// This pouch is powered by Workers!
var db = new PouchDB('mydb', {adapter: 'worker'});
Adapter plugin to use PouchDB over Web Workers and Service Workers. Transparently proxies all PouchDB API requests to the worker, so that the most expensive database operations are run in a separate thread.
Basically, worker-pouch allows you use the PouchDB API like you normally would, but your UI will suffer fewer hiccups, because any blocking operations (such as IndexedDB or checksumming) are run inside of the worker.
The worker-pouch adapter passes the full PouchDB test suite. It requires PouchDB 5.0.0+.
Topics
$ npm install worker-pouch
This plugin has two modes:
In Easy Mode, you don't need to set up the worker yourself, because the script is loaded in a Blob URL. Whereas in Custom Mode, you must manage the Web Worker or Service Worker yourself.
You can do Easy Mode either with prebuilt JavaScript or via Browserify/Webpack.
The client JS file is available at node_modules/worker-pouch/dist/pouchdb.worker-pouch.js
. Or you can just download it from Github above (in which case, it will be available as window.workerPouch
).
Then include it in your HTML, after PouchDB:
<script src="https://github.com/pouchdb-community/worker-pouch/raw/master/pouchdb.js"></script>
<script src="https://github.com/pouchdb-community/worker-pouch/raw/master/pouchdb.worker-pouch.js"></script>
Then you can create a worker-powered PouchDB using:
var db = new PouchDB('mydb', {adapter: 'worker'});
The same rules apply, but you have to notify PouchDB of the new adapter:
var PouchDB = require('pouchdb');
PouchDB.adapter('worker', require('worker-pouch'));
Unfortunately, creating workers via Blob URLs is not supported in all browsers. In particular, IE, Edge, Safari, and iOS are not supported. Luckily, Firefox and Chrome are the browsers that benefit the most from web workers. There is also an API to detect browser support, which you must use if you would like to support browsers other than Firefox and Chrome.
In this mode, you manage the Web Worker yourself, and you register the two endpoints so that worker-pouch can communicate with the "backend" and "frontend."
Since this doesn't require Blob URLs, and because you can use custom PouchDB objects, you can potentially support more browsers this way. It's much more flexible.
This mode only supports bundling via Browserify/Webpack/etc. There is no prebuilt option.
To use, you'll need this code on the client side:
// client-side code
var PouchDB = require('pouchdb');
PouchDB.adapter('worker', require('worker-pouch/client'));
var worker = new Worker('worker.js');
var db = new PouchDB('mydb', {
adapter: 'worker',
worker: worker
});
Note that you create the PouchDB
object passing in both adapter: 'worker'
and worker
, which points to your Worker
object.
Then you include this code on the worker side:
// worker-side code
var registerWorkerPouch = require('worker-pouch/worker');
var PouchDB = require('pouchdb');
// attach to global `self` object
registerWorkerPouch(self, PouchDB);
If you would like to customize how PouchDB
is created inside of the worker, then you can also pass in a custom PouchDB factory function, which is a function that takes an options object (e.g. {name: 'mydb', auto_compaction: true}
)
and returns a PouchDB
object.
This is useful in cases where PouchDB's IndexedDB adapter doesn't work inside of a worker (such as Safari), so for instance you can have the pouchCreator
function
return an in-memory PouchDB
object.
Here's an example:
var PouchDB = require('pouchdb');
require('pouchdb/extras/memory');
function pouchCreator(opts) {
opts.adapter = 'memory'; // force in-memory mode
return new PouchDB(opts);
}
var registerWorkerPouch = require('worker-pouch/worker');
registerWorkerPouch(self, pouchCreator);
The PouchDB worker code will listen for messages from the client side, but should ignore any non-worker-pouch messages, so you are free to still use worker.postMessage()
as desired.
Communicating with a Service Worker is the same as with a Web Worker. However, you have to wait for the Service Worker to install and start controlling the page. Here's an example:
navigator.serviceWorker.register('sw.js', {
scope: './'
}).then(function () {
if (navigator.serviceWorker.controller) {
// already active and controlling this page
return navigator.serviceWorker;
}
// wait for a new service worker to control this page
return new Promise(function (resolve) {
function onControllerChange() {
navigator.serviceWorker.removeEventListener('controllerchange', onControllerChange);
resolve(navigator.serviceWorker);
}
navigator.serviceWorker.addEventListener('controllerchange', onControllerChange);
});
}).then(function (serviceWorker) { // the worker is ready
db = new PouchDB('testdb', {
adapter: 'worker',
worker: function() {
return serviceWorker;
}
});
return db;
}).catch(console.log.bind(console));
});
Then inside your Service Worker:
// worker-side code
var registerWorkerPouch = require('worker-pouch/worker');
var PouchDB = require('pouchdb');
// attach to global `self` object
registerWorkerPouch(self, PouchDB);
self.addEventListener('activate', function(event) {
event.waitUntil(self.clients.claim()); // activate right now
});
These numbers were recorded using this site. The test involved inserting 10000 PouchDB documents, and was run on a 2013 MacBook Air. Browser data was deleted between each test.
