Stack

Package log15 provides an opinionated, simple toolkit for best-practice logging that is both human and machine readable. It is modeled after the standard library's io and net/http packages. This package enforces you to only log key/value pairs. Keys must be strings. Values may be any type that you like. The default output format is logfmt, but you may also choose to use JSON instead if that suits you. Here's how you log: This will output a line that looks like: To get started, you'll want to import the library: Now you're ready to start logging: Because recording a human-meaningful message is common and good practice, the first argument to every logging method is the value to the *implicit* key 'msg'. Additionally, the level you choose for a message will be automatically added with the key 'lvl', and so will the current timestamp with key 't'. You may supply any additional context as a set of key/value pairs to the logging function. log15 allows you to favor terseness, ordering, and speed over safety. This is a reasonable tradeoff for logging functions. You don't need to explicitly state keys/values, log15 understands that they alternate in the variadic argument list: If you really do favor your type-safety, you may choose to pass a log.Ctx instead: Frequently, you want to add context to a logger so that you can track actions associated with it. An http request is a good example. You can easily create new loggers that have context that is automatically included with each log line: This will output a log line that includes the path context that is attached to the logger: The Handler interface defines where log lines are printed to and how they are formated. Handler is a single interface that is inspired by net/http's handler interface: Handlers can filter records, format them, or dispatch to multiple other Handlers. This package implements a number of Handlers for common logging patterns that are easily composed to create flexible, custom logging structures. Here's an example handler that prints logfmt output to Stdout: Here's an example handler that defers to two other handlers. One handler only prints records from the rpc package in logfmt to standard out. The other prints records at Error level or above in JSON formatted output to the file /var/log/service.json This package implements three Handlers that add debugging information to the context, CallerFileHandler, CallerFuncHandler and CallerStackHandler. Here's an example that adds the source file and line number of each logging call to the context. This will output a line that looks like: Here's an example that logs the call stack rather than just the call site. This will output a line that looks like: The "%+v" format instructs the handler to include the path of the source file relative to the compile time GOPATH. The github.com/go-stack/stack package documents the full list of formatting verbs and modifiers available. The Handler interface is so simple that it's also trivial to write your own. Let's create an example handler which tries to write to one handler, but if that fails it falls back to writing to another handler and includes the error that it encountered when trying to write to the primary. This might be useful when trying to log over a network socket, but if that fails you want to log those records to a file on disk. This pattern is so useful that a generic version that handles an arbitrary number of Handlers is included as part of this library called FailoverHandler. Sometimes, you want to log values that are extremely expensive to compute, but you don't want to pay the price of computing them if you haven't turned up your logging level to a high level of detail. This package provides a simple type to annotate a logging operation that you want to be evaluated lazily, just when it is about to be logged, so that it would not be evaluated if an upstream Handler filters it out. Just wrap any function which takes no arguments with the log.Lazy type. For example: If this message is not logged for any reason (like logging at the Error level), then factorRSAKey is never evaluated. The same log.Lazy mechanism can be used to attach context to a logger which you want to be evaluated when the message is logged, but not when the logger is created. For example, let's imagine a game where you have Player objects: You always want to log a player's name and whether they're alive or dead, so when you create the player object, you might do: Only now, even after a player has died, the logger will still report they are alive because the logging context is evaluated when the logger was created. By using the Lazy wrapper, we can defer the evaluation of whether the player is alive or not to each log message, so that the log records will reflect the player's current state no matter when the log message is written: If log15 detects that stdout is a terminal, it will configure the default handler for it (which is log.StdoutHandler) to use TerminalFormat. This format logs records nicely for your terminal, including color-coded output based on log level. Becasuse log15 allows you to step around the type system, there are a few ways you can specify invalid arguments to the logging functions. You could, for example, wrap something that is not a zero-argument function with log.Lazy or pass a context key that is not a string. Since logging libraries are typically the mechanism by which errors are reported, it would be onerous for the logging functions to return errors. Instead, log15 handles errors by making these guarantees to you: - Any log record containing an error will still be printed with the error explained to you as part of the log record. - Any log record containing an error will include the context key LOG15_ERROR, enabling you to easily (and if you like, automatically) detect if any of your logging calls are passing bad values. Understanding this, you might wonder why the Handler interface can return an error value in its Log method. Handlers are encouraged to return errors only if they fail to write their log records out to an external source like if the syslog daemon is not responding. This allows the construction of useful handlers which cope with those failures like the FailoverHandler. log15 is intended to be useful for library authors as a way to provide configurable logging to users of their library. Best practice for use in a library is to always disable all output for your logger by default and to provide a public Logger instance that consumers of your library can configure. Like so: Users of your library may then enable it if they like: The ability to attach context to a logger is a powerful one. Where should you do it and why? I favor embedding a Logger directly into any persistent object in my application and adding unique, tracing context keys to it. For instance, imagine I am writing a web browser: When a new tab is created, I assign a logger to it with the url of the tab as context so it can easily be traced through the logs. Now, whenever we perform any operation with the tab, we'll log with its embedded logger and it will include the tab title automatically: There's only one problem. What if the tab url changes? We could use log.Lazy to make sure the current url is always written, but that would mean that we couldn't trace a tab's full lifetime through our logs after the user navigate to a new URL. Instead, think about what values to attach to your loggers the same way you think about what to use as a key in a SQL database schema. If it's possible to use a natural key that is unique for the lifetime of the object, do so. But otherwise, log15's ext package has a handy RandId function to let you generate what you might call "surrogate keys" They're just random hex identifiers to use for tracing. Back to our Tab example, we would prefer to set up our Logger like so: Now we'll have a unique traceable identifier even across loading new urls, but we'll still be able to see the tab's current url in the log messages. For all Handler functions which can return an error, there is a version of that function which will return no error but panics on failure. They are all available on the Must object. For example: All of the following excellent projects inspired the design of this library: code.google.com/p/log4go github.com/op/go-logging github.com/technoweenie/grohl github.com/Sirupsen/logrus github.com/kr/logfmt github.com/spacemonkeygo/spacelog golang's stdlib, notably io and net/http https://xkcd.com/927/

