This guide applies to all development within the OpenSearch Dashboards project and is recommended for the development of all OpenSearch Dashboards plugins.
Besides the content in this developer guide, the following developer guides may also apply to all development within the OpenSearch Dashboards project. Please make sure to also read them:
If you would like to install and run this project, please see the Downloads Page.
You need to have an OpenSearch server up and running to be able to run OpenSearch Dashboards. The easiest way to do it is using Docker.
We recommend using Node Version Manager to install the node version we need.
While OpenSearch is starting, you can already bootstrap OpenSearch Dashboards:
$ git clone https://github.com/opensearch-project/OpenSearch-Dashboards.git
$ cd OpenSearch-Dashboards
$ nvm use
$ npm i -g yarn
$ yarn osd bootstrap # This command will also install npm dependenciesThis step is only mandatory if you have https/authentication enabled, or if you use the OpenSearch Docker image in its default configuration.
Once the bootstrap of OpenSearch Dashboards is finished, you need to apply some
changes to opensearch_dashboards.yml in order to connect to OpenSearch.
opensearch.hosts: ["https://localhost:9200"]
opensearch.username: "admin" # Default username on the docker image
opensearch.password: "admin" # Default password on the docker image
opensearch.ssl.verificationMode: noneAfter you've installed OpenSearch Dashboards and configured it, you can start the development server:
$ yarn startWhen the server is up and ready, click on the link displayed in your terminal to access it.
To build the Docker image, run the following:
yarn osd bootstrap
yarn build --docker
All filenames should use snake_case.
Right: src/opensearch-dashboards/index_patterns/index_pattern.js
Wrong: src/opensearch-dashboards/IndexPatterns/IndexPattern.js
We use a version management system. If a line of code is no longer needed, remove it, don't simply comment it out.
We are gradually moving the OpenSearch Dashboards code base over to Prettier. All TypeScript code
and some JavaScript code (check .eslintrc.js) is using Prettier to format code. You
can run node script/eslint --fix to fix linting issues and apply Prettier formatting.
We recommend you to enable running ESLint via your IDE.
Whenever possible we are trying to use Prettier and linting over written developer guide rules. Consider every linting rule and every Prettier rule to be also part of our developer guide and disable them only in exceptional cases and ideally leave a comment why they are disabled at that specific place.
This part contains developer guide rules around general (framework agnostic) HTML usage.
Use camel case for the values of attributes such as id and data-test-subj selectors.
<button id="veryImportantButton" data-test-subj="clickMeButton">
Click me
</button>The only exception is in cases where you're dynamically creating the value, and you need to use hyphens as delimiters:
buttons.map(btn => (
<button
id={`veryImportantButton-${btn.id}`}
data-test-subj={`clickMeButton-${btn.id}`}
>
{btn.label}
</button>
)It's important that when you write CSS/SASS selectors using classes, IDs, and attributes (keeping in mind that we should never use IDs and attributes in our selectors), that the capitalization in the CSS matches that used in the HTML. HTML and CSS follow different case sensitivity rules, and we can avoid subtle gotchas by ensuring we use the same capitalization in both of them.
When labeling elements (and for some other accessibility tasks) you will often need ids. Ids must be unique within the page i.e. no duplicate ids in the rendered DOM at any time.
Since we have some components that are used multiple times on the page, you must
make sure every instance of that component has a unique id. To make the generation
of those ids easier, you can use the htmlIdGenerator service in the @elastic/eui.
A React component could use it as follows:
import { htmlIdGenerator } from '@elastic/eui';
render() {
// Create a new generator that will create ids deterministic
const htmlId = htmlIdGenerator();
return (<div>
<label htmlFor={htmlId('agg')}>Aggregation</label>
<input id={htmlId('agg')}/>
</div>);
}Each id generator you create by calling htmlIdGenerator() will generate unique but
deterministic ids. As you can see in the above example, that single generator
created the same id in the label's htmlFor as well as the input's id.
A single generator instance will create the same id when passed the same argument to the function multiple times. But two different generators will produce two different ids for the same argument to the function, as you can see in the following example:
const generatorOne = htmlIdGenerator();
const generatorTwo = htmlIdGenerator();
// Those statements are always true:
// Same generator
generatorOne('foo') === generatorOne('foo');
generatorOne('foo') !== generatorOne('bar');
// Different generator
generatorOne('foo') !== generatorTwo('foo');This allows multiple instances of a single React component to now have different ids. If you include the above React component multiple times in the same page, each component instance will have a unique id, because each render method will use a different id generator.
You can also use this service outside of React.
The following developer guide rules are targeting development of server side API endpoints.
API routes must start with the /api/ path segment, and should be followed by the plugin id if applicable:
Right: /api/marvel/nodes
Wrong: /marvel/api/nodes
OpenSearch Dashboards uses snake_case for the entire API, just like OpenSearch. All urls, paths, query string parameters, values, and bodies should be snake_case formatted.
Right:
POST /api/opensearch-dashboards/index_patterns
{
"id": "...",
"time_field_name": "...",
"fields": [
...
