Zap Concepts: Difference between revisions
Line 121: | Line 121: | ||
Encoding: "console", | Encoding: "console", | ||
EncoderConfig: encoderConfig, | EncoderConfig: encoderConfig, | ||
OutputPaths: []string{ | OutputPaths: []string{"stdout"}, | ||
ErrorOutputPaths: []string{ | ErrorOutputPaths: []string{"stdout"}, | ||
InitialFields: map[string]interface{}{ | InitialFields: map[string]interface{}{ | ||
"pid": os.Getpid(), | "pid": os.Getpid(), |
Latest revision as of 00:48, 15 March 2024
External
Internal
Overview
Zap is a logging framework for Go. It provides fast, structured, contextual and leveled logging.
The programming model involves creating a Logger
object and invoking its API, for example logger.Info(...)
. Creation of Logger
instances can be done in a number of way that are explored in the Logger Instance Creation section. There are actually two logging APIs, zap.Logger
and zap.SugaredLogger
. The former is appropriate for high-performance scenarios, but only supports a structured logging method signatures, while the latter has a more friendlier, and just slightly less performant syntax which is similar to that of fmt.Sprintf()
. The difference is discussed in Two Logging APIs section.
Each logging invocation creates a log event with key/value pairs, as explained in Structured Logging. The rendering of the log events can be changed through configuration. These aspects are discussed in the Log Rendering section.
There is no global logger that can be used right away, though one can be configured as shown in Global Logger Instance. Use of global loggers should be avoided, though.
Zap supports the standard DEBUG, INFO, WARN and ERROR logging levels. It comes with a few new ones: PANIC, DPANIC and FATAL. The ERROR level requires special attention. There is no TRACE. More details are available in the Logging Levels section.
By default, loggers are unbuffered. This aspect is discussed in the Buffering section.
Implementation aspects are discussed in Implementation Details.
Best Practices
Encapsulate the logic that configures and creates loggers into a logging
package. The package should create the loggers, and should manage the number of loggers used by the application.
If the application needs a logger, it should request it via logging.Get(...) *zap.Logger
kind of exported method. The Get()
methods could use the singleton initialization pattern based on sync.Once
synchronization primitive.
Zap documentation seems to recommend passing logger instances as function parameters. Why is this a good idea?
Can I use a single instance? Is it thread safe? Sharing or not sharing logger instances?
Understand how I can get a logger instance “out of the blue” and how I can build a set of predefined loggers. Must figure out logging situation in tests. I must be able to start a test in debug mode and see the log.
Use a typical usage section?
import "go.uber.org/zap"
...
var config zap.Config
var logger *zap.Logger
var sugaredLogger *zap.SugaredLogger
outputPath := "stderr" // ... or a local path
config = zap.NewDevelopmentConfig()
config.OutputPaths = []string{outputPath}
config.ErrorOutputPaths = []string{outputPath}
logger, err := config.Build()
if err != nil {
panic(err)
}
sugaredLogger = logger.Sugar()
sugaredLogger.Infof("somehting")
...
logger = sugaredLogger.Desugar()
Logger Instance Creation and Configuration
Pre-configured loggers are available. Custom-configured instances can be created via two other methods: building a configuration first and using the configuration to instantiate a logger, or programmatically creating and configuring a zapcore.Core
.
Pre-configured Loggers
The framework comes with three preset constructors: zap.NewExample()
, zap.NewProduction(...)
and zap.NewDevelopment(...)
. Each of these constructors internally create a logging core and set its encoder, output and level. These components are configured according to the type of logger. In case of the example logger, the encoder produces JSON structures that contain only two keys "level" and "msg", the default log level is DEBUG and the logs are send to stdout. In case of a production logger, the encoder produces JSON structures that contain "level", "ts", "msg" and "caller", the default log level is INFO, the "normal" logging is sent to stdout and error logging is sent to stderr and sampling is enabled. In case of a development logger, the encoder produces "console" encoding, which are human readable messages, the default log level is DEBUG, the "normal" logging is sent to stdout and error logging is sent to stderr and sampling is disabled.
logger := zap.NewExample()
logger, err := zap.NewProduction(...)
