New WebSockets client guide #40931

wbamberg · 2025-08-28T22:13:19Z

This came out of #40188, which wanted to talk about how to make WebSocket clients bfcache-compatible, and https://github.com/orgs/mdn/discussions/817, in which we find that the example that's supposed to underpin the current Writing WebSocket client applications page doesn't work.

In mdn/dom-examples#326 I added a new example that does work and is bfcache-compatible.

This PR updates the guide page to use that example as a backbone. I hope I haven't jettisoned anything too important from the old page.

I wonder if we ought to say more about URL schemes: this example uses ws which is maybe shouldn't, but it can't use wss and still work with localhost AFAICT. But it could I think use a relative URL, and maybe should?

I also don't talk about the protocols option, as we don't use it in the example and it would feel like quite a big digression. Maybe that would be better as a separate guide?

On the differences between different browsers: this is mostly just what I have observed by testing. But this: https://docs.google.com/document/d/1JtDCN9A_1UBlDuwkjn1HWxdhQ1H2un9K4kyPLgBqJUc/edit?tab=t.0#heading=h.58d6ijfz2say is also a useful doc.

I wasn't sure about the "Security considerations" bit: I've kept it as-is but AFAIK mixed content just isn't allowed any more, and I couldn't find any more detail on "Most browsers now only allow secure WebSocket connections" (though I would love to know more about this, because if so it would be good to document it in BCD and describe it better here...).

I wouldn't be particularly averse to adding a note in https://developer.mozilla.org/en-US/docs/Web/API/WebSockets_API about bfcache-compatibility as well, if we think it merits more visibility than this.

If this merges I should file a follow-up to update https://developer.mozilla.org/en-US/docs/Web/API/WebSockets_API/Writing_a_WebSocket_server_in_JavaScript_Deno with the new example, too.

cc @gilbertococchi and maybe @tunetheweb ?

Fixes #29029

github-actions · 2025-08-28T22:15:17Z

seems not a single file was built! 🙀

(comment last updated: 2025-09-09 15:02:02)

chrisdavidmills

Nice work @wbamberg; this is a massive improvement over what we had before, in terms of both demo and guide. I've left you a few comments, but nothing major really.

I would still like to see a tech review from the other folks on the thread.

files/en-us/web/api/websockets_api/writing_websocket_client_applications/index.md

chrisdavidmills · 2025-08-29T08:39:45Z

files/en-us/web/api/websockets_api/writing_websocket_client_applications/index.md

-    If you don't specify a protocol string, an empty string is assumed.
+The `WebSocket` constructor takes one mandatory argument, which is the URL of the WebSocket server to connect to. In this case, since we're running the server locally, we're using the address of localhost.
+
+The constructor takes another optional argument [`protocols`](/en-US/docs/Web/API/WebSocket/WebSocket#protocols), which allows a single server to implement multiple sub-protocols. We're not using this feature in our example.


I agree with your note that this probably should be discussed in a separate guide. It certainly doesn't belong in a basic usage guide, although from reading this and following the link, it feels like it'd be nicer if we provided some more info about it somewhere on MDN.

files/en-us/web/api/websockets_api/writing_websocket_client_applications/index.md

chrisdavidmills · 2025-08-29T09:09:41Z

files/en-us/web/api/websockets_api/writing_websocket_client_applications/index.md

+
+The back/forward cache, or {{glossary("bfcache")}}, enables much faster back and forward navigation between pages that the user has recently visited. It does this by storing a complete snapshot of the page, including the JavaScript heap.
+
+The browser pauses and then resumes JavaScript execution when a page is added to or loaded from the bfcache. This means that, depending on what the page is doing, it's not always safe for the browser to use the bfcache for the page. If the browser determines that it is not safe, then the page will not be added to the bfcache, and then the user will not get the performance benefit that it can bring.


Suggested change

The browser pauses and then resumes JavaScript execution when a page is added to or loaded from the bfcache. This means that, depending on what the page is doing, it's not always safe for the browser to use the bfcache for the page. If the browser determines that it is not safe, then the page will not be added to the bfcache, and then the user will not get the performance benefit that it can bring.

