...

Source file src/go.uber.org/zap/logger.go

Documentation: go.uber.org/zap

     1  // Copyright (c) 2016 Uber Technologies, Inc.
     2  //
     3  // Permission is hereby granted, free of charge, to any person obtaining a copy
     4  // of this software and associated documentation files (the "Software"), to deal
     5  // in the Software without restriction, including without limitation the rights
     6  // to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
     7  // copies of the Software, and to permit persons to whom the Software is
     8  // furnished to do so, subject to the following conditions:
     9  //
    10  // The above copyright notice and this permission notice shall be included in
    11  // all copies or substantial portions of the Software.
    12  //
    13  // THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
    14  // IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
    15  // FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
    16  // AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
    17  // LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
    18  // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
    19  // THE SOFTWARE.
    20  
    21  package zap
    22  
    23  import (
    24  	"fmt"
    25  	"io"
    26  	"os"
    27  	"strings"
    28  
    29  	"go.uber.org/zap/internal/bufferpool"
    30  	"go.uber.org/zap/internal/stacktrace"
    31  	"go.uber.org/zap/zapcore"
    32  )
    33  
    34  // A Logger provides fast, leveled, structured logging. All methods are safe
    35  // for concurrent use.
    36  //
    37  // The Logger is designed for contexts in which every microsecond and every
    38  // allocation matters, so its API intentionally favors performance and type
    39  // safety over brevity. For most applications, the SugaredLogger strikes a
    40  // better balance between performance and ergonomics.
    41  type Logger struct {
    42  	core zapcore.Core
    43  
    44  	development bool
    45  	addCaller   bool
    46  	onPanic     zapcore.CheckWriteHook // default is WriteThenPanic
    47  	onFatal     zapcore.CheckWriteHook // default is WriteThenFatal
    48  
    49  	name        string
    50  	errorOutput zapcore.WriteSyncer
    51  
    52  	addStack zapcore.LevelEnabler
    53  
    54  	callerSkip int
    55  
    56  	clock zapcore.Clock
    57  }
    58  
    59  // New constructs a new Logger from the provided zapcore.Core and Options. If
    60  // the passed zapcore.Core is nil, it falls back to using a no-op
    61  // implementation.
    62  //
    63  // This is the most flexible way to construct a Logger, but also the most
    64  // verbose. For typical use cases, the highly-opinionated presets
    65  // (NewProduction, NewDevelopment, and NewExample) or the Config struct are
    66  // more convenient.
    67  //
    68  // For sample code, see the package-level AdvancedConfiguration example.
    69  func New(core zapcore.Core, options ...Option) *Logger {
    70  	if core == nil {
    71  		return NewNop()
    72  	}
    73  	log := &Logger{
    74  		core:        core,
    75  		errorOutput: zapcore.Lock(os.Stderr),
    76  		addStack:    zapcore.FatalLevel + 1,
    77  		clock:       zapcore.DefaultClock,
    78  	}
    79  	return log.WithOptions(options...)
    80  }
    81  
    82  // NewNop returns a no-op Logger. It never writes out logs or internal errors,
    83  // and it never runs user-defined hooks.
    84  //
    85  // Using WithOptions to replace the Core or error output of a no-op Logger can
    86  // re-enable logging.
    87  func NewNop() *Logger {
    88  	return &Logger{
    89  		core:        zapcore.NewNopCore(),
    90  		errorOutput: zapcore.AddSync(io.Discard),
    91  		addStack:    zapcore.FatalLevel + 1,
    92  		clock:       zapcore.DefaultClock,
    93  	}
    94  }
    95  
    96  // NewProduction builds a sensible production Logger that writes InfoLevel and
    97  // above logs to standard error as JSON.
    98  //
    99  // It's a shortcut for NewProductionConfig().Build(...Option).
   100  func NewProduction(options ...Option) (*Logger, error) {
   101  	return NewProductionConfig().Build(options...)
   102  }
   103  
   104  // NewDevelopment builds a development Logger that writes DebugLevel and above
   105  // logs to standard error in a human-friendly format.
   106  //
   107  // It's a shortcut for NewDevelopmentConfig().Build(...Option).
   108  func NewDevelopment(options ...Option) (*Logger, error) {
   109  	return NewDevelopmentConfig().Build(options...)