// To turn the error into a panic if it occurs, and simplify the code:
logger := zap.Must(zap.NewProduction(...))
logger, err := zap.NewDevelopment(...)
These instances are OK if their configuration matches the use case.
Configuration-Driven Logger Creation
An alternative to the prepackaged configuration constructors zap.NewDevelopmentConfig()
and zap.NewProductionConfig()
is creating the logger by instantiating a zap.Config
struct, and then invoking Build()
on it to create the logger.
Note that zap.Config
intentionally supports only the most common options. More unusual logging setups, such as logging to network connections or message queues, splitting output between multiple files, etc. are possible, but require direct use of the zapcore
package, as shown in the Low-Level Logger Creation section.
An example of creating a logger by creating configuration instance first is:
config := zap.NewDevelopmentConfig()
// update configuration as you like ...
logger, err := config.Build()
zap.Config
contains many of the most common configuration options desired when creating a new Logger
.
The configuration can also be created from scratch:
// The 'loggingLevel' instance can be used to dynamically update the level of the active logger.
loggingLevel, err = zap.ParseAtomicLevel("info")
if err != nil {
...
}
encoderConfig := zapcore.EncoderConfig{
TimeKey: "timestamp",
LevelKey: "level",
NameKey: "logger",
CallerKey: "caller",
FunctionKey: zapcore.OmitKey,
MessageKey: "msg",
StacktraceKey: "stacktrace",
LineEnding: zapcore.DefaultLineEnding,
EncodeLevel: zapcore.LowercaseLevelEncoder,
EncodeTime: zapcore.ISO8601TimeEncoder,
EncodeDuration: zapcore.SecondsDurationEncoder,
EncodeCaller: zapcore.ShortCallerEncoder,
}
loggerConfig := zap.Config{
Level: loggingLevel,
Development: false,
DisableCaller: false,
DisableStacktrace: true,
Sampling: nil,
Encoding: "console",
EncoderConfig: encoderConfig,
OutputPaths: []string{"stdout"},
ErrorOutputPaths: []string{"stdout"},
InitialFields: map[string]interface{}{
"pid": os.Getpid(),
},
}
logger, err := loggerConfig.Build()
if err != nil {
...
}
Configuration Options
- level The minimum enabled logging level. The level can be dynamically changed at runtime with
Config.Level.SetLevel()
for all logger descended from theConfig
instance the function was called on. - development This field puts the logger in development mode, which changes the behavior of
DPanic
logging level. - disableCaller Setting this field to
true
stops annotating logs with the calling function's file name and line number. By default, all logs are annotated. - disableStacktrace Setting this field to
true
completely disables automatic stack trace capturing. By stack traces are captured forWarn
level and above logs in development andError
level and above in production. - sampling Set a sampling| policy. A
nil
SamplingConfig
disables sampling. - encoding Sets the logger's encoding. Valid values are "json", "console" and any third-party encoders registered via
RegisterEncoder
. Also see Log Rendering below. - encoderConfig Sets options for the chosen encoder.
- outputPaths A list of URLs or file paths to write logging output to.
- errorOutputPaths A list of URLs to write internal logger errors to. The default is standard error. Note that this setting only affects internal errors. For sample code that sends error-level logs to a different location from info- and debug-level logs, see the package-level AdvancedConfiguration example.
- initialFields Specifies global contextual fields that should be included in every long entry produced by each logger that was built from this
Config
object.
Low-Level Logger-Creation
An example of creating a logger by programmatically creating and configuring a zapcore.Core
instance is:
encoderCfg := zapcore.EncoderConfig{
MessageKey: "msg",
LevelKey: "level",
NameKey: "logger",
EncodeLevel: zapcore.LowercaseLevelEncoder,
EncodeTime: zapcore.ISO8601TimeEncoder,
EncodeDuration: zapcore.StringDurationEncoder,
}
core := zapcore.NewCore(zapcore.NewJSONEncoder(encoderCfg), os.Stdout, DebugLevel)
logger := zap.New(core)
Cores may form a tree hierarchy:
core := zapcore.NewTee(zapcore.NewCore(...), zapcore.NewCore(...))
