Golang Slog

Golang slog

Use this skill when implementing structured logging in Go with the standard library log/slog package.

Use it alongside the structured-logging skill:

• structured-logging defines the cross-language schema, field naming, safety, and observability rules
• golang-slog shows how to implement those rules idiomatically in Go with log/slog

When to Use

• Adding structured logging to a Go service or CLI
• Replacing ad-hoc log.Printf calls with structured events
• Standardising common log fields across Go services
• Building request-scoped, worker-scoped, or job-scoped loggers
• Implementing redaction, log schema normalisation, or dynamic log levels

Default Stance

Unless there is a clear reason not to, prefer these defaults:

• Use the standard library log/slog for Go 1.21+
• Prefer a Format config such as json or text, rather than a bare JSON boolean
• Use JSONHandler in production and TextHandler locally
• Construct the logger once during application startup
• Attach stable application metadata once with Logger.With(...)
• Keep the schema aligned with the structured-logging skill
• Prefer typed attrs like slog.String, slog.Int, slog.Int64, slog.Duration, slog.Bool
• Prefer InfoContext / ErrorContext / LogAttrs when a context.Context is available
• Use WithGroup or slog.Group(...) for nested structured context
• Use ReplaceAttr and LogValuer for schema normalisation and redaction
• Pass *slog.Logger as a dependency instead of hiding it in global state when possible

Version Notes

• log/slog is in the Go standard library starting with Go 1.21
• slog.MultiHandler is a newer addition; only use it if your project's Go version supports it
• If your codebase targets older Go versions, use your existing logging stack instead of forcing slog

Schema Alignment with structured-logging

The structured-logging skill recommends canonical fields like:

• timestamp
• level
• message
• service
• environment
• application_version
• application_commit
• request_id
• trace_id
• span_id
• user_id
• operation
• outcome
• duration_ms
• error_kind
• error_code
• error_message

By default, slog emits built-in keys such as:

• time
• level
• msg
• source

If you want Go logs to match the shared schema exactly, rename built-in keys with HandlerOptions.ReplaceAttr.

Recommended Logger Bootstrap

Centralise logger construction in main or a small bootstrap package.

package logging

import (
    "fmt"
    "io"
    "log/slog"
)

type Format string

const (
    FormatJSON Format = "json"
    FormatText Format = "text"
)

type config struct {
    environment        string
    service            string
    applicationVersion string
    applicationCommit  string
    format             Format
    addSource          bool
    level              slog.Leveler
    replaceAttr        func(groups []string, a slog.Attr) slog.Attr
}

type Option func(*config)

func WithEnvironment(v string) Option {
    return func(c *config) { c.environment = v }
}

func WithService(v string) Option {
    return func(c *config) { c.service = v }
}

func WithApplicationVersion(v string) Option {
    return func(c *config) { c.applicationVersion = v }
}

func WithApplicationCommit(v string) Option {
    return func(c *config) { c.applicationCommit = v }
}

func WithFormat(v Format) Option {
    return func(c *config) { c.format = v }
}

func WithAddSource(v bool) Option {
    return func(c *config) { c.addSource = v }
}

func WithLevel(v slog.Leveler) Option {
    return func(c *config) { c.level = v }
}

func WithReplaceAttr(fn func(groups []string, a slog.Attr) slog.Attr) Option {
    return func(c *config) { c.replaceAttr = fn }
}

func LevelVarFromString(v string) (*slog.LevelVar, error) {
    var level slog.LevelVar
    if err := level.UnmarshalText([]byte(v)); err != nil {
        return nil, fmt.Errorf("parse log level %q: %w", v, err)
    }
    return &level, nil
}

func New(w io.Writer, opts ...Option) (*slog.Logger, error) {
    cfg := config{
        format: FormatJSON,
        level:  slog.LevelInfo,
        replaceAttr: func(groups []string, a slog.Attr) slog.Attr {
            switch a.Key {
            case slog.TimeKey:
                a.Key = "timestamp"
            case slog.MessageKey:
                a.Key = "message"
            }
            return a
        },
    }

