A panic in the worker leaves it unable to process any subsequent requests for a period of time #166
Description
When a Rust-based worker panics, it leaves the worker dangling and unable to process subsequent requests from the same client (client A), which manifests itself in consistent 500s returned by Cloudflare. Requests from another client (client B) work fine, which suggests the sticky routing mechanism causes the requests sent by the client A to be routed to a dangling non-functional instance of the worker.
I created a minimal test case in this repository with the exact steps to reproduce.
Here's the relevant worker exhibiting this behaviour.
use worker::*;
#[event(fetch)]
pub async fn main(req: Request, env: Env, _context: Context) -> Result<Response> {
let router = Router::new();
router
.get("/helloworld", |_, _ctx| Response::ok("Hello World!"))
.get("/crashtheworker", |_, _ctx| panic!("CRASH"))
.run(req, env)
.await
}I am not sure if this is a bug in workers-rs – perhaps not – but hopefully it can be routed to the appropriate team.
Thank you!