Zap Concepts

From NovaOrdis Knowledge Base
Jump to navigation Jump to search

External

Internal

TODO

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.

Logger Instance Creation

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.

However, 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.

Configuration-Driven Logger Creation. A logger can be created by first creating a configuration with one of the configuration constructors (zap.NewDevelopmentConfig(), zap.NewProductionConfig()), or directly by instantiating a zap.Config struct, and then invoking Build() on it to create the logger: An example of creating a logger by creating configuration instance first is:

config := zap.NewDevelopmentConfig()
// update configuration as you like ...
logger, err := config.Build()

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 := New(core)

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. 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.Fields, 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 uses fmt.Sprint() to concatenate the arguments into the msg log entry field.
  • Infoln(): Same as above, except that it adds a new line to the message with fmt.Sprintln().
  • Infof(): the first argument is a format string, followed by arguments for conversion characters, in the fmt.Sprintf() style. The format string and the arguments are rendered into the content of the msg log entry field.
  • Infow(): the first argument is the message, followed by a variable number of key/value pairs, as zap.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

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>.

Also see:

Structured Logging

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.

Logging Levels

Internally, the logging level are implemented as constants (zap.DebugLevel, zap.InfoLevel, zap.WarnLevel, zap.ErrorLevel, zap.DPanicLevel, zap.PanicLevel, zap.FatalLevel).

Custom Logging Levels

Configuration

Buffering

Explain.

logger := zap.Must(zap.NewDevelopment(...))
defer logger.Sync()

Sampling

Other Subjects

Child loggers.

Implementation Details

Each zap.Logger wraps around a core, which is configured with an encoder, an output and a logging level:

Zap.png

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.

zap and zap core.

Advanced Use

Custom Encoders

Multi-Output Logging

Best Practices

Sharing or not sharing logger instances?

Is it thread-safe?

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.

Can I use a single instance? Is it thread safe?