Time (ms) | Blocked the DOM? | |
---|---|---|
Chrome 48 | ||
put() - normal |
50070 | No |
put() - worker |
56993 | No |
bulkDocs() - normal |
2740 | Yes |
bulkDocs() - worker |
3454 | No |
Firefox 43 | ||
put() - normal |
39595 | No |
put() - worker |
41425 | No |
bulkDocs() - normal |
1027 | Yes |
bulkDocs() - worker |
1130 | No |
Basic takeaway: put()
s avoid DOM-blocking (due to using many smaller transactions), but are much slower than bulkDocs()
. With worker-pouch, though, you can get nearly all the speed benefit of bulkDocs()
without blocking the DOM.
(Note that by "blocked the DOM," I mean froze the animated GIF for a significant amount of time - at least a half-second. A single dropped frame was not penalized. Try the test yourself, and you'll see the difference is pretty stark.)
In Easy Mode, this plugin doesn't support all browsers. So it provides a special API to dianogose whether or not the current browser supports worker-pouch. Here's how you can use it:
var workerPouch = require('worker-pouch');
workerPouch.isSupportedBrowser().then(function (supported) {
var db;
if (supported) {
db = new PouchDB('mydb', {adapter: 'worker'});
} else { // fall back to a normal PouchDB
db = new PouchDB('mydb');
}
}).catch(console.log.bind(console)); // shouldn't throw an error
The isSupportedBrowser()
API returns a Promise for a boolean, which will be true
if the browser is supported and false
otherwise.
If you are using this method to return the PouchDB object itself from a Promise, be sure to wrap it in an object, to avoid "circular promise" errors:
var workerPouch = require('worker-pouch');
workerPouch.isSupportedBrowser().then(function (supported) {
if (supported) {
return {db: new PouchDB('mydb', {adapter: 'worker'})};
} else { // fall back to a normal PouchDB
return {db: new PouchDB('mydb')};
}
}).then(function (dbWrapper) {
var db = dbWrapper.db; // now I have a PouchDB
}).catch(console.log.bind(console)); // shouldn't throw an error
worker-pouch uses debug for logging. So in the browser, you can enable debugging by using PouchDB's logger:
PouchDB.debug.enable('pouchdb:worker:*');
Yes, you can use pure PouchDB inside of a Web Worker or Service Worker. But the point of this plugin is to let you use PouchDB from outside a Web Worker or Service Worker, and then have it transparently proxy to another PouchDB that is isolated in a Web Worker or Service Worker.
Only those browsers that 1) Allow service workers or 2) Allow blob URLs for Web Worker scripts and allow IndexedDB inside of a Web Worker. Today, that means Chrome and Firefox.
Not right now, although map/reduce is supported.
Yes, but apparently this cost is less than that of IndexedDB, because the DOM is significanty less blocked when using worker-pouch. Another thing to keep in mind is that PouchDB's internal document representation in IndexedDB is more complex than the PouchDB documents you insert. So you clone a small PouchDB object to send it to the worker, and then inside the worker it's exploded into a more complex IndexedDB object. IndexedDB itself has to clone as well, but the more complex cloning is done inside the worker.
It's a bit subtle. The answer is yes, if you do this:
var local = new PouchDB('local', {adapter: 'worker'});
local.replicate.to('http://example.com/db');
However, the answer is no if you do:
var local = new PouchDB('local', {adapter: 'worker'});
var remote = new PouchDB('http://example.com/db');
local.replicate.to(remote);
The reason is that when you create a remote PouchDB using new PouchDB('http://example.com/db')
, then that runs inside the UI thread. However, when you .replicate.to('http://example.com/db')
, then that string is passed ver-batim to the worker thread, where worker-pouch
becomes responsible for creating the remote PouchDB. Hence replication will occur inside of the worker thread.
In general, if you are very concerned about performance implications of what runs inside of the woker vs what runs outside of the worker, you are encouraged to not use worker-pouch
and to instead just run PouchDB inside a worker and handle message-passing yourself (might I recommend promise-worker?). This is the only way to really ensure that all PouchDB operations are isolated to the worker.
npm install
npm run build
Your plugin is now located at dist/pouchdb.worker-pouch.js
and dist/pouchdb.worker-pouch.min.js
and is ready for distribution.
Run npm run dev
and then point your favorite browser to http://127.0.0.1:8000/test/index.html.
The query param ?grep=mysearch
will search for tests matching mysearch
.
You can run e.g.
CLIENT=selenium:firefox npm test
CLIENT=selenium:phantomjs npm test
This will run the tests automatically and the process will exit with a 0 or a 1 when it's done. Firefox uses IndexedDB, and PhantomJS uses WebSQL.
Run:
npm run test-custom
Or to debug:
npm run test-custom-local