    for _, opt := range opts {
        opt(&cfg)
    }

    handlerOptions := &slog.HandlerOptions{
        AddSource:   cfg.addSource,
        Level:       cfg.level,
        ReplaceAttr: cfg.replaceAttr,
    }

    var handler slog.Handler
    switch cfg.format {
    case FormatJSON:
        handler = slog.NewJSONHandler(w, handlerOptions)
    case FormatText:
        handler = slog.NewTextHandler(w, handlerOptions)
    default:
        return nil, fmt.Errorf("unsupported log format: %q", cfg.format)
    }

    logger := slog.New(handler).With(
        slog.String("service", cfg.service),
        slog.String("environment", cfg.environment),
        slog.String("application_version", cfg.applicationVersion),
        slog.String("application_commit", cfg.applicationCommit),
    )

    return logger, nil
}

Example use:

level, err := logging.LevelVarFromString("info")
if err != nil {
    return err
}

logger, err := logging.New(os.Stdout,
    logging.WithService("users-api"),
    logging.WithEnvironment("production"),
    logging.WithApplicationVersion("1.8.0"),
    logging.WithApplicationCommit("a1b2c3d4"),
    logging.WithFormat(logging.FormatJSON),
    logging.WithLevel(level),
)
if err != nil {
    return err
}

Guidance:

• Build one base logger at startup
• Prefer functional options when the bootstrap needs several optional settings
• Add service, environment, application_version, and application_commit there
• Prefer a format enum/string like json or text over a JSON bool
• Use JSONHandler for production ingestion
• Use TextHandler only when human readability is more important than strict machine parsing

Global vs Injected Logger

Prefer:

• Constructing the logger at startup
• Injecting *slog.Logger into servers, services, workers, and repositories that need it

Avoid:

• Scattering calls to slog.Default() throughout the codebase without a reason
• Using package globals for everything

SetDefault is acceptable at the process boundary if you explicitly want top-level slog.Info(...) and legacy log.Print(...) calls to flow through the same handler, but keep that decision near startup.

Common Attribute Strategy

Stable Application Metadata

Attach these once to the base logger:

baseLogger := logger.With(
    slog.String("service", "users-api"),
    slog.String("environment", "production"),
    slog.String("application_version", "1.8.0"),
    slog.String("application_commit", "a1b2c3d4"),
)

Request- or Job-Scoped Metadata

Attach these at the request/job boundary:

requestLogger := baseLogger.With(
    slog.String("request_id", requestID),
    slog.String("trace_id", traceID),
    slog.String("span_id", spanID),
    slog.String("user_id", userID),
)

Prefer deriving child loggers with With(...) rather than repeating the same attrs on every line.

Context Usage

If a context.Context is available, prefer the ...Context forms:

• logger.InfoContext(ctx, ...)
• logger.ErrorContext(ctx, ...)
• logger.LogAttrs(ctx, ...)

Why:

• custom handlers may extract trace/span metadata from the context
• it keeps logs aligned with request-scoped and tracing-aware systems

Rules:

• pass context when you already have one
• do not create fake contexts just to log
• do not treat context as a generic service locator

HTTP Request Pattern

Derive a request logger at the boundary and reuse it for the whole flow.

func (s *Server) handleGetUser(w http.ResponseWriter, r *http.Request) {
    start := time.Now()

    logger := s.logger.With(
        slog.String("request_id", requestIDFromRequest(r)),
        slog.String("trace_id", traceIDFromContext(r.Context())),
        slog.String("http_method", r.Method),
        slog.String("http_route", "/users/{id}"),
    )

    user, err := s.userService.GetByID(r.Context(), chi.URLParam(r, "id"))
    if err != nil {
        logger.ErrorContext(r.Context(), "get user failed",
            slog.String("operation", "get_user"),
            slog.String("outcome", "failure"),
            slog.String("error_kind", "user_lookup_failed"),
            slog.String("error_message", "failed to fetch user"),
            slog.Int64("duration_ms", time.Since(start).Milliseconds()),
        )
        return
    }

    logger.InfoContext(r.Context(), "http request completed",
        slog.String("operation", "get_user"),
        slog.String("outcome", "success"),
        slog.Int("http_status_code", http.StatusOK),
        slog.Int64("duration_ms", time.Since(start).Milliseconds()),
    )

