@goodware/log: Winston3-based logging to console, file, and/or AWS CloudWatch Logs

Links

• Release History
• npm
• git
• API
• RunKit Notebook

Requirements

As of version 9, NodeJS version 14+ is required. Versions 8 and earlier work with NodeJS 8 and later. When the CloudWatch Logs transport is used, NodeJS support depends upon that of @aws-sdk.

ulidx complains that Node 16+ is required but 14 works fine.

Installation

npm i --save @goodware/log

Introduction

This package extends Winston3 with additional features such as tag-based filtering. It is intended for NodeJS runtimes including AWS Lambda. The package can log scalars, objects (even with cyclical references), arrays, and graphs of Error objects quickly and reliably. Log entries can be sent to the following transports: console (via winston-console-format), files (via winston-daily-rotate-file) in JSON format, and CloudWatch Logs (via winston-cloudwatch) in JSON format.

@goodware/log features a large unit test suite and has been servicing a commercial SaaS product for several years. It is BSD licensed.

Features

1. In addition to a message (an object or scalar), a log entry may consist of tags (an array of strings), context (an object or scalar), and data (an object or scalar). message, and data may also be arrays.
2. Log entries can be filtered by tags on a per-transport basis
3. Redaction of specific object keys which can be enabled and disabled via tags
4. Safely logs large objects and arrays - even those with circular references
5. Embedded Error objects passed via 'message' and 'data' are logged separately (e.g., in the 'cause' and 'error' properties), grouping multiple log entries via a unique identifier
6. Promotes object properties to a configurable subset of 'meta' properties
7. On-demand flushing to CloudWatch Logs
8. Does not interfere with other code that uses Winston
9. Transformation/redaction of logged objects via events

Performance

Call isLevelEnabled() (and even memoize it) to avoid creating expensive messages that won't be logged, especially in loops.

Transports Supported

The following transports can be utilized. They are all optional.

• The console (stdout) via winston-console-format
• Local JSON files via winston-daily-rotate-file
• AWS CloudWatch Logs via winston-cloudwatch in JSON format

What's Missing

The ability to add additional transports

Usage

Loggers

The Loggers class is a container that manages logger instances that have unique category names. Each logger has its own settings, such as logging levels and transports.

Any number of Loggers instances can exist at any given time. This is useful if, say, independent libraries use this package with different logging levels and other settings. The only caveat is Winston's design flaw that prevents assigning different colors to the same level.

Logging

Log messages via log(), default(), and methods that are named after logging levels (aka level()), such as info(). The list of available logging levels and the console color for each can be provided via options.

The concept of tags was borrowed from the HAPI project. Tags are a superset of logging levels. Log entries have only one level; however, tags are logged as an array in order to facilitate searches. For example, when the tags are (info, web) the log entry is logged at the 'info' level. When tags contain multiple level names, predecence rules apply (see logLevel below; otherwise, the tag in the smallest array index wins.

Log entries are created from five components (all optional): tags, message, data, context, and category. This information can be passed as traditional ordered parameters or by passing a single object for named parameters. When named parameters are used, extra provided properties are logged as data; for example, the following object can be logged: { tags: 'disk', message: 'A message', error: new Error('An error') }.

Winston's splat formatter is not enabled. However, any type of data can be logged, such as strings, objects, arrays, and Errors (including their stack traces and related errors).

log()'s tag parameter can be a string, array, or an object whose properties are tag names and their values are evaluated for truthiness that indicates whether the tag is enabled. When named parameters are used, the 'tags' argument can be an object, array, or string.

level() methods optionally accept an array of tags as the first parameter. If an object is provided as the first parameter, it is treated either as name parameters or as a message to log.

The values for 'message,' 'data,' and 'context' can be of any type except 'function.'