The browser pauses and then resumes JavaScript execution when a page is added to or loaded from the bfcache. This means that, depending on what the page is doing, it's not always safe for the browser to use the bfcache for the page. If the browser determines that it is not safe, the page will not be added to the bfcache, and the user will not get the performance benefit that it can bring.

Also, maybe link to https://developer.mozilla.org/en-US/docs/Web/API/Performance_API/Monitoring_bfcache_blocking_reasons for more information on why pages might not be added to the bfcache.

I wonder why you are deleting my "then"s? I like my thens. "If....then" I think is clearer structurally than an implicit then.

I think it's a shame that this is presented as a page on how to use a Chrome-only feature. It would be better to have a page under https://developer.mozilla.org/en-US/docs/Web/Performance, that talks about the bfcache as it works across browsers, and the various reasons pages might not be bfcache-friendly in various browsers, and links to notRestoredReasons as a tool you can use in that context.

I would say "restored from the bfcache" rather than "loaded from the bfcache".

The bfcache isn't a "cache" really. More a preservation and restore of a page's state.

I think it's a shame that this is presented as a page on how to use a Chrome-only feature. It would be better to have a page under https://developer.mozilla.org/en-US/docs/Web/Performance, that talks about the bfcache as it works across browsers, and the various reasons pages might not be bfcache-friendly in various browsers, and links to notRestoredReasons as a tool you can use in that context.

That's fair. The concepts are generally the same. It's just that only Chrome currently exposes the reasons in an easy manner. I think testing each in each browser is gonna be quite painful without that as could run into false positives (e.g. you think it didn't use bfcache for X reasons, but it was actually for Y reason).

Either way I'm in two minds whether you should link it. I mean it is the only definitive list (and is part of the HTML spec btw!), but also is only queryable from Chrome as you say.

Yes, sorry, I didn't mean to sound so negative about it! It is a good article and obviously a useful API. And we should link to it. I just think it would be great to have a browser-agnostic article about bfcache compatibility on MDN, that's not tied to specific APIs. But I'm not planning to write that any time soon.

-> ef5e49c and 86ebca5

chrisdavidmills · 2025-08-29T09:14:06Z

files/en-us/web/api/websockets_api/writing_websocket_client_applications/index.md

+});
+```
+
+Conversely, by listening for the {{domxref("Window.pageshow_event", "pageshow")}} event, you can seamlessly start the connection again when the page is loaded, or when it is retrieved from the bfcache. So in our example we will add all the code to initialize our WebSocket and set up its event listeners inside the `pageshow` event handler:


Suggested change

Conversely, by listening for the {{domxref("Window.pageshow_event", "pageshow")}} event, you can seamlessly start the connection again when the page is loaded, or when it is retrieved from the bfcache. So in our example we will add all the code to initialize our WebSocket and set up its event listeners inside the `pageshow` event handler:

Conversely, by listening for the {{domxref("Window.pageshow_event", "pageshow")}} event, you can seamlessly start the connection again when the page is loaded or retrieved from the bfcache. In our example, we add all the code to initialize our WebSocket and set up its event listeners inside the `pageshow` event handler:

Again I think I like my extra words around ", or when it is". I think it makes the structure clearer. I suppose this is subjective?

pageshow is also fired on page load. So might wanna mention that (to avoid people opening a connection twice on pageload).

Again I think I like my extra words around ", or when it is". I think it makes the structure clearer. I suppose this is subjective?

Yes, I think so. They seem a little wordy to me, but it is not wrong. Up to you whether you want to commit these suggestions.

pageshow is also fired on page load. So might wanna mention that (to avoid people opening a connection twice on pageload).

I do mention this:

you can seamlessly start the connection again when the page is loaded or retrieved from the bfcache.

good enough?

Mmm.... it's the "start the connection again" which is not quite working for me. If you're using this as the main way of initialising the web socket on initial page load, then it't not again. Also "when the page is loaded or retrieved from the bfcache" could easily be misread to mean "when the page is (loaded or retrieved) from the bfcache" rather than "when the page (is loaded) or (retrieved from the bfcache)"

WDYT about something more explicit like this?:

Conversely, by listening for the {{domxref("Window.pageshow_event", "pageshow")}} event, you can seamlessly start the connection again when the page is retrieved from the bfcache. Since the pageshow event also fires on page load, this can also be used to initiate the websocket connection on initial page load."

ah, makes sense, I missed the "again". Does f15529d work for you?

