@chee/o-errors

Origami Client Side tracking library

Usage no npm install needed!

<script type="module">
  import cheeOErrors from 'https://cdn.skypack.dev/@chee/o-errors';
</script>

README

o-errors Build Status

This module provides a decoupled events-based mechanism for modules to report errors and an API for products to report client-side errors.

Important note

o-errors has an event driven API so it can be used without tightly coupling any application to using sentry & raven (which is a fairly large js library). When using o-errors within a component, prefer to use the event driven API over its methods. This leaves the decision about whether to include the library up to the application that consumes your component.

Requirements

Error tracking requires a project to be configured with Sentry as o-errors reports errors to this service.

See the Sentry documentation for setup specifics.


Quick Start

Include o-errors in the build and include a configuration <script> tag on your page configuring the Sentry DSN specific to your application (see Requirements).

<script type="application/json" data-o-errors-config>
{
    "sentryEndpoint": "https://dsn@app.getsentry.com/appid"
}
</script>

This will automatically configure o-errors on the o.DOMContentLoaded event and any uncaught errors will be reported.

Usage

Components that are included on the page might throw uncaught exceptions or emit an oErrors.log CustomEvent. o-errors listens on the document root for this event and aggregates errors to the configured Sentry endpoint.

Component developers should ensure that { bubbles: true } is set when constructing the CustomEvent and should dispatch it on the component's owned DOM.

Uncaught errors are handled by a window.onerror function installed on initialisation. Note uncaught errors will not be reported automatically if they occur before initialisation, although any errors reported using the o-errors API will be buffered and reported once initialised.

Reporting errors

Report errors using the oErrors.report method.

var oErrors = require('o-errors');

oErrors.report(new Error("My error"));

If the module has not been initialised, these errors are buffered and sent once initialised. Following initialisation uncaught errors are automatically recorded.

Configuration

Configure o-errors using the oErrors.init method, or using a <script> tag (see Using markup to configure o-errors).

Some options are only configurable when using the init method as they must accept a function.

Available options

sentryEndpoint - required

The DSN specific to your application.

{
    sentryEndpoint: "https:/dsn@app.getsentry.com/appid"
}
siteVersion - optional

The version of the product or site on which o-errors is configured. This is useful for associating errors with releases in the Sentry interface.

{
    siteVersion: "v1.0.0"
}
environment - optional

Track the environment name inside Sentry for the product or site on which o-errors is configured. This is useful for highlighting new issues in lower environments in the Sentry interface.

{
    environment: "production"
}
tags - optional

A hash of tags to apply to each error event. This is useful for associating errors to specfic parts of your application and the ability to filter within the Sentry interface.

{
    tags: {
        appName: "o-errors"
    }
}
logLevel - optional

o-errors provides a logging API specific to logging console messages. These messages are recorded so that additional contextual information can be added to any reported errors. Calling one of these functions does not report anything.

The log level could be one of the following options:

off - Turns off all logging by turning the logging API into a no-operation. This is the default and is reccommended for most production scenarios as most javascript environments will optimise the no-operation away.

contextonly - Turns on log context tracking, oErrors.log, oErrors.warn, and oErrors.error will store the last 10 log messages in a fast circular buffer. When an error is reported, these log lines are attached to the error as additional context in the context:log field and can be viewed in the Sentry interface. This can be particularly useful when understanding what the state of the application was before an error occurs. While this has a very low overhead it can increase the size of the error payload so is not the default.

debug - Turns on context tracking and outputs the messages through console.warn, console.log, and console.error for oErrors.warn, oErrors.log, and oErrors.error respectively. This should not be used in production as it will call the native console APIs.

consoleonly - Turns on console only logging. This will stop the Raven client from being initialised and reporting errors to sentry. Messages will be output through console.warn, console.log, and console.error for oErrors.warn, oErrors.log, and oErrors.error respectively. This should not be used in production.

{
    logLevel: "off"
}
enabled - optional

If false oErrors will be configured with no-operation methods, and errors will not be reported. Defaults to true.

{
    enabled: false
}
filterError - optional

A function that can be used to filter errors before they are reported. The function should accept one argument, a complicated looking object that comes from the n-raven client, with a bunch of fields, of which the most important is exception:

{
  ...
  "exception": {
    "values": [
      {
        "type": "TypeError",
        "value": "_nUi is undefined",
        "stacktrace": {
          "frames": [
            {
              "filename": "http://local.ft.com:5005/__dev/assets/public/main.js",
              "lineno": 1,
              "colno": 11,
              "function": "?",
              "in_app": true
            },
            ...
          ]
        }
      }
    ]
  },
  ...
}

The function should return a boolean inidicating whether the error should be sent or not, if true or coerced to a truthy value, the error will be sent, if false the error will be filtered.

This function should not mutate the data object or its fields. See transformError if you wish to mutate the data.

filterError is run after transformError.

Note: this may only be configured through the init method, it will report an Error and continue without filtering enabled if this is configured declaratively.