Error objects are treated specially: their stacks and dependency graphs are also logged. In most cases, 'message' is a string that is used as the log entry's message whereas context and data are objects that appear in the corresponding properties in log entries.

log(), default(), and level() optionally accept an Error object as the first parameter, followed by message, data, context, and category. 'error' is automatically added to the tags; however, 'level' for level() methods takes precedence. For example, info(new Error('An error')) is logged at the info level.

Loggers and Loggers Methods

• tags() merges multiple objects into a single object
• category() returns the default category (specified via options) if the provided value is blank
• logger(category) returns a logger associated with a category. Log entries have a category property.
• context() Returns the context associated with a child loggers combined with the context of its parent

Winston Loggers

One Winston logger is created for each unique category name. winstonLogger() returns a Winston logger given a category name.

Child Loggers (flyweight)

logger() and child() return objects with their own tags, context, data, and category values. These objects have the same interface as the Loggers class. Child loggers are not real Winston Logger instances and are therefore lightweight.

Child loggers and Loggers instances have the following methods:

Context can be built up by chaining calls to logger() and/or child().

loggers.logger('dog').child('dogTag').child(null, { userId: 101 }).logger('anotherCategory').debug('Wow!');

Caching Loggers

A logger can be associated with a category via setLogger(name, logger). This changes the value that is returned by logger() when the same category is provided. For example:

loggers.setLogger('dog', loggers.child('dogTag'));

Redaction

The 'redact' option setting allows the removal of properties from data objects (top-level only or recursive). Input data is never changed - properties are removed from copies.

The following would remove the password key from all input data.

redact: {
  password: undefined,
}

Events

The Loggers class inherits from EventEmitter.

redact Event

redact events are emitted (potentially multiple times) when one log-related method is called. This event is intended for redacting properties from message, data, and context objects. These events are emitted once for each object to be logged, including nested Error objects.

Event listeners' return values are ignored. Event listeners are sent an event object consisting of category, level, data, and tags properties. Listeners can modify the event object's data property. Alternatively, the data object can be modified, but mutate input data at your own risk!

Event objects have the following properties when their 'type' property value is 'message,' or 'data:'

• arg: Contains the data to be altered (context, message, or data). It could be an array, string, or object.
• property: Contains the object property name where 'arg' is located
• tags
• category
• context
• level

Event objects have the following properties when their 'type' property value is 'context:'

• arg: Contains the data to be altered (context). It could be an array, string, or object.
• property: Contains the object property name where 'arg' is located
• tags
• category
• level: This is undefined except when the logging level is known, which is only when a context value is provided to log() or to a log level method.

The 'arg' property can be modified to avoid mutating input data. For example, the following removes the response attribute from Error objects:

loggers.on('redact', (item) => {
  const { arg } = item;
  // only process Errors with response properties
  if (!(arg instanceof Error) || !arg.response) return;
  // modify item's arg property instead of modifying the error object
  // Make a shallow copy and remove the response property
  const copy = { ...arg };
  delete copy.response;
  item.arg = copy;
});

const error = new Error('An http error occurred');
error.response = 'I will be redacted';
error.statusCode = 404;
loggers.log(error);

log Event

'log' events are emitted prior to calling Winston's log() method. Due to error object traversal, multiple log events can be emitted when one log-related method is called. Because of transport tag filtering, the entry might not be logged.

Event objects have the following properties:

• tags
• level
• message
• data
• context
• category

Stopping and Flushing

stop() is a heavyweight asynchronous method that safely closes all transports. It can be reversed via start(). restart() is also available.

Only CloudWatch Logs transports can be flushed without stopping them. flush() is therefore an alias for restart(). Use flushCloudWatchTransports() when you only need to flush CloudWatch Logs transports because it's substantially faster.

Unhandled exceptions and Promise rejections

While a Loggers instance is active, uncaught exceptions and unhandled Promise rejections are logged using the category @log/unhandled. The process is not terminated after logging uncaught exceptions.