Global Logger Instance
Unlike most other logging packages for Go, Zap does not provide a pre-configured global logger for use. It does provide the API for it, so if you prefer to use a global logger, you could use zap.ReplaceGlobals()
, perhaps in your package's init()
function, as shown below. However, use of global loggers should be avoided.
zap.ReplaceGlobal(zap.Must(zap.NewProduction()) // or use any other custom configuration you desire
...
zap.L().Info("test")
Two Logging APIs
Zap provides to APIs for logging: zap.Logger
and zap.SugaredLogger
.
Choosing between Logger
and SugaredLogger
does not need to be an application-wide decision, the instances implementing these APIs can be converted into each other with negligible performance penalty as shown below. A typical pattern is to use the SugaredLogger
API throughout the code, for convenience, then convert it to a Logger
at the boundaries of performance-sensitive code.
logger := ...
sugaredLogger := logger.Sugar()
plainLogger := sugar.Desugar()
zap.Logger
zap.Logger
is a low-level API aimed for performance-sensitive scenarios. It only supports a structured logging method signatures, with strongly typed fields. It strives to avoid serialization overhead and allocations whenever possible. The logger has a single method fore each supported log levels (Debug()
, Info()
, Warn()
, etc.), and each method accept a message parameter, which will surface as the "msg" key/value pair in the structured log entry, and zero or more zap.Field
s, which are strongly typed key/value pairs and will surface as key/value pairs in the structured log entry.
log.Info("this is the message", zap.String("someString", "blue"), zap.Int("someInt", 15))
will generate:
{"level":"info","msg":"this is the message","someString":"blue","someInt":15}
zap.SugaredLogger
zap.SugaredLogger has a friendlier, loosely typed, more convenient API that just slightly less performant. The syntax which is similar to that of fmt.Sprintf()
. Internally, zap.SugaredLogger delegates to a zap.Logger instance. The logger exposes a series of methods for each logging level:
Info()
: all arguments are loosely typed. The method usesfmt.Sprint()
to concatenate the arguments into themsg
log entry field.Infoln()
: Same as above, except that it adds a new line to the message withfmt.Sprintln()
.Infof()
: the first argument is a format string, followed by arguments for conversion characters, in thefmt.Sprintf()
style. The format string and the arguments are rendered into the content of themsg
log entry field.Infow()
: the first argument is the message, followed by a variable number of key/value pairs, aszap.String()
, etc. The method allows you to add a mix of strongly and loosely typed key/value pairs. The log message is the first argument. Subsequent arguments are expected to be key/value pairs. Note that for the loosely typed key/value pairs, the keys are always expected to be strings, while the values can be of any type. If you use a non-string key, the program will panic in development and a separate error will logged, and the key/value pair will be ignored in production. Passing an orphaned key behaves similarly. Due to all these caveats, we recommend using strongly-typed contextual fields at all times.
Structured Logging
Structured logging is a technique that generates log events rendered in a well-defined format, typically JSON or key-value pairs. This approach allows efficient searching and filtering, making the log entires machine-readable.
Each logging invocation creates a log event. Zap calls these events log entries. The log entries are implemented as zapcore.Entry
struct instances, so each carry by default the structure's fields, also known as logging fields: logging level, timestamp, logger name, message, caller and stack. The caller is a string containing the source code location where the log invocation was done, in form of <package>/<file-name>.go:<line>
.
Zap API consistently allows specifying a message in addition to key-value pairs. Subjectively, we find it helpful to accompany structured context with a brief description. This is not critical during development, but it makes debugging and operating unfamiliar system much easier.
Also see:
Contextual Logging
Contextual logging with Zap is done by passing strongly typed key/value pairs alongside the log message.