{
    filterError: function(data) { return true; }
}
transformError - optional

A function that can be used to transform errors before they are reported - this may be to add/remove data, or alter its format. The function should accept one argument, an object with two fields: an error field which contains the reported error, and a context field, which contains any additional reported context.

The function may return an object with at least an error field, if context is missing it will be filled in with an empty object. If no return type is given, the input value is used. Because the input value is an object, any fields or references that are mutated in the transform function are preserved.

This function may mutate the data object.

transformError is run before filterError.

Note: this may only be configured through the init method, it will report an Error and continue without transforms enabled if this is configured declaratively.

{
    transformError: function(data) { return data; }
}
errorBuffer - optional

An additional array of error objects which will be pushed into the internal buffer queue to be sent on initialisation. This can be useful if you have caught errors inline on your page before oErrors.init() has been ran. Each error in the array should be an object with two fields: an error field which contains the reported Error, and a context field, which contains any additional reported context.

Note: this may only be configured through the init method.

{
    errorBuffer: [
        {
            error: new Error(),
            context: {
                additional: "info"
            }
        }
    ]
}
transportFunction - optional

Function to override Sentry's HTTP transport handler. It accepts one options argument which is an object.

This function is passed on directly to Sentry to its transport configuration option and you can find the documentation for it here.

Using markup to configure oErrors

Include a <script> tag containing 'JSON' describing the configuration object.

<script type="application/json" data-o-errors-config>
{
    sentryEndpoint: "https:/dsn@app.getsentry.com/appid"
}
</script>

Using oErrors.init

Pass an object to oErrors.init with the appropriate configuration. If sentryEndpoint is not configured in the options, an Error is thrown.

var oErrors = require('o-errors');
oErrors.init({
    sentryEndpoint: "https://dsn@app.getsentry.com/appid",
    siteVersion:    "v1.0.0",                              // Optional
    logLevel:       "off"                                  // Optional
});

Product usage

This section outlines some typical use cases when integrating oErrors into a product.

Due to the nature of the Raven Sentry client, o-errors is a singleton. This means when you require it using browserify you'll always get the same instance.

Wrapping functions in an error handler

o-errors allows you to add context to errors to help the debug process. This is difficult to do when only using the global window.onerror method of catching uncaught errors.

Instead you can wrap blocks of code in an error handler and assign additional context to it:

oErrors.wrapWithContext(context, fn) - Wrap a function so that any uncaught errors thrown while executing the function are caught and reported to the error aggregator, along with the additional context data.

function renderAll(components) {
  components.forEach(render);
}

var components = [ componentA, componentB ];

var wrappedFunction = oErrors.wrapWithContext({ componentsToRender: components }, renderAll);

wrappedFunction(components);

Similar to wrapWithContext is wrap which does not accept a context argument.

wrapWithContext and wrap do no execute the given function, instead they return a wrapped version of the function.

oErrors.wrap(fn) - Wrap a function so that any uncaught errors thrown while executing the function are caught and reported.

Log data and associating logs with errors

o-errors allows you to log arbitrary strings using a console.log, console.warn, and console.error like API.

Instead of logging to the console however, if the logLevel configuration option is set to contextonly, the last few log lines are kept in memory. Then, when an uncaught error is logged to the error aggregator, the log lines are included as additional context. Calling these APIs does not report anything to the aggregator by themselves.

In this example, a message is logged before calling exampleSyncData. If exampleSyncData errors, then "Syncing data to APIs" would be attached to the error context and sent to the error aggregator.

oErrors.log("Syncing data to APIs")'
exampleSyncData();

This is not always desirable and extensive logging could affect performance. It can be turned on and off using the logLevel configuration variable when initialising the module. When logLevel is "off" the operations become 'noops' which are compiled out of your code by modern javascript engines.

oErrors.log(message) - Creates a 'log' level message, semantically equivalent to console.log oErrors.warn(message) - Creates a 'warn' level message, semantically equivalent to console.warn oErrors.error(message) - Creates a 'error' level message, semantically equivalent to console.error

Events

Events are primarily useful for reporting errors from components. If a component fires an oErrors.log event, if o-errors is configured, the error can be reported.

oErrors.log

A component can fire an oErrors.log event on its owned DOM to send an error report if oErrors has been configured on the page.

detail:

{
   error: e,  // the Error object that's been caught
   info:  i   // an object with further useful debug info
}

Because o-errors listens on document, the event must bubble.

Example:

o-mycomponent uses Promises to handle async events, it lives inside the DOM element, myComponentElement:

doThis().then(that).catch(function(e) {
    // Send error to oErrors
    var event = new CustomEvent('oErrors.log', {
        bubbles: true,
        detail: {
            error: e,
            info: { additional: "context" }
        }
    });

    // Dispatch on owned DOM
    myComponentElement.dispatchEvent(event);

    // re-throw event
    throw e;
});

License

Copyright (c) 2016 Financial Times Ltd. All rights reserved.

This software is published under the MIT licence.