A fully asynchronous, contextual logger for Go.
Logr is inspired by Logrus and Zap but addresses a number of issues:
-
Logr is fully asynchronous, meaning that all formatting and writing is done in the background. Latency sensitive applications benefit from not waiting for logging to complete.
-
Logr provides custom filters which provide more flexibility than Trace, Debug, Info... levels. If you need to temporarily increase verbosity of logging while tracking down a problem you can avoid the fire-hose that typically comes from Debug or Trace by using custom filters.
-
Logr generates much less allocations than Logrus, and is close to Zap in allocations.
| entity | description |
|---|---|
| Logr | Engine instance typically instantiated once; used to configure logging.lgr,_ := logr.New() |
| Logger | Provides contextual logging via fields; lightweight, can be created once and accessed globally, or created on demand.logger := lgr.NewLogger()logger2 := logger.With(logr.String("user", "Sam")) |
| Target | A destination for log items such as console, file, database or just about anything that can be written to. Each target has its own filter/level and formatter, and any number of targets can be added to a Logr. Targets for file, syslog and any io.Writer are built-in and it is easy to create your own. You can also use any Logrus hooks via a simple adapter. |
| Filter | Determines which logging calls get written versus filtered out. Also determines which logging calls generate a stack trace.filter := &logr.StdFilter{Lvl: logr.Warn, Stacktrace: logr.Fatal} |
| Formatter | Formats the output. Logr includes built-in formatters for JSON and plain text with delimiters. It is easy to create your own formatters or you can also use any Logrus formatters via a simple adapter.formatter := &format.Plain{Delim: " | "} |
// Create Logr instance.
lgr,_ := logr.New()
// Create a filter and formatter. Both can be shared by multiple
// targets.
filter := &logr.StdFilter{Lvl: logr.Warn, Stacktrace: logr.Error}
formatter := &formatters.Plain{Delim: " | "}
// WriterTarget outputs to any io.Writer
t := targets.NewWriterTarget(filter, formatter, os.StdOut, 1000)
lgr.AddTarget(t)
// One or more Loggers can be created, shared, used concurrently,
// or created on demand.
logger := lgr.NewLogger().With("user", "Sarah")
// Now we can log to the target(s).
logger.Debug("login attempt")
logger.Error("login failed")
// Ensure targets are drained before application exit.
lgr.Shutdown()Fields allow for contextual logging, meaning information can be added to log statements without changing the statements themselves. Information can be shared across multiple logging statements thus allowing log analysis tools to group them.
Fields can be added to a Logger via Logger.With or included with each log record:
lgr,_ := logr.New()
// ... add targets ...
logger := lgr.NewLogger().With(
logr.Any("user": user),
logr.String("role", role)
)
logger.Info("login attempt", logr.Int("attempt_count", count))
// ... later ...
logger.Info("login", logr.String("result", result))Logr fields are inspired by and work the same as Zap fields.
Logr supports the traditional seven log levels via logr.StdFilter: Panic, Fatal, Error, Warning, Info, Debug, and Trace.
// When added to a target, this filter will only allow
// log statements with level severity Warn or higher.
// It will also generate stack traces for Error or higher.
filter := &logr.StdFilter{Lvl: logr.Warn, Stacktrace: logr.Error}Logr also supports custom filters (logr.CustomFilter) which allow fine grained inclusion of log items without turning on the fire-hose.
// create custom levels; use IDs > 10.
LoginLevel := logr.Level{ID: 100, Name: "login ", Stacktrace: false}
LogoutLevel := logr.Level{ID: 101, Name: "logout", Stacktrace: false}
lgr,_ := logr.New()
// create a custom filter with custom levels.
filter := &logr.CustomFilter{}
filter.Add(LoginLevel, LogoutLevel)
formatter := &formatters.Plain{Delim: " | "}
tgr := targets.NewWriterTarget(filter, formatter, os.StdOut, 1000)
lgr.AddTarget(tgr)
logger := lgr.NewLogger().With(logr.String("user": "Bob"), logr.String("role": "admin"))
logger.Log(LoginLevel, "this item will get logged")
logger.Debug("won't be logged since Debug wasn't added to custom filter")Both filter types allow you to determine which levels force a stack trace to be output. Note that generating stack traces cannot happen fully asynchronously and thus add some latency to the calling goroutine.
There are built-in targets for outputting to syslog, file, TCP, or any io.Writer. More will be added.
You can use any Logrus hooks via a simple adapter.
You can create your own target by implementing the simple Target interface.
Example target that outputs to io.Writer:
type Writer struct {
out io.Writer
}
func NewWriterTarget(out io.Writer) *Writer {
w := &Writer{out: out}
return w
}
// Called once to initialize target.
func (w *Writer) Init() error {
return nil
}
// Write will always be called by a single internal Logr goroutine, so no locking needed.
func (w *Writer) Write(p []byte, rec *logr.LogRec) (int, error) {
return w.out.Write(buf.Bytes())
}
// Called once to cleanup/free resources for target.
func (w *Writer) Shutdown() error {
return nil
}Logr has two built-in formatters, one for JSON and the other plain, delimited text.
You can use any Logrus formatters via a simple adapter.
You can create your own formatter by implementing the Formatter interface:
Format(rec *LogRec, stacktrace bool, buf *bytes.Buffer) (*bytes.Buffer, error)When creating the Logr instance, you can set configuration options. For example:
lgr, err := logr.New(
logr.MaxQueueSize(1000),
logr.StackFilter("mypackage1", "mypackage2"),
)Some options are documented below. See options.go for all available configuration options.
Called any time an internal logging error occurs. For example, this can happen when a target cannot connect to its data sink.
It may be tempting to log this error, however there is a danger that logging this will simply generate another error and so on. If you must log it, use a target and custom level specifically for this event and ensure it cannot generate more errors.
Called on an attempt to add a log record to a full Logr queue. This generally means the Logr maximum queue size is too small, or at least one target is very slow. Logr maximum queue size can be changed before adding any targets via:
lgr, err := logr.New(logr.MaxQueueSize(2000))Returning true will drop the log record. False will block until the log record can be added, which creates a natural throttle at the expense of latency for the calling goroutine. The default is to block.
Called on an attempt to add a log record to a full target queue. This generally means your target's max queue size is too small, or the target is very slow to output.
As with the Logr queue, returning true will drop the log record. False will block until the log record can be added, which creates a natural throttle at the expense of latency for the calling goroutine. The default is to block.
OnExit and OnPanic are called when the Logger.FatalXXX and Logger.PanicXXX functions are called respectively.
In both cases the default behavior is to shut down gracefully, draining all targets, and calling os.Exit or panic respectively.
When adding your own handlers, be sure to call Logr.Shutdown before exiting the application to avoid losing log records.
StackFilter sets a list of package names to exclude from the top of stack traces. The Logr packages are automatically filtered.