Adding Stack Traces of API Callers

When a log entry's level is one of the values specified in the 'logStackLevels' options setting, the 'stack' meta property is set to the stack trace of the caller of the logging method (such as info()). This behavior is manually enabled and disabled via the 'logStack' meta tag.

Concepts

options

An object provided to Loggers' constructor. Options are described by optionsObject.

logger

A logger sends log entries to transports. Loggers instances are loggers. child() and logger() return loggers.

A logger implements the following interface:

• start()
• stop()
• restart()
• ready (property)
• log()
• default() The default level is specified via option
• level() where level is a logging level name. Example: info()
• child()
• stack()
• logger(category [, loggerObject])
• parent() For child loggers to access their parent, which is either a Loggers object or a child logger
• loggers() For child loggers to access the Loggers object that created them
• winstonLogger()
• tags(), for combining tags
• data(), for combining data objects
• category() (mostly useful for retrieving the category assigned to Logger objects)
• flush()
• flushCloudWatchTransports()

category

Per Winston terminology, a category is the name of a logger. When a category is not specified, the category specified via the 'defaultCategory' options setting is used. 'defaultCategory' defaults to 'general.' Transport filtering (based on tags) is specified on a per-category basis. Loggers inherit the settings of the 'default' category (not the 'defaultCategory' options setting).

log entry

An object that is sent to a transport. A log entry consists of meta and data.

meta

The top-level keys of a log entry. Meta properties contain scalar values except for tags and transports which are arrays. Meta properties are: timestamp, ms (elpased time between log entries), level, message, tags, category, id, groupId, depth, hostId, stage, version, service, stack, and transports. Certain properties in 'both' can be copied to meta properties, optionally renaming them, via the 'metaProperties' options setting.

data

The properties remaining in 'both' after meta properties are removed (see 'both' below)

context

An object provided optionally to log() methods and to the Logger constructor. Context objects are merged. For example, if a child logger (Logger) is created with: {a: 1, b: 5} and {b: 2, c: 3} is passed as a context argument to log(), {a: 1, b: 2, c: 3} is logged.

level (severity)

Levels are associated with natural numbers. Per Winston's convention, a lower value indicates greater severity. Therefore, 0 means "most severe."

Levels and their colors can be specified via the second argument provided to the constructor using the same object that is provided when creating a Winston logger.

By default, Loggers uses Winston's default levels (aka npm log levels with the addition of three levels:

1. 'fail' is more severe than 'error' and has the color red
2. 'more' is between 'info' and 'verbose' and has the color cyan
3. 'db' is between 'verbose' and 'http' and has the color yellow

From highest to lowest severity, the levels are: fail, error, warn, notice, info, more, db, http, verbose, debug, silly.

A custom set of levels can be provided to the Loggers class's constructor; however, the Loggers class assumes there is an 'error' level and the options model (via the defaults) assumes the following levels exist: error, warn, debug.

default level

When a level is not found in the provided tags, the default level, 'debug', is assumed. The default level is specified via the 'defaultLevel' options setting.

level methods

Methods that are named after logging levels, such as error(). log(tags, message, data, category) is an alternative to the level methods. Level methods accept variant parameters. If the first parameter to a level method is an array, the parameter list is (tags, message, data, category). Otherwise, it's (message, data, category).

level filtering

A log entry is sent to a transport only when its severity level is equal to or greater than the transport's level.

tag

Tags are logged as an array of strings. Tags are specified via a single string, an array of strings, or an object in which each key value is evaluated for truthiness. Tags are combined via the static method tags(a, b) where b's tags override a's tags.

tag and level

Tags are a superset of levels. A log entry's level is, by default, set to the tag with the highest severity. Level methods override this behavior such that the level associated with the method is chosen. The level can be specified via the logLevel meta tag. The level can also be modified via tag filtering options (see "tag filtering" below).