   110  }
   111  
   112  // Must is a helper that wraps a call to a function returning (*Logger, error)
   113  // and panics if the error is non-nil. It is intended for use in variable
   114  // initialization such as:
   115  //
   116  //	var logger = zap.Must(zap.NewProduction())
   117  func Must(logger *Logger, err error) *Logger {
   118  	if err != nil {
   119  		panic(err)
   120  	}
   121  
   122  	return logger
   123  }
   124  
   125  // NewExample builds a Logger that's designed for use in zap's testable
   126  // examples. It writes DebugLevel and above logs to standard out as JSON, but
   127  // omits the timestamp and calling function to keep example output
   128  // short and deterministic.
   129  func NewExample(options ...Option) *Logger {
   130  	encoderCfg := zapcore.EncoderConfig{
   131  		MessageKey:     "msg",
   132  		LevelKey:       "level",
   133  		NameKey:        "logger",
   134  		EncodeLevel:    zapcore.LowercaseLevelEncoder,
   135  		EncodeTime:     zapcore.ISO8601TimeEncoder,
   136  		EncodeDuration: zapcore.StringDurationEncoder,
   137  	}
   138  	core := zapcore.NewCore(zapcore.NewJSONEncoder(encoderCfg), os.Stdout, DebugLevel)
   139  	return New(core).WithOptions(options...)
   140  }
   141  
   142  // Sugar wraps the Logger to provide a more ergonomic, but slightly slower,
   143  // API. Sugaring a Logger is quite inexpensive, so it's reasonable for a
   144  // single application to use both Loggers and SugaredLoggers, converting
   145  // between them on the boundaries of performance-sensitive code.
   146  func (log *Logger) Sugar() *SugaredLogger {
   147  	core := log.clone()
   148  	core.callerSkip += 2
   149  	return &SugaredLogger{core}
   150  }
   151  
   152  // Named adds a new path segment to the logger's name. Segments are joined by
   153  // periods. By default, Loggers are unnamed.
   154  func (log *Logger) Named(s string) *Logger {
   155  	if s == "" {
   156  		return log
   157  	}
   158  	l := log.clone()
   159  	if log.name == "" {
   160  		l.name = s
   161  	} else {
   162  		l.name = strings.Join([]string{l.name, s}, ".")
   163  	}
   164  	return l
   165  }
   166  
   167  // WithOptions clones the current Logger, applies the supplied Options, and
   168  // returns the resulting Logger. It's safe to use concurrently.
   169  func (log *Logger) WithOptions(opts ...Option) *Logger {
   170  	c := log.clone()
   171  	for _, opt := range opts {
   172  		opt.apply(c)
   173  	}
   174  	return c
   175  }
   176  
   177  // With creates a child logger and adds structured context to it. Fields added
   178  // to the child don't affect the parent, and vice versa. Any fields that
   179  // require evaluation (such as Objects) are evaluated upon invocation of With.
   180  func (log *Logger) With(fields ...Field) *Logger {
   181  	if len(fields) == 0 {
   182  		return log
   183  	}
   184  	l := log.clone()
   185  	l.core = l.core.With(fields)
   186  	return l
   187  }
   188  
   189  // WithLazy creates a child logger and adds structured context to it lazily.
   190  //
   191  // The fields are evaluated only if the logger is further chained with [With]
   192  // or is written to with any of the log level methods.
   193  // Until that occurs, the logger may retain references to objects inside the fields,
   194  // and logging will reflect the state of an object at the time of logging,
   195  // not the time of WithLazy().
   196  //
   197  // WithLazy provides a worthwhile performance optimization for contextual loggers
   198  // when the likelihood of using the child logger is low,
   199  // such as error paths and rarely taken branches.
   200  //
   201  // Similar to [With], fields added to the child don't affect the parent, and vice versa.
   202  func (log *Logger) WithLazy(fields ...Field) *Logger {
   203  	if len(fields) == 0 {
   204  		return log
   205  	}
   206  	return log.WithOptions(WrapCore(func(core zapcore.Core) zapcore.Core {
   207  		return zapcore.NewLazyWith(core, fields)
   208  	}))
   209  }
   210  
   211  // Level reports the minimum enabled level for this logger.
   212  //
   213  // For NopLoggers, this is [zapcore.InvalidLevel].
   214  func (log *Logger) Level() zapcore.Level {
   215  	return zapcore.LevelOf(log.core)
   216  }
   217  
   218  // Check returns a CheckedEntry if logging a message at the specified level
   219  // is enabled. It's a completely optional optimization; in high-performance
   220  // applications, Check can help avoid allocating a slice to hold fields.
