Roarr

JSON logger for Node.js and browser.

• Motivation
• Usage
  • Prepending context using the global state
  • Filtering logs
    • jq primer
• Log message format
• API
  • child
  • trace
  • debug
  • info
  • warn
  • error
  • fatal
• CLI program
• Transports
• Environment variables
• Conventions
  • Context property names
  • Using Roarr in an application
  • Using Roarr in modules
• Recipes
  • Logging errors
  • Using with Elasticsearch

Motivation

For a long time I have been a big fan of using debug. debug is simple to use, works in Node.js and browser, does not require configuration and it is fast. However, problems arise when you need to parse logs. Anything but one-line text messages cannot be parsed in a safe way.

To log structured data, I have been using Winston and Bunyan. These packages are great for application-level logging. I have preferred Bunyan because of the Bunyan CLI program used to pretty-print logs. However, these packages require program-level configuration – when constructing an instance of a logger, you need to define the transport and the log-level. This makes them unsuitable for use in code designed to be consumed by other applications.

Then there is pino. pino is fast JSON logger, it has CLI program equivalent to Bunyan, it decouples transports, and it has sane default configuration. Unfortunately, you still need to instantiate logger instance at the application-level. This makes it more suitable for application-level logging just like Winston and Bunyan.

I needed a logger that:

• Does not require initialisation.
• Produces structured data.
• Decouples transports.
• Has a CLI program.
• Works in Node.js and browser.
• Configurable using environment variables and global namespace.

In other words,

• a logger that I can use in an application code and in dependencies.
• a logger that allows to correlate logs between the main application code and the dependency code.
• a logger that works well with transports in external processes.

Roarr is this logger.

Usage

Roarr logging is disabled by default. To enable logging, you must start program with an environment variable ROARR_LOG set to true, e.g.

ROARR_LOG=true node ./index.js

import log from 'roarr';

log('foo');

log('bar %s', 'baz');

const debug = log.child({
  logLevel: 10
});

debug('qux');

debug({
  quuz: 'corge'
}, 'quux');

Produces output:

{"context":{},"message":"foo","sequence":0,"time":1506776210000,"version":"1.0.0"}
{"context":{},"message":"bar baz","sequence":1,"time":1506776210000,"version":"1.0.0"}
{"context":{"logLevel":10},"message":"qux","sequence":2,"time":1506776210000,"version":"1.0.0"}
{"context":{"logLevel":10,"quuz":"corge"},"sequence":3,"message":"quux","time":1506776210000,"version":"1.0.0"}

Prepending context using the global state

Prepending context using the global state will affect all roarr logs.

import log from 'roarr';

log('foo');

global.ROARR.prepend = {
  taskId: 1
};

log('bar');

global.ROARR.prepend = {};

log('baz');

Produces output:

{"context":{},"message":"foo","sequence":0,"time":1506776210000,"version":"1.0.0"}
{"context":{"taskId":1},"message":"bar","sequence":1,"time":1506776210000,"version":"1.0.0"}
{"context":{},"message":"baz","sequence":2,"time":1506776210000,"version":"1.0.0"}

Prepending context using the global state is useful when the desired result is to associate all logs with a specific context for a duration of an operation, e.g. to correlate the main process logs with the dependency logs.

import log from 'roarr';
import foo from 'foo';

const taskIds = [
  1,
  2,
  3
];

for (const taskId of taskIds) {
  global.ROARR = global.ROARR || {};
  global.ROARR.prepend = {
    taskId
  };

  log('starting task ID %d', taskId);

  // In this example, `foo` is an arbitrary third-party dependency that is using
  // roarr logger.
  foo(taskId);

  log('successfully completed task ID %d', taskId);

  global.ROARR.prepend = {};
}

Produces output:

{"context":{"taskId":1},"message":"starting task ID 1","sequence":0,"time":1506776210000,"version":"1.0.0"}
{"context":{"taskId":1},"message":"foo","sequence":1,"time":1506776210000,"version":"1.0.0"}
{"context":{"taskId":1},"message":"successfully completed task ID 1","sequence":2,"time":1506776210000,"version":"1.0.0"}
[...]

Filtering logs

Roarr is designed to print all or none logs (refer to the ROARR_LOG environment variable documentation).

To filter logs you need to use a JSON processor, e.g. jq.

jq primer

jq allows you to filter JSON messages using select(boolean_expression), e.g.

ROARR_LOG=true node ./index.js | jq 'select(.context.logLevel > 40)'

Combine it with roarr pretty-print to pretty-print a subset of the logs:

ROARR_LOG=true node ./index.js | jq -cM 'select(.context.logLevel > 40)'

(Notice the use of -cM parameters to disable JSON colarization and formatting.)

If your application outputs non-JSON output, jq will fail with an error similar to:

parse error: Invalid numeric literal at line 1, column 5
Error: write EPIPE
    at _errnoException (util.js:1031:13)
    at WriteWrap.afterWrite (net.js:873:14)

To ignore the non-JSON output, use jq -R flag (raw input) in combination with fromjson, e.g.