meta tag

Some tags alter logging behavior. A tag's value (tags can be specified as an object) enables and disables the feature, based on their truthiness. Meta tags are not logged. Meta tag names start with 'log' and 'noLog.' Meta tags tags that start with noLog negate their corresponding meta tags. For example, {logStack: true} is identical to 'logStack'. {logStack: false} is identical to 'noLogStack.' Meta tag names are:

logLevel

Use the meta tag's value as a log entry's logging level

logStack

Whether to add the current stack to meta. When true, populates the 'stack' meta property. This is the default behavior when the log entry's level is 'error.'

tag-based filtering

Tags can be used to filter messages beyond logging levels. Tags that are named after logging levels do not participate in tag filtering. All tags are enabled by default. When using tag filtering, when the log entry's severity level is 'warn' higher, tags are enabled; however, this behavior can be overidden via the 'allowLevel' setting. Tags are enabled and disabled on a per-category basis. The 'default' category specifies the default behavior for unlisted categories.

The following example enables the tag 'sql' for only two categories: one and two. Category 'one' changes the level to 'more' and sends log entries only to the file and console transports. Category 'two' sends log entries to all transports.

categories: {
  default: {
    tags: {
      sql: 'off', // Disable for all transports for all categories. Log entries with warn and error levels are logged.
    },
  },
  one: {
    tags: {
      sql: {
        // Fine-tune filtering for category 'one.' All of these keys are optional.
        allowLevel: 'off', // Enable tag filtering for all log entries regardless of their levels. 'off' is needed
          // because the default is 'warn'
        level: 'more', // Set the log entry's level to 'more'
        // Log entries are sent to all transports by default (console, file, errorFile, cloudWatch). Each transport
        // can be overridden:
        file: 'verbose', // Send a log entry with the 'sql' tag to the file transport if the log entry's severity
          // level is equal to or greater than 'verbose'
        console: 'on', // Send a log entry with the 'sql' tag to the console transport
        other: 'off', // Do not send a log entry with the 'sql' tag to CloudWatch
      },
    },
  },
  two: {
    tags: {
      sql: 'on', // Send all log entries with the 'sql' tag to all transports for category two
    },
  },
}

host id

Uniquely identifies the system that is running node

stage

Identifies the environment in which node is running, such as 'dev' or 'prod'

data

An optional string, array, or object to log with message. Two 'data' objects are combined via the static method data(a, b) in which b's keys override a's keys if they overlap.

message

A scalar, array, or object to log. If an object is provided, its 'message' property is moved to meta and other properties can be copied to meta. The list of keys to copy to meta is altered via the 'metaProperties' options setting. Properties are copied to meta if their values are scalar and their names are specified in metaProperties.

both

message and 'data' are shallow copied and combined into a new object called 'both.' If message's keys overlap with those in 'data,' 'data' is logged separately; both log entries will have the same groupId meta value.

Errors

If both.error is truthy and both.message is falsey, both.message is set to both.error.asString().

Error objects that are discovered in the top-level keys of both are logged separately, in a parent-child fashion, and recursively. This allows the stack trace and other details of every Error in a chain to be logged using applicable redaction rules. Each log entry contains the same groupId meta value. The data properties of parent entries contain the result of converting Error strings. For example, if both.error is an Error object, data.error will contain the Error object converted to a string. This process is performed recursively. Circular references are handled gracefully. The depth meta property contains a number, starting from 1, that indicates the recursion depth from both. The maximum recursion depth is specified via the 'maxErrorDepth' options setting. The maximum number of errors to log is specified via the 'maxErrors' options setting.

The following example produces three three log entries. error3 will be logged first, followed by error2, followed by error1. error1's corresponding log entry contains a data.cause key with a string value of 'Error: error2.'

const error = new Error('error1');
const error2 = new Error('error2');
const error3 = new Error('error3');
error1.cause = error2;
error2.error = error3;
logger.log('error', error);

transport

A transport sends log entries to one of the following destinations:

• file

  Writes log entries with level equal to or higher than a specified level (defaults to 'info') to a file named category-timestamp.log.

