I am in the context of a web application, where each request is assigned a unique correlation ID.
I am running in a wasm environment with the wasm32-unknown-unknown target. One request is always served by one thread, and the entire environment is torn down afterwards.
I would like to register a panic handler that if a request panics, it also logs this request ID.
This has proven to be difficult, as anything that has to go into the set_hook method needs the 'static lifetime constraint, which a request ID obviously doesn't have.
I would like code along the following lines to compile
// Assume we have a request here from somewhere.
let request = get_request_from_framework();
// This is at the start of the request
panic::set_hook(Box::new(|info| {
let request_id = request.get_request_id();
// Log panic messages here with request_id here
}));
I have a few potential approaches. I am not sure which one is best, or if there are any approaches that I am missing.
As I know my environment is torn down after each request, one way to get a String moved into the 'static lifetime to leak it is like this
let request_id = uuid::Uuid::new_v4().to_string();
let request_id: &'static str = Box::leak(request_id.into_boxed_str());
request_id
This will work in practice, as the request id is theoretically 'static (as after the request is served, the application is closed) - however it has the disadvantage that if I ever move this code into a non-wasm environment, we'll end up leaking memory pretty quickly.
As I know that each request is served by one thread, I could stuff the request id into a ThreadLocal, and read from that ThreadLocal on panics.
pub fn create_request_id() -> &'static str {
let request_id = uuid::Uuid::new_v4().to_string();
CURRENT_REQUEST_ID.with(|current_request_id| {
*current_request_id.borrow_mut() = request_id;
});
}
thread_local! {
pub static CURRENT_REQUEST_ID: RefCell<String> = RefCell::new(uuid::Uuid::new_v4().to_string());
}
// And then inside the panic handler get the request_id with something like
let request_id = CURRENT_REQUEST_ID.with(|current_request_id| {
let current_request_id = current_request_id.try_borrow();
match current_request_id {
Ok(current_request_id) => current_request_id.clone(),
Err(err) => "Unknown".to_string(),
}
});
This seems like the "best" solution I can come up with. However I'm not sure what the perf. implications are of initializing a ThreadLocal on each request is, particularly because we panic extremely rarely, so I'd hate to pay a big cost up-front for something I almost never use.
I experimented with the catch_unwind API, as that seemed like a good choice. I would then wrap the handling of each request with catch_unwind. However it seems like wasm32-unknown-unknown currently doesn't respect catch_unwind
What is the best solution here? Is there any way to get something that's heap-allocated into a Rust panic hook that I'm not aware of?
As per your example, you could move the id into the clusure:
// Assume we have a request here from somewhere.
let request = get_request_from_framework();
let request_id = request.get_request_id();
// This is at the start of the request
panic::set_hook(Box::new(move |info| {
let panic_message = format!("Request {} failed", request_id);
// Log panic messages here with request_id here
}));
Playground
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With