LGTM! Thanks.

chrisdavidmills · 2025-08-29T09:17:47Z

files/en-us/web/api/websockets_api/writing_websocket_client_applications/index.md

 ```

-It may be helpful to examine the socket's {{domxref("WebSocket.bufferedAmount", "bufferedAmount")}} attribute before attempting to close the connection to determine if any data has yet to be transmitted on the network.
-If this value isn't 0, there's pending data still, so you may wish to wait before closing the connection.
+If you run our example, try navigating to a different page, then back to the example. In Chrome, you should see that the example starts the connection again, and keeps its original context: so, for example, it remembers the count of exchanged messages.


In Firefox and Safari, it just seems to start the app again from its initial state, which is a shame. Worth mentioning?

This is not what I see in Safari, and I would like to know why we see different things here.

But it is what I see in Firefox, and that's expected: because the page isn't added to the bfcache, then it is reloaded, so the log starts out empty and the counter starts again at zero.

In Safari, after a forward/back navigation, I see this:

Here, I clicked "forward" after exchange 4, and then "back". I think what happens here is that:

the page was added to/loaded from the bfcache, so the HTML/JS is not reloaded, and we keep the contents of the log and the counter value

but because Safari closed the network request as soon as we navigated forward, we get a WebSocket error notification

At least this is my understanding of the situation. It is perhaps worth spelling this out in the guide, maybe not to the screenshot level of detail though.

Hi @wbamberg, thanks for raising this.

I would love to hear from someone from WebKit if possible if they can share what Safari is doing with WebSocket.

I also noticed that Safari is behaving somehow differently from Chrome and Firefox.

It looks like pages using WebSocket without closing the connection via pagehide are still eligible for BFCache on Safari somehow.
From my tests it looked like the approach of closing the connection via pagehide and restarting the connection via pageshow was also working in Safari.

Can you please share the error you are getting?

Correction: Safari is definitively throwing a readyState 3 error (CLOSED), although I see the connection restarting on Safari too on my test page (https://output.jsbin.com/sepojum).

Did you see the WebSocket connection not working on a BFCache restored page on Safari while testing?

It looks like pages using WebSocket without closing the connection via pagehide are still eligible for BFCache on Safari somehow.

This is what I would have expected, given the info in https://docs.google.com/document/d/1JtDCN9A_1UBlDuwkjn1HWxdhQ1H2un9K4kyPLgBqJUc/edit?tab=t.0#heading=h.yqabnzb880lx :

Safari caches such pages but cancels active network requests.

This is not what I see in Safari, and I would like to know why we see different things here.

Ah, hold on. I've tried it again in Safari. When I navigate to a new page by typing the URL into the address bar and pressing Enter (or following a link), then press "Back", I get the behavior I describe (I am using gamespot.com as a test).

When I then press "Forward" to go back to the new page, then press "Back", I get the behavior your screenshot is showing.

I've tried it with a few other example pages, and get the same effects,

Yes, I see that as well. In 1b0a049 I've been less specific about what the browsers do, so I hope that means we don't have to worry about this.

wbamberg · 2025-08-29T13:12:11Z

Thanks for the review, Chris! I'm out until Tuesday but will look when I get back.

gilbertococchi · 2025-08-29T14:49:54Z

Thanks everyone for your support for the WebSocket guide, sorry for late replying but I had been busy with lots of conflicting priorities these weeks.

I will review this first thing next week and will review the changes.

Thank you so much again for helping developers awareness about this opportunity for their sites when using WebSocket!

Josh-Cena · 2025-09-02T01:28:17Z

Looks like this may fix #29029, though I'm not sure if completely.

Co-authored-by: Chris Mills <[email protected]>

tunetheweb

Mostly LGTM but have made some comments.

Note that I'm far from a WebSockets expert, but I do know a fair bit about bfcache.

tunetheweb · 2025-09-02T17:28:37Z

