marshal

   • Entire encoding process is documented on godoc.
   • With zap, We found ourselves often implementing zap's ObjectMarshaler to log Go structures. This was verbose and most of the time we ended up only implementing fmt.Stringer and using zap.Stringer instead.

7. slog takes inspiration from Go's stdlib and implements slog.Helper which works just like t.Helper

   • It marks the calling function as a helper and skips it when reporting location info.
   • We had many helper functions for logging but we wanted the line reported to be of the parent function. zap has an API for this but it's verbose and requires passing the logger around explicitly.

8. Tight integration with stdlib's testing package

   • You can configure slogtest to exit on any ERROR logs and it has a global stateless API that takes a testing.TB so you do not need to create a logger first.
   • Test assertion helpers are provided in slogtest/assert.
   • zap has zaptest but the API surface is large and doesn't integrate well. It does not support any of the features described above.