Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Runaway GC in Safari\Webkit on refresh without disconnect() #398

Open
WalrusSoup opened this issue Aug 2, 2024 · 4 comments
Open

Runaway GC in Safari\Webkit on refresh without disconnect() #398

WalrusSoup opened this issue Aug 2, 2024 · 4 comments

Comments

@WalrusSoup
Copy link

WalrusSoup commented Aug 2, 2024

Running into an issue where a simple pubnub chat causes Safari to eat up 90% CPU when any channels have been joined. I've isolated it to opening a connection to channels. As soon as I call connect, it will have this issue after refreshing. It wont happen every refresh, but it happens 90% of the time.

Safari Version: Version 17.5 (19618.2.12.11.6)

Prior to refreshing, a lot of errors pop up in the console (~10 times):
Fetch API cannot load [x] due to access control checks.

This occurs even in a simple boilerplate page with nothing on it. This is from a vanilla vite project compilation, with an empty page:

Chat.init({
  publishKey: import.meta.env.VITE_PUBNUB_PUB_KEY,
  subscribeKey: import.meta.env.VITE_PUBNUB_SUB_KEY,
  userId: "1111111",
  storeUserActivityTimestamps: true,
  }).then((chat) => {
  const channel = await chat.getChannel("50")
  // connect or join
  channel.join((msg) => {
    console.log(msg);
  });
});

I do not see the errors Fetch API cannot load [x] due to access control checks when reloading via the developer console.

muh_cpu.mp4

If you force the browser to disconnect before reloading, we don't see the same behavior:

window.onbeforeunload = () => {
  console.log('unloading!!!!', window.chatRef);
  window.chatRef.sdk.disconnect();
}

On the development side, since we're in a Webkit view, i cannot seem to fix it for a reload. But i can at least prevent it from happening during hot module reloads:

import.meta.hot.on('vite:beforeFullReload', () => {
  if(window.chat) {
    console.log('!!!!UNLOADING CHAT!!!!');
    window.chat.sdk.disconnect();
    console.log('!!!!SDK DISCONNECTED!!!!');
  }
});

However the webkit environment does not fire the events onbeforeunload etc when refreshing which means this bug is basically stuck.

On reload, it appears that the timers that were scheduled are simply running forever. I set up a setTimeout() to simply disconnect Pubnub and set the object to undefined and the timers still hang out forever.

If I load, then reload the window and call destroy() on the current instance:

  setTimeout(() => {
    console.log('Calling Destroy()');
    window.chatRef.sdk.destroy();
  }, 5000)

When we compare snapshots we see that Safari seems to be hanging onto the SDK and throwing partial GCs hundreds of times:

Screenshot 2024-08-02 at 9 43 24 AM

Likewise, when calling destroy() on the SDK underneath, we still hang onto memory because there is no destroy() function on the ChatSDK. I can manually clean up both and THAT appears to fix the issue. However, this does not work when reloading. This must be called before we reload the window.

  setTimeout(() => {
    console.log('Calling Destroy()');
    window.chatRef.sdk.destroy();
    delete window.chatRef;
  }, 5000)

Perhaps I am missing something here?

@parfeon
Copy link
Contributor

parfeon commented Aug 2, 2024

@WalrusSoup thank you for sharing information about this issue.
It looks like the WKWebView way of handling JS context is the reason for what is happening. There are two ways to solve it:

  1. Initialize new WKWebView instance: this will force JS contexts to be destroyed
  2. The one which you mentioned: window.onbeforeunload - this can be done differently:
    a. The user handles this case and destroys the current SDK client / Chat SDK instance
    b. If the web version of JS SDK used, it can add onbeforeunload handler and destroy itself.

For 2.b. I will need to speak with the team about whether it will affect anyone if we will destroy by default.

@WalrusSoup
Copy link
Author

WKWebView does not seem to reliably trigger onbeforeunload. Should this be a bug report to webkit then since it's so easily and reliably recreated?

I managed to get it functional by hitting it from all sides in our app - but jeeze. This doesn't feel great.

if (import.meta.hot && !window.importHook) {
  import.meta.hot.on('vite:beforeFullReload', () => {
    if (window.chat) {
      // @ts-ignore
      window.chat.sdk.destroy();
    }
  });
  window.importHook = true;
}

reload via hotkey:

if (!window.refreshHook) {
  document.addEventListener('keydown', async (e) => {
    e = e || window.event;
    if (e.keyCode == 116) {
      window.chat.sdk.destroy();
      window.location.reload();
    }
  });
}

context menu needs to be replaced with one that can hook reload, so we disabled it.

document.addEventListener('contextmenu', function (e) {
    e.preventDefault();
});

@fwebdev
Copy link

fwebdev commented Oct 24, 2024

Is there any progress on this?

Our Customers also run into this Issue. The Regression was introduced in Version 8.2.7.
When I look at the Diff there is not much that changed. How could that cause the Error?

We had to rollback because of Customer Complaints.

@tance77
Copy link

tance77 commented Nov 13, 2024

@parfeon Any updates on this?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

4 participants