    _ = user
}

Keep boundary logs summary-oriented. Avoid logging every internal function call in production.

Prefer Typed Attrs

Prefer:

logger.Info("user authenticated",
    slog.String("user_id", userID),
    slog.String("operation", "login"),
    slog.Int64("duration_ms", duration.Milliseconds()),
)

Over large alternating key/value calls when they get dense or performance-sensitive.

Notes:

• alternating key/value syntax is valid and ergonomic for small calls
• typed attrs are clearer in larger calls
• typed attrs also avoid mistakes that produce !BADKEY

Hot Paths and LogAttrs

If profiling shows logging overhead matters, prefer LogAttrs with typed attrs.

logger.LogAttrs(ctx, slog.LevelInfo, "cache refresh completed",
    slog.String("operation", "refresh_cache"),
    slog.String("outcome", "success"),
    slog.Int64("duration_ms", duration.Milliseconds()),
)

Use this especially in high-frequency code paths.

Grouping Attributes

Use groups when nested context improves clarity or prevents key collisions.

logger.Info("request finished",
    slog.Group("request",
        slog.String("method", r.Method),
        slog.String("route", "/users/{id}"),
    ),
    slog.Group("auth",
        slog.String("user_id", userID),
        slog.String("actor_type", "user"),
    ),
)

Use WithGroup("http") or slog.Group("http", ...) when multiple subsystems would otherwise reuse keys like id, status, or method.

Error Logging

Log errors as structured events, not only as formatted strings.

Prefer:

logger.ErrorContext(ctx, "invoice payment failed",
    slog.String("operation", "charge_invoice"),
    slog.String("outcome", "failure"),
    slog.String("error_kind", "payment_gateway_error"),
    slog.String("error_code", "card_declined"),
    slog.String("error_message", "payment provider declined charge"),
    slog.String("user_id", userID),
    slog.Int64("duration_ms", duration.Milliseconds()),
)

Guidance:

• prefer stable domain fields like error_kind and error_code
• keep error_message safe and sanitised
• include the raw error value only when useful and safe
• avoid logging the same failure at every layer

When you do need the raw error object:

logger.ErrorContext(ctx, "dependency call failed",
    slog.Any("error", err),
    slog.String("dependency", "billing-api"),
)

JSONHandler formats error-valued attrs as strings.

Redaction and Sensitive Data

Never log secrets directly.

Use LogValuer for type-level redaction:

type Token string

func (Token) LogValue() slog.Value {
    return slog.StringValue("REDACTED")
}

Use ReplaceAttr for cross-cutting redaction or key normalisation:

opts := &slog.HandlerOptions{
    ReplaceAttr: func(groups []string, a slog.Attr) slog.Attr {
        if a.Key == "authorization" || a.Key == "token" {
            return slog.String(a.Key, "REDACTED")
        }
        return a
    },
}

Prefer type-local redaction with LogValuer when the data type itself is sensitive. Use ReplaceAttr when you need a handler-wide policy.

Performance Guidance

• Use Logger.With(...) for common attrs instead of repeating them
• Prefer LogAttrs in hot paths
• Use typed attrs for common scalar values
• Avoid expensive string formatting before checking whether a level is enabled
• Pass values directly when possible instead of eagerly calling .String()
• Use LogValuer to defer expensive computation for debug logs

Example:

if logger.Enabled(ctx, slog.LevelDebug) {
    logger.DebugContext(ctx, "computed plan",
        slog.Any("plan", buildDebugPlan(input)),
    )
}

Avoid:

logger.DebugContext(ctx, "computed plan",
    slog.String("plan", buildDebugPlan(input).String()),
)

if buildDebugPlan is expensive and debug logging may be disabled.

