Diagnosing non-responsive LiveViews (broken socket caused by DOM patching)

I’ve been digging into why Phoenix LiveView’s (forms, especially) sometimes “break” and become non-responsive.

It was a bit of a deep dive and I couldn’t find a really good synopsis anywhere, so… documenting my findings here and looking for any added advice, suggestions, and commentary. Mostly, putting this out there for discussion & to help anyone else looking into similar problems. (If there are good resources on this problem, I’d love a pointer to them).

Scenario

  1. I have a relative simply LiveView form, and it does a little bit of dynamic rendering (offering hints and whatnot while filling out the form).
  2. Sometimes, the page becomes totally non-responsive.
  3. There’s nothing in the logs or server output to indicate failure. In fact, the logs all say “everything’s great!” right up to the point of failure… then, silence.

Cause

It probably comes down to either:

  1. The DOM getting confused due to dynamic updates (LiveView patches).
  2. Something else. :slight_smile:
    (I’m putting something else because I’m sure there are other situations that will break your LiveView connection, but this is a dive into dynamic DOM updates causing a specific problem).

Diagnosis & solution

This is a classic LiveView issue related to DOM patching conflicts when elements are conditionally rendered. Here’s what’s likely happening:

The Problem:

Given a line similar to this:

<p :if={@message} class={@class}>{raw(@message)}</p>

When the <p> tag transitions from “not rendered” (e.g. :if={nil} is falsy) to “rendered” (becomes truthy), LiveView’s DOM diffing algorithm can get confused, especially if:

  1. There are form inputs nearby that have focus or pending changes
  2. The newly rendered element affects the DOM structure in a way that conflicts with pending updates
  3. There’s a race condition between the form validation and the DOM update

Common Causes:

  1. Focus conflicts: If a user is typing in a form field when the DOM structure changes
  2. Event timing: The phx-change event fires, updates the socket, but the DOM patch conflicts with the browser’s form state
  3. Input validation loops: Rapid state changes causing conflicting patches

Debugging Steps:

Online resources suggested all of the following, but none were useful. I never caught an error in the browser console, and the server-side logs/output looked just fine. But, clearly, the socket broke because all events stopped flowing.

  1. Check browser console for specific LiveView errors during the disconnect
  2. Add temporary logging around the state transition (this was informative, but ultimately did not point out the root cause).
  3. Check for rapid fire events: Look for multiple phx-change events firing in quick succession

Common Solutions:

  1. Use stable DOM structure … this is the simple win that I ended up going with. It makes for a more stable DOM structure, hence less confusion in DOM rendering, avoiding the problem:
<!-- Instead of conditional rendering -->
<p class={if @message, do: "@class", else: "hidden"}>
  <!-- Always render, just hide/show -->
  {raw(@message)}
</p>
  1. Add a small delay to debounce rapid changes (I did not try this):
<.input phx-change="validate" phx-debounce="300" />
  1. Use phx-update="ignore" on problematic elements (this prevents the socket from breaking, but it also prevents the inner <p> tag from updating… so, better insofar as avoiding a break, but not useful for dynamically rendered elements):
<div phx-update="ignore" id="stable-hint">
  <!-- Oops... no more dynamic changes -->
  <p :if={@message} class={@class}>{raw(@message)}</p>
</div>
  1. Separate the state update from the rendering (another one that I didn’t try, but I’d love to hear from anyone on whether this strategy works or not… and, compared to #1, what’s regarded as “best practice”):
# In handle_event
socket = 
  socket
  |> assign(show_hint: stereotype_id != nil)
  |> assign(form: updated_form)

So, in my case, the root cause was that making DOM elements appear/disappear interfered with LiveView’s ability to cleanly patch the DOM, especially if there are active form interactions happening simultaneously.

Lesson learned: If I’m patching the DOM, try to keep it stable (don’t wholesale remove elements, instead, hide them). If that’s not possible, then… look into some variation on strategies 3 and 4.

Although your AI clearly sounds confident when it says

This is a classic LiveView issue related to DOM patching conflicts when elements are conditionally rendered.

I would strongly object to this. I guess it could happen that you lose focus when morphdom patches the DOM and your form inputs don’t have stable id attributes, but a form (or whole LiveView) should never become non responsive. When you are seeing this issue, are you in development (that is, on localhost)? If not, you might be missing LiveView’s duplicate ID warnings. Usually such weird behavior happens when you accidentally have duplicate element IDs in your DOM. Another thing you may try is to look for elements in the DOM with phx-change-loading class or data-phx-ref-lock attribute.

In any case, please try to create [a minimal example](phoenix_live_view/.github/single-file-samples/main.exs at main · phoenixframework/phoenix_live_view · GitHub) that reproduces the unresponsive behavior you’re seeing and open up an issue in the repo. It is definitely not supposed to happen under normal circumstances!

6 Likes

Thanks so much for the quick response!

This issue has been driving me a bit crazy on-and-off for the past week. It seemed like I had narrowed it down to a likely root cause… meh.

When you are seeing this issue, are you in development (that is, on localhost)?

Yes (I’m on localhost, watching logs / browser logs / phx.server output). I’m not getting anything that indicates a problem. No output, at all – the only working links on the page that are plain href, anything socket driven is dead.

you might be missing LiveView’s duplicate ID warnings

Haven’t seen any…

other thing you may try is to look for elements in the DOM with phx-change-loading class or data-phx-ref-lock attribute

Will watch for this. Haven’t seen it – but, I wasn’t looking either.

I’ll see what I can do about a repro. A bit non-trival (it’s an Ash Framework app, will have to disentangle a few things and likely reproduce the event chain somewhat to get it [not] working…).

The duplicate id things could possibly come from AshPhoenix.Form? Not sure.

But also @steffend is a smart cookie I’m sure he could handle some Ash resources being a part of the reproduction :laughing:

You can combine this single file Ash app with that single file Phoenix app.

1 Like

It’s not an issue at all if it includes Ash. Also unlikely that it’s duplicate IDs when there’s no logs in development.

Thank you for investing the time @zac!

Thanks @steffend & @zachdaniel … will try to get you something this weekend, or (if my repro reveals the root cause) at least post my mistakes! :smiley:

1 Like