ROARR_LOG=true node ./index.js | jq -cRM 'fromjson? | select(.context.logLevel > 40)'

Log message format

Example:

{
  "context": {
    "application": "task-runner",
    "hostname": "curiosity.local",
    "instanceId": "01BVBK4ZJQ182ZWF6FK4EC8FEY",
    "taskId": 1
  },
  "message": "starting task ID 1",
  "sequence": 0,
  "time": 1506776210000,
  "version": "1.0.0"
}

API

roarr package exports a function that accepts the following API:

export type LoggerType =
  (
    context: MessageContextType,
    message: string,
    c?: SprintfArgumentType,
    d?: SprintfArgumentType,
    e?: SprintfArgumentType,
    f?: SprintfArgumentType,
    g?: SprintfArgumentType,
    h?: SprintfArgumentType,
    i?: SprintfArgumentType,
    k?: SprintfArgumentType
  ) => void |
  (
    message: string,
    b?: SprintfArgumentType,
    c?: SprintfArgumentType,
    d?: SprintfArgumentType,
    e?: SprintfArgumentType,
    f?: SprintfArgumentType,
    g?: SprintfArgumentType,
    h?: SprintfArgumentType,
    i?: SprintfArgumentType,
    k?: SprintfArgumentType
  ) => void;

To put it into words:

• First parameter can be either a string (message) or an object.
  • If first parameter is an object (context), the second parameter must be a string (message).
• Arguments after the message parameter are used to enable printf message formatting.
  • Printf arguments must be of a primitive type (string | number | boolean | null).
  • There can be up to 9 printf arguments (or 8 if the first parameter is the context object).

Refer to the Usage documentation for common usage examples.

child

Creates a child logger appending the provided context object to the previous logger context.

type ChildType = (context: MessageContextType) => LoggerType;

trace

debug

info

warn

error

fatal

Convenience methods for logging a message with logLevel context property value set to the name of the convenience method, e.g.

import log from 'roarr';

log.trace('foo');
log.debug('foo');
log.info('foo');
log.warn('foo');
log.error('foo');
log.fatal('foo');

Produces output:

{"context":{"logLevel":10},"message":"foo","sequence":0,"time":1506776210000,"version":"1.0.0"}
{"context":{"logLevel":20},"message":"foo","sequence":1,"time":1506776210000,"version":"1.0.0"}
{"context":{"logLevel":30},"message":"foo","sequence":2,"time":1506776210000,"version":"1.0.0"}
{"context":{"logLevel":40},"message":"foo","sequence":3,"time":1506776210000,"version":"1.0.0"}
{"context":{"logLevel":50},"message":"foo","sequence":4,"time":1506776210000,"version":"1.0.0"}
{"context":{"logLevel":60},"message":"foo","sequence":5,"time":1506776210000,"version":"1.0.0"}

CLI program

Roarr comes with a CLI program used to pretty-print logs for development purposes.

To format the logs, pipe the program output to roarr pretty-print program, e.g.

$ npm install roarr -g
$ ROARR_LOG=true node index.js | roarr pretty-print

Provided that the index.js program produced an output such as:

{"context":{"package":"forward-proxy","namespace":"createHttpProxyServer","logLevel":30},"message":"Internal SSL Server running on localhost:62597","sequence":0,"time":1506803138704,"version":"1.0.0"}
{"context":{"package":"forward-proxy","namespace":"createRequestProcessor","logLevel":30},"message":"request start -> http://localhost:62595/","sequence":1,"time":1506803138741,"version":"1.0.0"}
{"context":{"package":"forward-proxy","namespace":"createLogInterceptor","logLevel":20,"headers":{"host":"localhost:62595","connection":"close"}},"message":"received request","sequence":2,"time":1506803138741,"version":"1.0.0"}
{"context":{"package":"forward-proxy","namespace":"createRequestProcessor","logLevel":30},"message":"request finished <- http://localhost:62595/","sequence":3,"time":1506803138749,"version":"1.0.0"}
{"context":{"package":"forward-proxy","namespace":"createLogInterceptor","logLevel":30,"method":"GET","requestHeaders":{"host":"localhost:62595","connection":"close"},"responseHeaders":{"date":"Sat, 30 Sep 2017 20:25:38 GMT","connection":"close","content-length":"7","x-forward-proxy-request-id":"2b746d92-1a8b-4f36-b3cc-5bff57dad94d","x-forward-proxy-cache-hit":"false"},"statusCode":200,"url":"http://localhost:62595/"},"message":"response","sequence":4,"time":1506803138755,"version":"1.0.0"}
{"context":{"package":"forward-proxy","namespace":"createLogInterceptor","logLevel":30,"method":"GET","requestHeaders":{"host":"localhost:62595","connection":"close"},"responseHeaders":{"date":"Sat, 30 Sep 2017 20:25:38 GMT","content-length":"7","x-forward-proxy-request-id":"2b746d92-1a8b-4f36-b3cc-5bff57dad94d","x-forward-proxy-cache-hit":"true"},"statusCode":200,"url":"http://localhost:62595/"},"message":"response","sequence":5,"time":1506803138762,"version":"1.0.0"}

roarr CLI program will format the output to look like this:

• @ prefixed value denotes the name of the package.
• # prefixed value denotes the namespace.