Package logr defines a general-purpose logging API and abstract interfaces to back that API. Packages in the Go ecosystem can depend on this package, while callers can implement logging with whatever backend is appropriate. Logging is done using a Logger instance. Logger is a concrete type with methods, which defers the actual logging to a LogSink interface. The main methods of Logger are Info() and Error(). Arguments to Info() and Error() are key/value pairs rather than printf-style formatted strings, emphasizing "structured logging". With Go's standard log package, we might write: With logr's structured logging, we'd write: Errors are much the same. Instead of: We'd write: Info() and Error() are very similar, but they are separate methods so that LogSink implementations can choose to do things like attach additional information (such as stack traces) on calls to Error(). Error() messages are always logged, regardless of the current verbosity. If there is no error instance available, passing nil is valid. Often we want to log information only when the application in "verbose mode". To write log lines that are more verbose, Logger has a V() method. The higher the V-level of a log line, the less critical it is considered. Log-lines with V-levels that are not enabled (as per the LogSink) will not be written. Level V(0) is the default, and logger.V(0).Info() has the same meaning as logger.Info(). Negative V-levels have the same meaning as V(0). Error messages do not have a verbosity level and are always logged. Where we might have written: We can write: Logger instances can have name strings so that all messages logged through that instance have additional context. For example, you might want to add a subsystem name: The WithName() method returns a new Logger, which can be passed to constructors or other functions for further use. Repeated use of WithName() will accumulate name "segments". These name segments will be joined in some way by the LogSink implementation. It is strongly recommended that name segments contain simple identifiers (letters, digits, and hyphen), and do not contain characters that could muddle the log output or confuse the joining operation (e.g. whitespace, commas, periods, slashes, brackets, quotes, etc). Logger instances can store any number of key/value pairs, which will be logged alongside all messages logged through that instance. For example, you might want to create a Logger instance per managed object: With the standard log package, we might write: With logr we'd write: Logger has very few hard rules, with the goal that LogSink implementations might have a lot of freedom to differentiate. There are, however, some things to consider. The log message consists of a constant message attached to the log line. This should generally be a simple description of what's occurring, and should never be a format string. Variable information can then be attached using named values. Keys are arbitrary strings, but should generally be constant values. Values may be any Go value, but how the value is formatted is determined by the LogSink implementation. Logger instances are meant to be passed around by value. Code that receives such a value can call its methods without having to check whether the instance is ready for use. The zero logger (= Logger{}) is identical to Discard() and discards all log entries. Code that receives a Logger by value can simply call it, the methods will never crash. For cases where passing a logger is optional, a pointer to Logger should be used. Keys are not strictly required to conform to any specification or regex, but it is recommended that they: These guidelines help ensure that log data is processed properly regardless of the log implementation. For example, log implementations will try to output JSON data or will store data for later database (e.g. SQL) queries. While users are generally free to use key names of their choice, it's generally best to avoid using the following keys, as they're frequently used by implementations: Implementations are encouraged to make use of these keys to represent the above concepts, when necessary (for example, in a pure-JSON output form, it would be necessary to represent at least message and timestamp as ordinary named values). Implementations may choose to give callers access to the underlying logging implementation. The recommended pattern for this is: Logger grants access to the sink to enable type assertions like this: Custom `With*` functions can be implemented by copying the complete Logger struct and replacing the sink in the copy: Don't use New to construct a new Logger with a LogSink retrieved from an existing Logger. Source code attribution might not work correctly and unexported fields in Logger get lost. Beware that the same LogSink instance may be shared by different logger instances. Calling functions that modify the LogSink will affect all of those.

Package errors provides errors that have stack-traces. This is particularly useful when you want to understand the state of execution when an error was returned unexpectedly. It provides the type *Error which implements the standard golang error interface, so you can use this library interchangably with code that is expecting a normal error return. For example: This can be called as follows: This package was original written to allow reporting to Bugsnag, but after I found similar packages by Facebook and Dropbox, it was moved to one canonical location so everyone can benefit.

Package stack is a pluggable framework for microservices