]
}
The following developer guide rules apply for working with TypeScript/JavaScript files.
Whenever possible, write code in TypeScript instead of JavaScript, especially if it's new code. Check out TYPESCRIPT.md for help with this process.
You should prefer modern language features in a lot of cases, e.g.:
- Prefer
classoverprototypeinheritance - Prefer arrow function over function expressions
- Prefer arrow function over storing
this(noconst self = this;) - Prefer template strings over string concatenation
- Prefer the spread operator for copying arrays (
[...arr]) overarr.slice() - Use optional chaining (
?.) and nullish Coalescing (??) overlodash.get(and similar utilities)
Wherever possible, do not rely on mutable state. This means you should not reassign variables, modify object properties, or push values to arrays. Instead, create new variables, and shallow copies of objects and arrays:
// good
function addBar(foos, foo) {
const newFoo = { ...foo, name: 'bar' };
return [...foos, newFoo];
}
// bad
function addBar(foos, foo) {
foo.name = 'bar';
foos.push(foo);
}Since TypeScript 3.0 and the introduction of the
unknown type there are rarely any
reasons to use any as a type. Nearly all places of former any usage can be replace by either a
generic or unknown (in cases the type is really not known).
You should always prefer using those mechanisms over using any, since they are stricter typed and
less likely to introduce bugs in the future due to insufficient types.
If you’re not having any in your plugin or are starting a new plugin, you should enable the
@typescript-eslint/no-explicit-any
linting rule for your plugin via the .eslintrc.js config.
You should try avoiding non-null assertions (!.) wherever possible. By using them you tell
TypeScript, that something is not null even though by it’s type it could be. Usage of non-null
assertions is most often a side-effect of you actually checked that the variable is not null
but TypeScript doesn’t correctly carry on that information till the usage of the variable.
In most cases it’s possible to replace the non-null assertion by structuring your code/checks slightly different or using user defined type guards to properly tell TypeScript what type a variable has.
Using non-null assertion increases the risk for future bugs. In case the condition under which we assumed that the variable can’t be null has changed (potentially even due to changes in compeltely different files), the non-null assertion would now wrongly disable proper type checking for us.
If you’re not using non-null assertions in your plugin or are starting a new plugin, consider enabling the
@typescript-eslint/no-non-null-assertion
linting rule for you plugin in the .eslintrc.js config.
To avoid deep nesting of if-statements, always return a function's value as early as possible. And where possible, do any assertions first:
// good
function doStuff(val) {
if (val > 100) {
throw new Error('Too big');
}
if (val < 0) {
return false;
}
// ... stuff
}
// bad
function doStuff(val) {
if (val >= 0) {
if (val < 100) {
// ... stuff
} else {
throw new Error('Too big');
}
} else {
return false;
}
}This helps avoid temporary references and helps prevent typo-related bugs.
// best
function fullName({ first, last }) {
return `${first} ${last}`;
}
// good
function fullName(user) {
const { first, last } = user;
return `${first} ${last}`;
}
// bad
function fullName(user) {
const first = user.first;
const last = user.last;
return `${first} ${last}`;
}Directly accessing array values via index should be avoided, but if it is necessary, use array destructuring:
const arr = [1, 2, 3];
// good
const [first, second] = arr;
// bad
const first = arr[0];
const second = arr[1];These are numbers (or other values) simply used in line in your code. Do not use these, give them a variable name so they can be understood and changed easily.
// good
const minWidth = 300;
if (width < minWidth) {
...
}
// bad
if (width < 300) {
...
}Module dependencies should be written using native ES2015 syntax wherever possible (which is almost everywhere):
// good
import { mapValues } from 'lodash';
export mapValues;
// bad
const _ = require('lodash');
module.exports = _.mapValues;
// worse
define(['lodash'], function (_) {
...
});In those extremely rare cases where you're writing server-side JavaScript in a file that does not pass run through webpack, then use CommonJS modules.
In those even rarer cases where you're writing client-side code that does not run through webpack, then do not use a module loader at all.
The files inside a module are implementation details of that module. They should never be imported directly. Instead, you must only import the top-level API that's exported by the module itself.
Without a clear mechanism in place in JS to encapsulate protected code, we make a broad assumption that anything beyond the root of a module is an implementation detail of that module.
On the other hand, a module should be able to import parent and sibling modules.
// good
import foo from 'foo';
import child from './child';
import parent from '../';
import ancestor from '../../../';
import sibling from '../foo';
// bad
import inFoo from 'foo/child';
import inSibling from '../foo/child';Don't do this. Everything should be wrapped in a module that can be depended on by other modules. Even things as simple as a single value should be a module.
And never use multiple ternaries together, because they make it more difficult to reason about how different values flow through the conditions involved. Instead, structure the logic for maximum readability.
// good, a situation where only 1 ternary is needed
const foo = a === b ? 1 : 2;
// bad
const foo = a === b ? 1 : a === c ? 2 : 3;Any non-trivial conditions should be converted to functions or assigned to descriptively named variables. By breaking up logic into smaller, self-contained blocks, it becomes easier to reason about the higher-level logic. Additionally, these blocks become good candidates for extraction into their own modules, with unit-tests.
// best
function isShape(thing) {
return thing instanceof Shape;
}
function notSquare(thing) {
return !(thing instanceof Square);
}
if (isShape(thing) && notSquare(thing)) {
...
}
// good
const isShape = thing instanceof Shape;
const notSquare = !(thing instanceof Square);
if (isShape && notSquare) {
...
}
// bad
if (thing instanceof Shape && !(thing instanceof Square)) {
...
}// good
const validPassword = /^(?=.*\d).{4,}$/;
if (password.length >= 4 && validPassword.test(password)) {
console.log('password is valid');
}
// bad
if (password.length >= 4 && /^(?=.*\d).{4,}$/.test(password)) {
console.log('losing');
}Keep your functions short. A good function fits on a slide that the people in the last row of a big room can comfortably read. So don't count on them having perfect vision and limit yourself to ~15 lines of code per function.
For expressiveness sake, and so you can be mix dynamic and explicit arguments.
// good
function something(foo, ...args) {
...
}
// bad
function something(foo) {
const args = Array.from(arguments).slice(1);
...
}Always use the default argument syntax for optional arguments.
// good
function foo(options = {}) {
...
}
// bad
function foo(options) {
if (typeof options === 'undefined') {
options = {};
}
...
}And put your optional arguments at the end.
// good
function foo(bar, options = {}) {
...
}
// bad
function foo(options = {}, bar) {
...
}For trivial examples (like the one that follows), thunks will seem like overkill, but they encourage isolating the implementation details of a closure from the business logic of the calling code.
// good
function connectHandler(client, callback) {
return () => client.connect(callback);
}
setTimeout(connectHandler(client, afterConnect), 1000);
// not as good
setTimeout(() => {
client.connect(afterConnect);
}, 1000);
// bad
setTimeout(() => {
client.connect(() => {
...
});
}, 1000);Use slashes for both single line and multi line comments. Try to write comments that explain higher level mechanisms or clarify difficult segments of your code. Don't use comments to restate trivial things.
Exception: Comment blocks describing a function and its arguments
(docblock) should start with /**, contain a single * at the beginning of
each line, and end with */.
// good
// 'ID_SOMETHING=VALUE' -> ['ID_SOMETHING=VALUE', 'SOMETHING', 'VALUE']
const matches = item.match(/ID_([^\n]+)=([^\n]+)/));
/**
* Fetches a user from...
* @param {string} id - id of the user
* @return {Promise}
*/
function loadUser(id) {
// This function has a nasty side effect where a failure to increment a
// redis counter used for statistics will cause an exception. This needs
// to be fixed in a later iteration.
...
}
const isSessionValid = (session.expires < Date.now());
if (isSessionValid) {
...
}
// bad
// Execute a regex
const matches = item.match(/ID_([^\n]+)=([^\n]+)/));
// Usage: loadUser(5, function() { ... })
function loadUser(id, cb) {
// ...
}
// Check if the session is valid
const isSessionValid = (session.expires < Date.now());
// If the session is valid
if (isSessionValid) {
...
}Feel free to use getters that are free from side effects, like providing a length property for a collection class.
Do not use setters, they cause more problems than they can solve.
When writing a new component, create a sibling SASS file of the same name and import directly into the top of the JS/TS component file. Doing so ensures the styles are never separated or lost on import and allows for better modularization (smaller individual plugin asset footprint).
All SASS (.scss) files will automatically build with the EUI & OpenSearch Dashboards invisibles (SASS variables, mixins, functions) from the globals_[theme].scss file.
While the styles for this component will only be loaded if the component exists on the page, the styles will be global and so it is recommended to use a three letter prefix on your classes to ensure proper scope.
Example:
// component.tsx
import './component.scss';
// All other imports below the SASS import
export const Component = () => {
return (
<div className="plgComponent" />
);
}// component.scss
.plgComponent { ... }Do not use the underscore _ SASS file naming pattern when importing directly into a javascript file.
The following developer guide rules are specific for working with the React framework.
When using ngReact to embed your react components inside Angular HTML, prefer the
reactDirective service over the react-component directive.
You can read more about these two ngReact methods here.
Using react-component means adding a bunch of components into angular, while reactDirective keeps them isolated, and is also a more succinct syntax.
Good:
<hello-component
fname="person.fname"
lname="person.lname"
watch-depth="reference"
></hello-component>Bad:
<react-component name="HelloComponent" props="person" watch-depth="reference" />Name action functions in the form of a strong verb and passed properties in the form of on. E.g:
<sort-button onClick={action.sort}/>
<pagerButton onPageNext={action.turnToNextPage} />Parts of the JavaScript developer guide were initially forked from the node style guide created by Felix Geisendörfer which is licensed under the CC BY-SA 3.0 license.