   221  func (log *Logger) Check(lvl zapcore.Level, msg string) *zapcore.CheckedEntry {
   222  	return log.check(lvl, msg)
   223  }
   224  
   225  // Log logs a message at the specified level. The message includes any fields
   226  // passed at the log site, as well as any fields accumulated on the logger.
   227  // Any Fields that require  evaluation (such as Objects) are evaluated upon
   228  // invocation of Log.
   229  func (log *Logger) Log(lvl zapcore.Level, msg string, fields ...Field) {
   230  	if ce := log.check(lvl, msg); ce != nil {
   231  		ce.Write(fields...)
   232  	}
   233  }
   234  
   235  // Debug logs a message at DebugLevel. The message includes any fields passed
   236  // at the log site, as well as any fields accumulated on the logger.
   237  func (log *Logger) Debug(msg string, fields ...Field) {
   238  	if ce := log.check(DebugLevel, msg); ce != nil {
   239  		ce.Write(fields...)
   240  	}
   241  }
   242  
   243  // Info logs a message at InfoLevel. The message includes any fields passed
   244  // at the log site, as well as any fields accumulated on the logger.
   245  func (log *Logger) Info(msg string, fields ...Field) {
   246  	if ce := log.check(InfoLevel, msg); ce != nil {
   247  		ce.Write(fields...)
   248  	}
   249  }
   250  
   251  // Warn logs a message at WarnLevel. The message includes any fields passed
   252  // at the log site, as well as any fields accumulated on the logger.
   253  func (log *Logger) Warn(msg string, fields ...Field) {
   254  	if ce := log.check(WarnLevel, msg); ce != nil {
   255  		ce.Write(fields...)
   256  	}
   257  }
   258  
   259  // Error logs a message at ErrorLevel. The message includes any fields passed
   260  // at the log site, as well as any fields accumulated on the logger.
   261  func (log *Logger) Error(msg string, fields ...Field) {
   262  	if ce := log.check(ErrorLevel, msg); ce != nil {
   263  		ce.Write(fields...)
   264  	}
   265  }
   266  
   267  // DPanic logs a message at DPanicLevel. The message includes any fields
   268  // passed at the log site, as well as any fields accumulated on the logger.
   269  //
   270  // If the logger is in development mode, it then panics (DPanic means
   271  // "development panic"). This is useful for catching errors that are
   272  // recoverable, but shouldn't ever happen.
   273  func (log *Logger) DPanic(msg string, fields ...Field) {
   274  	if ce := log.check(DPanicLevel, msg); ce != nil {
   275  		ce.Write(fields...)
   276  	}
   277  }
   278  
   279  // Panic logs a message at PanicLevel. The message includes any fields passed
   280  // at the log site, as well as any fields accumulated on the logger.
   281  //
   282  // The logger then panics, even if logging at PanicLevel is disabled.
   283  func (log *Logger) Panic(msg string, fields ...Field) {
   284  	if ce := log.check(PanicLevel, msg); ce != nil {
   285  		ce.Write(fields...)
   286  	}
   287  }
   288  
   289  // Fatal logs a message at FatalLevel. The message includes any fields passed
   290  // at the log site, as well as any fields accumulated on the logger.
   291  //
   292  // The logger then calls os.Exit(1), even if logging at FatalLevel is
   293  // disabled.
   294  func (log *Logger) Fatal(msg string, fields ...Field) {
   295  	if ce := log.check(FatalLevel, msg); ce != nil {
   296  		ce.Write(fields...)
   297  	}
   298  }
   299  
   300  // Sync calls the underlying Core's Sync method, flushing any buffered log
   301  // entries. Applications should take care to call Sync before exiting.
   302  func (log *Logger) Sync() error {
   303  	return log.core.Sync()
   304  }
   305  
   306  // Core returns the Logger's underlying zapcore.Core.
   307  func (log *Logger) Core() zapcore.Core {
   308  	return log.core
   309  }
   310  
   311  // Name returns the Logger's underlying name,
   312  // or an empty string if the logger is unnamed.
   313  func (log *Logger) Name() string {
   314  	return log.name
   315  }
   316  
   317  func (log *Logger) clone() *Logger {
   318  	clone := *log
   319  	return &clone
   320  }
   321  
   322  func (log *Logger) check(lvl zapcore.Level, msg string) *zapcore.CheckedEntry {
   323  	// Logger.check must always be called directly by a method in the
   324  	// Logger interface (e.g., Check, Info, Fatal).