files/en-us/web/api/websockets_api/writing_websocket_client_applications/index.md

+
+The back/forward cache, or {{glossary("bfcache")}}, enables much faster back and forward navigation between pages that the user has recently visited. It does this by storing a complete snapshot of the page, including the JavaScript heap.
+
+The browser pauses and then resumes JavaScript execution when a page is added to or loaded from the bfcache. This means that, depending on what the page is doing, it's not always safe for the browser to use the bfcache for the page. If the browser determines that it is not safe, then the page will not be added to the bfcache, and then the user will not get the performance benefit that it can bring.


I would say "restored from the bfcache" rather than "loaded from the bfcache".

The bfcache isn't a "cache" really. More a preservation and restore of a page's state.

tunetheweb · 2025-09-02T17:33:42Z

files/en-us/web/api/websockets_api/writing_websocket_client_applications/index.md

+Different browsers use different criteria for adding a page to the bfcache. If a page has an active WebSocket connection, then:

-### Text data format
+- Firefox and Chrome will not add the page to the bfcache.
+- Safari will add the page to the bfcache, but will cancel any active network requests. This is likely to generate an error in your WebSocket connection, leading to the connection being closed.


I'd probably be less explicit here as this ius subject to change.

For example Chrome is looking into whether just using a WebSocket really should disqualify a page from the bfcache. Or if we can still pout it in the bfcache but evict it only if a message is received on that WebSocket connection.

So maybe just leave as "Different browsers use different criteria for adding a page to the bfcache. For maximum support of the bfcache it is recommended to close any WebSocket clients on the {{domxref("Window.pagehide_event", "pagehide")}} event..."

I chose slightly different wording, hope this is OK? -> 1b0a049

This also gets us out of puzzling over Safari's behavior here.

tunetheweb · 2025-09-02T17:44:21Z

files/en-us/web/api/websockets_api/writing_websocket_client_applications/index.md

+});
+```
+
+Conversely, by listening for the {{domxref("Window.pageshow_event", "pageshow")}} event, you can seamlessly start the connection again when the page is loaded, or when it is retrieved from the bfcache. So in our example we will add all the code to initialize our WebSocket and set up its event listeners inside the `pageshow` event handler:


pageshow is also fired on page load. So might wanna mention that (to avoid people opening a connection twice on pageload).

tunetheweb · 2025-09-02T17:50:29Z

I wouldn't be particularly averse to adding a note in https://developer.mozilla.org/en-US/docs/Web/API/WebSockets_API about bfcache-compatibility as well, if we think it merits more visibility than this.

I think this would still be good as a short note (perhaps linking to the relevant section in this guide?). This updated guide is good, but it feels a little out of the way.

wbamberg · 2025-09-09T03:39:14Z

I wouldn't be particularly averse to adding a note in https://developer.mozilla.org/en-US/docs/Web/API/WebSockets_API about bfcache-compatibility as well, if we think it merits more visibility than this.

I think this would still be good as a short note (perhaps linking to the relevant section in this guide?). This updated guide is good, but it feels a little out of the way.

-> 21592f9 (I deleted a previous note that I didn't think was particularly useful)

wbamberg · 2025-09-09T03:40:04Z

Thanks for the reviews, people! I think this is ready for another look.

tunetheweb · 2025-09-09T05:51:17Z

Looks great now. One nit on the page_show on page load to consider, but other than that LGTM.

…plications/index.md Co-authored-by: wbamberg <[email protected]>

chrisdavidmills

@wbamberg One minor language nitpick, which you can take or leave. Apart from that, LGTM!

files/en-us/web/api/websockets_api/writing_websocket_client_applications/index.md

…plications/index.md Co-authored-by: Chris Mills <[email protected]>

New WebSockets client guide

c8f0fa3

github-actions bot added Content:WebAPI Web API docs size/m [PR only] 51-500 LoC changed labels Aug 28, 2025

wbamberg marked this pull request as ready for review August 28, 2025 22:24