Dynamic Levels

Use slog.LevelVar when the level may change at runtime.

var level = new(slog.LevelVar)
level.Set(slog.LevelInfo)

logger := slog.New(slog.NewJSONHandler(os.Stdout, &slog.HandlerOptions{
    Level: level,
}))

If your level comes from config, flags, or environment variables, a small helper is reasonable:

func LevelVarFromString(v string) (*slog.LevelVar, error) {
    var level slog.LevelVar
    if err := level.UnmarshalText([]byte(v)); err != nil {
        return nil, fmt.Errorf("parse log level %q: %w", v, err)
    }
    return &level, nil
}

This is appropriate for CLIs, services with admin-configurable verbosity, or temporary incident debugging.

Bridging the Old log Package

If you still have legacy log.Print / log.Printf calls, you can centralise output with:

slog.SetDefault(logger)

That allows the default log package to emit through the configured slog handler.

Use this only as a deliberate application-wide choice, not as a substitute for dependency injection.

Testing Guidance

For tests around logging:

• write to a bytes.Buffer
• use ReplaceAttr to remove or normalise time/source fields
• assert on structured output, not on fragile formatting details
• prefer JSON output in tests when machine-readable assertions are easier

Example:

var buf bytes.Buffer

logger := slog.New(slog.NewJSONHandler(&buf, &slog.HandlerOptions{
    ReplaceAttr: func(groups []string, a slog.Attr) slog.Attr {
        if a.Key == slog.TimeKey {
            return slog.Attr{}
        }
        return a
    },
}))

Review Checklist

When reviewing Go slog usage, check for:

• log/slog used consistently instead of mixed ad-hoc logging styles
• base logger created centrally at startup
• stable common attrs attached with With(...)
• shared schema aligned with the structured-logging skill
• ReplaceAttr used when canonical key renaming is needed
• InfoContext / ErrorContext used when context already exists
• LogAttrs used in hot paths where it matters
• structured error fields instead of only formatted strings
• redaction for secrets and sensitive values
• no repeated noisy logs for the same failure across layers
• JSON logs used for production ingestion

Anti-Patterns

Avoid these unless there is a strong reason:

• logging with fmt.Sprintf into one giant string
• using slog.Any for everything when typed attrs are clearer
• eagerly computing expensive debug fields when debug is disabled
• forcing every function to call slog.Default()
• burying the logger in context as a universal service locator
• logging secrets, tokens, or raw auth headers
• leaving built-in keys as time/msg when the rest of the platform expects timestamp/message
• emitting text logs in production when downstream systems expect JSON
• logging every internal function call instead of boundary summaries

Quick Template

level := new(slog.LevelVar)
level.Set(slog.LevelInfo)

logger := slog.New(slog.NewJSONHandler(os.Stdout, &slog.HandlerOptions{
    Level: level,
    ReplaceAttr: func(groups []string, a slog.Attr) slog.Attr {
        switch a.Key {
        case slog.TimeKey:
            a.Key = "timestamp"
        case slog.MessageKey:
            a.Key = "message"
        }
        return a
    },
})).With(
    slog.String("service", "users-api"),
    slog.String("environment", "production"),
    slog.String("application_version", "1.0.0"),
    slog.String("application_commit", "abcdef12"),
)

logger.Info("user authenticated",
    slog.String("request_id", requestID),
    slog.String("trace_id", traceID),
    slog.String("user_id", userID),
    slog.String("operation", "login"),
    slog.String("outcome", "success"),
    slog.Int64("duration_ms", duration.Milliseconds()),
)

Source: brpaz/agent-skills — distributed by TomeVault.