   325  	// This skips Logger.check and the Info/Fatal/Check/etc. method that
   326  	// called it.
   327  	const callerSkipOffset = 2
   328  
   329  	// Check the level first to reduce the cost of disabled log calls.
   330  	// Since Panic and higher may exit, we skip the optimization for those levels.
   331  	if lvl < zapcore.DPanicLevel && !log.core.Enabled(lvl) {
   332  		return nil
   333  	}
   334  
   335  	// Create basic checked entry thru the core; this will be non-nil if the
   336  	// log message will actually be written somewhere.
   337  	ent := zapcore.Entry{
   338  		LoggerName: log.name,
   339  		Time:       log.clock.Now(),
   340  		Level:      lvl,
   341  		Message:    msg,
   342  	}
   343  	ce := log.core.Check(ent, nil)
   344  	willWrite := ce != nil
   345  
   346  	// Set up any required terminal behavior.
   347  	switch ent.Level {
   348  	case zapcore.PanicLevel:
   349  		ce = ce.After(ent, terminalHookOverride(zapcore.WriteThenPanic, log.onPanic))
   350  	case zapcore.FatalLevel:
   351  		ce = ce.After(ent, terminalHookOverride(zapcore.WriteThenFatal, log.onFatal))
   352  	case zapcore.DPanicLevel:
   353  		if log.development {
   354  			ce = ce.After(ent, terminalHookOverride(zapcore.WriteThenPanic, log.onPanic))
   355  		}
   356  	}
   357  
   358  	// Only do further annotation if we're going to write this message; checked
   359  	// entries that exist only for terminal behavior don't benefit from
   360  	// annotation.
   361  	if !willWrite {
   362  		return ce
   363  	}
   364  
   365  	// Thread the error output through to the CheckedEntry.
   366  	ce.ErrorOutput = log.errorOutput
   367  
   368  	addStack := log.addStack.Enabled(ce.Level)
   369  	if !log.addCaller && !addStack {
   370  		return ce
   371  	}
   372  
   373  	// Adding the caller or stack trace requires capturing the callers of
   374  	// this function. We'll share information between these two.
   375  	stackDepth := stacktrace.First
   376  	if addStack {
   377  		stackDepth = stacktrace.Full
   378  	}
   379  	stack := stacktrace.Capture(log.callerSkip+callerSkipOffset, stackDepth)
   380  	defer stack.Free()
   381  
   382  	if stack.Count() == 0 {
   383  		if log.addCaller {
   384  			fmt.Fprintf(log.errorOutput, "%v Logger.check error: failed to get caller\n", ent.Time.UTC())
   385  			_ = log.errorOutput.Sync()
   386  		}
   387  		return ce
   388  	}
   389  
   390  	frame, more := stack.Next()
   391  
   392  	if log.addCaller {
   393  		ce.Caller = zapcore.EntryCaller{
   394  			Defined:  frame.PC != 0,
   395  			PC:       frame.PC,
   396  			File:     frame.File,
   397  			Line:     frame.Line,
   398  			Function: frame.Function,
   399  		}
   400  	}
   401  
   402  	if addStack {
   403  		buffer := bufferpool.Get()
   404  		defer buffer.Free()
   405  
   406  		stackfmt := stacktrace.NewFormatter(buffer)
   407  
   408  		// We've already extracted the first frame, so format that
   409  		// separately and defer to stackfmt for the rest.
   410  		stackfmt.FormatFrame(frame)
   411  		if more {
   412  			stackfmt.FormatStack(stack)
   413  		}
   414  		ce.Stack = buffer.String()
   415  	}
   416  
   417  	return ce
   418  }
   419  
   420  func terminalHookOverride(defaultHook, override zapcore.CheckWriteHook) zapcore.CheckWriteHook {
   421  	// A nil or WriteThenNoop hook will lead to continued execution after
   422  	// a Panic or Fatal log entry, which is unexpected. For example,
   423  	//
   424  	//   f, err := os.Open(..)
   425  	//   if err != nil {
   426  	//     log.Fatal("cannot open", zap.Error(err))
   427  	//   }
   428  	//   fmt.Println(f.Name())
   429  	//
   430  	// The f.Name() will panic if we continue execution after the log.Fatal.
   431  	if override == nil || override == zapcore.WriteThenNoop {
   432  		return defaultHook
   433  	}
   434  	return override
   435  }
   436  

View as plain text