• errorFile

  Writes log entries with level equal to or higher than a specified level (defaults to 'error') to a file named category-error-timestamp.log. Log entries are also sent to the file transport.

• cloudWatch

  Sends log entries to CloudWatch AWS service

• console

  Writes log entries to the process's STDOUT filehandle

transport level

Use the 'categories' options setting to configure transports. It is not necessary to specify every category that is actually used. The 'default' category specifies the base configuration for all categories. For example:

categories: {
  default: { console: 'on', cloudWatch: 'info',
             file: 'default' },
  api: { console: 'off' },
}

Level filtering for each transport is configured via a level name, 'default,' 'on,' or 'off.' 'off' is the default. Each transport type treats 'on' slightly differently:

• file: on -> info
• errorFile: on -> error
• cloudWatch: on -> warn
• console: on -> info, off -> warn if file and cloudWatch are both off

console

The behavior of console transports is altered via the 'console' options setting.

When 'colors' is true, log entries sent to the console are colorized. To override the provided value, set the

CONSOLE_COLORS

environment variable such that blank, 0, and 'false' are false and all other values are true.

When 'data' is true, the maximum amount of information is sent to the console, including meta, data, embedded errors, and stack traces. When it is false, a subset of meta properties are sent to the console with the log entry's message. To override the value for 'data', set the

CONSOLE_DATA

environment variable such that blank, 0, and 'false' are false and all other values are true.

Inside the categories: property, the console propery can contain a string or an object containing the data and colors keys that override the 'console' options settings.

file and errorFile

Log entries are written to files as JSON strings to a directory specified via the 'file' options setting. If no directory in the provided array does not exist and can be created, the file-related transports are disabled. File names contain the category and the date and hour of the local time when this object was instantiated. Category names may contain operating-system directory separators and must conform to the filesystem rules of the operating system. For error log files, '-error' is appended to the category. Files have the extension .log. An example file name is: general-error-2020-07-18-18.log. Files are rotated and zipped on an hourly basis. The maximum number of archived log files defaults to 14 days and can be specified via the 'file' options setting.

cloudWatch

CloudWatch transports are configured with a log group name and an optional AWS region. Log entries are sent to CloudWatch as JSON strings. One Loggers instance uses the same stream name for all cloudWatch transports. The log group, on the other hand, can be specified on a per-category basis. The log stream name contains the date and time (including the millisecond) when Loggers was instantiated followed by the host id. Any errors that occur while sending log entries to CloudWatch are written to the console and to files named cloudwatch-error*.log if a file directory is specified.

If an AWS region is not specified, the environment variables are used in the following order:

1. AWS_CLOUDWATCH_LOGS_REGION
2. AWS_REGION

Begin/End/Error Utility Classes

Utility functions are provided for logging begin and end messages for common operations (database, http, etc.). Begin log entries are tagged with 'begin.' End log entries are tagged with 'end.' The operationId property is added to both log entries with the same unique identifier generated for the operation. If an exception is thrown, an error is logged. These functions are implemented as static methods.

The following classes are available:

• TaskLogger: For logging calls to aynchronous and asynchronous functions
• GeneratorLogger: Creates an object that is useful for logging operations that produce data (aka generators), usually via events or iterators
• MySqlLogger: For logging SQL statement execution via mysql2
• RequestLogger: For logging http requests via request-promise

TaskLogger

Example

The following example sets result to 'Some data.'

TaskLogger = require('@goodware/log/TaskLogger');
Loggers = require('@goodware/log');

const logger = new Loggers({ defaultLevel: 'info' });

let result;
TaskLogger.execute(logger, async () => 'Some data', 'Doing it').then((value) => {
  result = value;
});

Maintainer Notes

Deployment

First, push to git.

1. Change the version number in package.json
2. npm run prepub
3. Commit and push
4. Publish to npm: npm run pub2