The roarr pretty-print CLI program is using the context property names suggested in the conventions to pretty-print the logs for the developer inspection purposes.

Explore other CLI commands and options using roarr --help.

Transports

A transport in most logging libraries is something that runs in-process to perform some operation with the finalised log line. For example, a transport might send the log line to a standard syslog server after processing the log line and reformatting it.

Roarr does not support in-process transports.

Roarr does not support in-process transports because Node processes are single threaded processes (ignoring some technical details). Given this restriction, Roarr purposefully offloads handling of the logs to external processes so that the threading capabilities of the OS can be used (or other CPUs).

Depending on your configuration, consider one of the following log transports:

• Beats for aggregating at a process level (written in Go).
• logagent for aggregating at a process level (written in JavaScript).
• Fluentd for aggregating logs at a container orchestration level (e.g. Kubernetes) (written in Ruby).

Environment variables

When running the script in a Node.js environment, use environment variables to control roarr behaviour.

When using ROARR_STREAM=STDERR, use 3>&1 1>&2 2>&3 3>&- to pipe stderr output.

Conventions

Context property names

Roarr does not have reserved context property names. However, I encourage use of the following conventions:

The roarr pretty-print CLI program is using the context property names suggested in the conventions to pretty-print the logs for the developer inspection purposes.

Log levels

The roarr pretty-print CLI program translates logLevel values to the following human-readable names:

Using Roarr in an application

To avoid code duplication, you can use a singleton pattern to export a logger instance with predefined context properties (e.g. describing the application).

I recommend to create a file Logger.js in the project directory. Use this file to create an child instance of Roarr with context parameters describing the project and the initialisation instance, e.g.

/**
 * @file Example contents of a Logger.js file.
 */

import log from 'roarr';
import ulid from 'ulid';

// Instance ID is useful for correlating logs in high concurrency environment.
//
// See `roarr augment --append-instance-id` option as an alternative way to
// append instance ID to all logs.
const instanceId = ulid();

// The reason we are using `global.ROARR.prepend` as opposed to `roarr#child`
// is because we want this information to be prepended to all logs, including
// those of the "my-application" dependencies.
//
// Note: If you are adding logger to a package intended to be consumed by other
// packages, you must not set `global.ROARR.prepend`. Instead, use `roarr#child`.
global.ROARR.prepend = {
  ...global.ROARR.prepend,
  application: 'my-application',
  instanceId
};

const Logger = log.child({
  // .foo property is going to appear only in the logs that are created using
  // the current instance of a Roarr logger.
  foo: 'bar'
});

export default Logger;

Using Roarr in modules

If you are developing a code that is designed to be consumed by other applications/ modules, then you should avoid using global.ROARR (though, there are valid use cases). However, you should still start the project by defining a Logger.js file and use log.child instead.

/**
 * @file Example contents of a Logger.js file.
 */
import Roarr from 'roarr';

export default Roarr.child({
  domain: 'database',
  package: 'my-package'
});

Roarr does not have reserved context property names. However, I encourage use of the conventions. The roarr pretty-print CLI program is using the context property names suggested in the conventions to pretty-print the logs for the developer inspection purposes.

Recipes

Logging errors

This is not specific to Roarr – this suggestion applies to any kind of logging.

If you want to include an instance of Error in the context, you must serialize the error.

The least-error prone way to do this is to use an existing library, e.g. serialize-error.

import log from 'roarr';
import serializeError from 'serialize-error';

// [..]

send((error, result) => {
  if (error) {
    log.error({
      error: serializeError(error)
    }, 'message not sent due to a remote error');

    return;
  }

  // [..]
});

Without using serialisation, your errors will be logged without the error name and stack trace.

Using with Elasticsearch

If you are using Elasticsearch, you will want to create an index template.

The following serves as the ground work for the index template. It includes the main Roarr log message properties (context, message, time) and the context properties suggested in the conventions.

{
  "mappings": {
    "log_message": {
      "_source": {
        "enabled": true
      },
      "dynamic": "strict",
      "properties": {
        "context": {
          "dynamic": true,
          "properties": {
            "application": {
              "type": "keyword"
            },
            "hostname": {
              "type": "keyword"
            },
            "instanceId": {
              "type": "keyword"
            },
            "logLevel": {
              "type": "integer"
            },
            "namespace": {
              "type": "text"
            },
            "package": {
              "type": "text"
            }
          }
        },
        "message": {
          "type": "text"
        },
        "time": {
          "format": "epoch_millis",
          "type": "date"
        }
      }
    }
  },
  "template": "logstash-*"
}