wbamberg requested a review from a team as a code owner August 28, 2025 22:24

wbamberg requested review from sideshowbarker and removed request for a team August 28, 2025 22:24

wbamberg added 2 commits August 28, 2025 15:43

Use fewer words

b4aa283

More about the example

4d16661

wbamberg requested review from chrisdavidmills and removed request for sideshowbarker August 28, 2025 23:03

wbamberg added 2 commits August 28, 2025 16:04

'Availaible in workers' is not for guides

e894765

typoe

0e2f4a0

chrisdavidmills requested changes Aug 29, 2025

View reviewed changes

Apply suggestions from code review

97fcbd5

Co-authored-by: Chris Mills <[email protected]>

tunetheweb reviewed Sep 2, 2025

View reviewed changes

wbamberg added 7 commits September 8, 2025 18:09

Add a note about protocols

9e4076f

fix conflicts

f6ac042

send() clarification

768ae6e

loaded->restored

ef5e49c

Link to notRestoredReasons article

86ebca5

Be a bit vaguer about browsers and bfcache criteria

1b0a049

Add a note about bfcache in the API overview page

21592f9

wbamberg requested a review from chrisdavidmills September 9, 2025 03:39

wbamberg requested a review from tunetheweb September 9, 2025 03:39

wbamberg and others added 2 commits September 8, 2025 23:13

More clarity about pageshow

f15529d

Update files/en-us/web/api/websockets_api/writing_websocket_client_ap…

0a3b9a1

…plications/index.md Co-authored-by: wbamberg <[email protected]>

chrisdavidmills approved these changes Sep 9, 2025

View reviewed changes

files/en-us/web/api/websockets_api/writing_websocket_client_applications/index.md Outdated Show resolved Hide resolved

wbamberg and others added 2 commits September 9, 2025 07:58

Update files/en-us/web/api/websockets_api/writing_websocket_client_ap…

f651367

…plications/index.md Co-authored-by: Chris Mills <[email protected]>

Merge branch 'main' into websockets-bfcache

13fab76

wbamberg mentioned this pull request Sep 9, 2025

Add WebSocket Connection best practices note to handle BFCache support #40188

Closed

wbamberg merged commit 20c3765 into mdn:main Sep 9, 2025
8 checks passed


		The back/forward cache, or {{glossary("bfcache")}}, enables much faster back and forward navigation between pages that the user has recently visited. It does this by storing a complete snapshot of the page, including the JavaScript heap.

		The browser pauses and then resumes JavaScript execution when a page is added to or loaded from the bfcache. This means that, depending on what the page is doing, it's not always safe for the browser to use the bfcache for the page. If the browser determines that it is not safe, then the page will not be added to the bfcache, and then the user will not get the performance benefit that it can bring.

	Conversely, by listening for the {{domxref("Window.pageshow_event", "pageshow")}} event, you can seamlessly start the connection again when the page is loaded, or when it is retrieved from the bfcache. So in our example we will add all the code to initialize our WebSocket and set up its event listeners inside the `pageshow` event handler:
	Conversely, by listening for the {{domxref("Window.pageshow_event", "pageshow")}} event, you can seamlessly start the connection again when the page is loaded or retrieved from the bfcache. In our example, we add all the code to initialize our WebSocket and set up its event listeners inside the `pageshow` event handler:

New WebSockets client guide #40931

New WebSockets client guide #40931

Uh oh!

Conversation

wbamberg commented Aug 28, 2025 • edited by Josh-Cena Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Aug 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chrisdavidmills left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

wbamberg Sep 3, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

wbamberg Sep 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

chrisdavidmills Sep 4, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

wbamberg commented Aug 29, 2025

Uh oh!

gilbertococchi commented Aug 29, 2025

Uh oh!

Josh-Cena commented Sep 2, 2025

Uh oh!

tunetheweb left a comment

wbamberg commented Aug 28, 2025 •

edited by Josh-Cena

Loading

github-actions bot commented Aug 28, 2025 •

edited

Loading

wbamberg Sep 3, 2025 •

edited

Loading

wbamberg Sep 2, 2025 •

edited

Loading

chrisdavidmills Sep 4, 2025 •

edited

Loading