Child loggers can be used to add contextual properties to all the logs produced in a specific scope. This helps avoiding unnecessary repetition at log point. A child logger is created calling With()
on a parent logger:
log := ...
childLog := log.With(zap.String("color", "red"))
childLog.Info(...)
A similar mechanism can be applied to zapcore.Core
.
Also see:
Log Rendering
The external representation of a log entry is produced by an encoder. The default encoder generates JSON structures, where the log entry's fields are represented as JSON map elements.
zapcore.NewConsoleEncoder()
zapcore.NewJSONEncoder()
The encoder constructors take EncoderConfig
instances. The encoder for a logger is configured with the encoding
field of the configuration.
Logging Levels
Internally, the logging level are implemented as constants (zap.DebugLevel
, zap.InfoLevel
, zap.WarnLevel
, zap.ErrorLevel
, zap.DPanicLevel
, zap.PanicLevel
, zap.FatalLevel
).
DebugLevel
Debugging messages. Debug logs are typically voluminous, and are usually disabled in production.
InfoLevel
Messages describing normal application operations. This is the default logging priority.
WarnLevel
Messages indicating that something unusual happened that may need attention before it escalates to a more severe issue. Warning log entries are more important than info log entries, but don't need individual human review..
ErrorLevel
An unexpected error condition in the program. Error log entries are high-priority. If an application is running smoothly, it shouldn't generate any error-level logs.
zap.Error*()
methods issue log events configured by default with ErrorLevel
. A stacktrace
field is automatically added to the log entry by the Error*()
methods.
DPanicLevel
A severe error condition. It behaves like PanicLevel
in development and ErrorLevel
in production. The behavior of the logger can be changed by configuring the development
field of the configuration the logger was created from.
PanicLevel
Calls panic()
after logging an error condition.
FatalLevel
The zap.Fatal*()
methods issue FatalLevel
log entries. After logging such an event, the method calls os.Exit(1)
.
Buffering
By default, loggers are unbuffered. However, since Zap's low-level API allows buffering, it's a good practice to call Sync()
before the process exits:
logger := zap.Must(zap.NewDevelopment(...))
defer logger.Sync()
Loggers and context.Context
Also see:
Implementation Details
Each zap.Logger
wraps around a core, which is configured with an encoder, an output and a logging level:
An encoder is the component that renders the external representation of the log entry instances. The default encoder is a JSON encoder, which renders entries as JSON structures. The encoder instance can be configured with a zapcore.EncoderConfig
structure.
Advanced Use
Disable Stack Trace Rendering
Stack trace on Warning is probably not necessary. However, this disable stack trace rendering altogether:
var config zap.Config
config = zap.NewDevelopmentConfig()
config.DisableStacktrace = true
Change Log Level Dynamically
Create the logger instance with an external zap.AtomicLevel
instance. Updating the level on that instance updates it dynamically for the logger:
atomicLevel := zap.NewAtomicLevelAt(zap.DebugLevel)
var config zap.Config
config = zap.NewDevelopmentConfig()
config.Level = atomicLevel // use the instance, updating the level on that instance will dynamically update the logging level
logger, err := config.Build()
...
// this changes the logging level dynamically
atomicLevel.SetLevel(zap.InfoLevel)
Multi-Output Logging
Custom Logging Levels
Log Rotation
The Lumberjack package https://pkg.go.dev/gopkg.in/natefinch/lumberjack.v2 may be used to rotate log files. However, the recommended practice is to rely on external tools like logrotate
instead of doing it in the application itself.
Log Sampling
Log sampling is a technique used to reduce application log volume by selectively capturing and recording only a subset of log events. Its purpose is to strike a balance between the need for comprehensive logging and the potential performance impact of logging too much data. Rather than capturing every single log event, log sampling selects a representative subset based on specific criterial. This way, the amount of generated log data is greatly reduced, which can be beneficial in high-throughput systems. Zap sampling algorithm uses the log message to identify duplicate entries. This is a practical middle ground between random sampling, which may drop the exact entry that you need while debugging, and hashing the complete entry, which is prohibitively expensive. The sampling policy is configured with the sampling
field of the configuration.