Contribute to this guide

guideWatchdog

Every non-trivial piece of software has bugs. Despite our high quality standards like 100% code coverage, regression testing and manual tests before every release, CKEditor 5 is not free of bugs. Neither is the browser used by the user, your application in which CKEditor 5 is integrated, or any third-party addons that you used.

In order to limit the effect of an editor crash on the user experience, you can automatically restart the WYSIWYG editor with the content saved just before the crash.

The Watchdog utility allows you to do exactly that. It ensures that an editor instance is running, despite a potential crash. It works by detecting that an editor crashed, destroying it, and automatically creating a new instance of that editor with the content of the previous editor.

# Usage

Note: Watchdog can be used only with an editor built from source.

Install the @ckeditor/ckeditor5-watchdog package:

npm install --save @ckeditor/ckeditor5-watchdog

And then change your ClassicEditor.create() call to watchdog.create() as follows:

import ClassicEditor from '@ckeditor/ckeditor5-editor-classic/src/classiceditor';
import Watchdog from '@ckeditor/ckeditor5-watchdog/src/watchdog';

import Essentials from '@ckeditor/ckeditor5-essentials/src/essentials';
import Paragraph from '@ckeditor/ckeditor5-paragraph/src/paragraph';
import Bold from '@ckeditor/ckeditor5-basic-styles/src/bold';
import Italic from '@ckeditor/ckeditor5-basic-styles/src/italic';

// Create a watchdog for the given editor type.
const watchdog = Watchdog.for( ClassicEditor );

// Create a new editor instance.
watchdog.create( document.querySelector( '#editor' ), {
    plugins: [ Essentials, Paragraph, Bold, Italic ],
    toolbar: [ 'bold', 'italic', 'alignment' ]
} );

In other words, your goal is to create a watchdog instance and make the watchdog create an instance of the editor you want to use. The watchdog will then create a new editor and if it ever crashes, restart it by creating a new editor.

A new editor instance is created every time the watchdog detects a crash. Thus, the editor instance should not be kept in your application’s state. Use the Watchdog#editor property instead.

It also means that any code that should be executed for any new editor instance should be either loaded as an editor plugin or executed in the callbacks defined by Watchdog#setCreator() and Watchdog#setDestructor(). Read more about controlling the editor creation and destruction in the next section.

# Controlling editor creation and destruction

For more control over the creation and destruction of editor instances, you can use the Watchdog#setCreator() and Watchdog#setDestructor() methods:

// Instantiate the watchdog manually (do not use the for() helper).
const watchdog = new Watchdog();

watchdog.setCreator( ( elementOrData, editorConfig ) => {
    return ClassicEditor
        .create( elementOrData, editorConfig )
        .then( editor => {
            // Do something with the new editor instance.
        } );
} );

watchdog.setDestructor( editor => {
    // Do something before the editor is destroyed.

    return editor
        .destroy()
        .then( () => {
            // Do something after the editor is destroyed.
        } );
 } );

watchdog.create( elementOrData, editorConfig );

# API

Other useful methods, properties and events:

watchdog.on( 'error', () => { console.log( 'Editor crashed.' ) } );
watchdog.on( 'restart', () => { console.log( 'Editor was restarted.' ) } );

// Destroy the watchdog and the current editor instance.
watchdog.destroy();

// The current editor instance.
watchdog.editor;

// The current state of the editor.
// The editor might be in one of the following states:
// * `initializing` - before the first initialization, and after crashes, before the editor is ready,
// * `ready` - a state when a user can interact with the editor,
// * `crashed` - a state when an error occurs - it quickly changes to `initializing` or `crashedPermanently` depending on how many and how frequency errors have been caught recently,
// * `crashedPermanently` - a state when the watchdog stops reacting to errors and keeps the editor crashed,
// * `destroyed` - a state when the editor is manually destroyed by the user after calling `watchdog.destroy()`.
// This property is observable.
watchdog.state;

// Listen to state changes.
watchdog.on( 'change:state' ( evt, name, currentState, prevState ) => {
    console.log( `State changed from ${ currentState } to ${ prevState }` );

    if ( currentState === 'crashedPermanently' ) {
        watchdog.editor.isReadOnly = true;
    }
} );

// An array of editor crashes info.
watchdog.crashes.forEach( crashInfo => console.log( crashInfo ) );

# Configuration

Both, the Watchdog#constructor and the Watchdog.for methods accept a {configuration object with the following optional properties:

  • crashNumberLimit - A threshold specifying the number of editor errors (defaults to 3). After this limit is reached and the time between last errors is shorter than minimumNonErrorTimePeriod the watchdog changes its state to crashedPermanently and it stops restarting the editor. This prevents an infinite restart loop.
  • minimumNonErrorTimePeriod - An average amount of milliseconds between last editor errors (defaults to 5000). When the period of time between errors is lower than that and the crashNumberLimit is also reached the watchdog changes its state to crashedPermanently and it stops restarting the editor. This prevents an infinite restart loop.
  • saveInterval - A minimum number of milliseconds between saving editor data internally (defaults to 5000). Note that for large documents this might have an impact on the editor performance.
const watchdog = new Watchdog( {
    minimumNonErrorTimePeriod: 2000,
    crashNumberLimit: 4,
    saveInterval: 1000
} )

# Limitations

The CKEditor 5 watchdog listens to uncaught errors which can be associated with the editor instance created by that watchdog. Currently, these errors are CKEditorError errors so those explicitly thrown by the editor (by its internal checks). This means that not every runtime error that crashed the editor can be caught which, in turn, means that not every crash can be detected.

However, with time, the most “dangerous” places in the API will be covered with checks and try-catch blocks (allowing detecting unknown errors).

The watchdog does not handle errors thrown during the editor initialization (by Editor.create()) and editor destruction (Editor#destroy()). Errors thrown at these stages mean that there is a serious problem in the code integrating the editor with your application and such problem cannot be easily fixed by restarting the editor.