Rust 1.78 introduced a panic in our application #41
Comments
|
cc @SFBdragon and #37, would you be able to help investigate this? |
|
Also, @elpiel, could you detail how to reproduce this with the repository you linked? |
|
|
Although it might not be related, I think it's good to link these issues and repos as well. They mention a nightly bug: |
|
I'm willing to investigate but I can't say for sure when I'll be able to. Hopefully within a few days. At this point I'm very tempted to write a "safe" memory allocator to help comprehensively detect allocation issues at runtime, given the (relatively small, I imagine!) number of allocation bugs I've diagnosed in the Rust WASM ecosystem from bug reports. In the meantime, I'm curious what happens if you substitute |
|
We are also seeing this behaviour internally; our WASM modules are dying at random, likely after some allocation of a few dozen kilobytes. I'll try to collect more data and report back. The initial issue we commented upon was rustwasm/wasm-pack#1389, but I suspect that this is the root cause issue? |
|
@philpax Thanks for letting us know. Given this seems to be effecting quite a few people, I'll try to make time to dive into this today, then 👍 |
|
I wrote a quick test WASM application to determine if it was a fragmentation issue, but I can't reproduce the issue on 1.78.0: [package]
name = "dlmalloc_test"
version = "0.1.0"
edition = "2021"
[lib]
crate-type = ["cdylib"]
[dependencies]
wasm-bindgen = "0.2"use wasm_bindgen::prelude::*;
static mut LAST_BYTES: Vec<Vec<u8>> = Vec::new();
#[wasm_bindgen]
pub fn allocate(bytes: usize) -> usize {
unsafe {
if LAST_BYTES.len() > 10 {
LAST_BYTES.remove(0);
}
LAST_BYTES.push(vec![0; bytes]);
LAST_BYTES.iter().map(|b| b.len()).sum()
}
}<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8" />
<title>dlmalloc test</title>
<script type="module">
import init, { allocate } from "./pkg/dlmalloc_test.js";
async function run() {
await init();
setInterval(() => {
let bytesNow = Math.floor(Math.random() * 1000000);
let bytesActive = allocate(bytesNow);
let output = document.getElementById("output");
let msg = `Total: ${bytesActive}b | Now: ${bytesNow}b`;
output.textContent = msg;
console.log(msg);
}, 20);
}
run();
</script>
</head>
<body>
<p id="output"></p>
</body>
</html>This runs indefinitely without crashing. I was hoping this would catch it so that I could alter the asserts to print the values, but it looks like it'll take a bit more work to reproduce! This is definitely using dlmalloc and has the assert available: |
|
Update: I don't think this is an issue with Error and stack traceI suspect we just have a bad deallocation on our hands; this wouldn't be the first time rustwasm/wasm-bindgen#3801 The reason for this is that I'll poke around a bit more to try narrow down the cause. (I suspect |
Inspired by #41 but hasn't actually found any issues.
|
Thanks for taking a look at this @SFBdragon! Your conclusion, a buggy alloc/dealloc in |
Update 2:For others also hunting down the bug, we are indeed looking for a badly sized deallocation. I logged all the allocations/deallocations/reallocations: That size is very consistent. (Note that this likely isn't the same exact allocation |
|
Thank you very much for spend time on this bug! I will follow up on the other issue in wasm-pack then. |
|
For anyone landing on this issue, it seems that in the latest revision on
https://github.com/rustwasm/wasm-bindgen/blob/0.2.92/CHANGELOG.md#fixed-1 OR
https://github.com/rustwasm/wasm-bindgen/blob/0.2.92/CHANGELOG.md#0287 |
The default allocator (dlmalloc-rs) caused panic in workspaces with very large number of files and bumping wasm-bindgen to latest didn't help. See the issue: alexcrichton/dlmalloc-rs#41
First of all, thank you very much for working on this allocator (it seems that it's still the only one feature-rich enough for wasm in Rust)!
After a recent upgrade to Rust 1.78 our application started getting a panic message:
I've tried re-compiling the wasm module on 1.76 and 1.77 and did not get the panic message.
I believe this could be related to recent changes in the Rust compiler which are causing this misbehavior:
https://blog.rust-lang.org/2024/05/02/Rust-1.78.0.html#asserting-unsafe-preconditions
https://blog.rust-lang.org/2024/05/02/Rust-1.78.0.html#deterministic-realignment
Honestly, I don't feel competent enough to dig into the
dlmalloccode and debug our wasm module but I'll be happy to get some guidance or help with debugging this issue.PS: Our application is open-source and the wasm build can be found here: https://github.com/Stremio/stremio-core-web
The text was updated successfully, but these errors were encountered: