fix panic in tracing example

Multiple calls to init will cause a panic because a tracing subscriber has already been set globally.

examples/tracing/src/lib.rs

@@ -10,8 +10,10 @@ fn test() {
     tracing::warn!(foo = "bar", "baz");
 }
 
-#[event(fetch)]
-async fn main(req: Request, _env: Env, _ctx: Context) -> Result<Response> {
+// Multiple calls to `init` will cause a panic as a tracing subscriber is already set.
+// So we use the `start` event to initialize our tracing subscriber when the worker starts.
+#[event(start)]
+fn start() {
     let fmt_layer = tracing_subscriber::fmt::layer()
         .json()
         .with_ansi(false) // Only partially supported across JavaScript runtimes
@@ -22,7 +24,10 @@ async fn main(req: Request, _env: Env, _ctx: Context) -> Result<Response> {
         .with(fmt_layer)
         .with(perf_layer)
         .init();
+}
 
+#[event(fetch)]
+async fn main(req: Request, _env: Env, _ctx: Context) -> Result<Response> {
     tracing::info!(request=?req, "Handling request");
     test();