[pbs-devel] [PATCH backup 5/7] proxy: implement 'reload-certificate' command

[Wolfgang Bumiller]

> On 05/12/2021 9:42 AM Dietmar Maurer <dietmar at proxmox.com> wrote:
> 
>  
> Stupid questzioon, but why cant we do:
> 
> diff --git a/src/bin/proxmox-backup-proxy.rs b/src/bin/proxmox-backup-proxy.rs
> index fc773459..29298a22 100644
> --- a/src/bin/proxmox-backup-proxy.rs
> +++ b/src/bin/proxmox-backup-proxy.rs
> @@ -223,7 +223,6 @@ async fn accept_connection(
>      // Note that these must not be moved out/modified directly, they get pinned in the loop and
>      // "rearmed" after waking up:
>      let mut reload_tls = notify_tls_cert_reload.notified();
> -    let mut accept = listener.accept();
>  
>      loop {
>          let sock;
> @@ -231,7 +230,9 @@ async fn accept_connection(
>          // normally we'd use `tokio::pin!()` but we need this to happen outside the loop and we
>          // need to be able to "rearm" the futures:
>          let reload_tls_pin = unsafe { Pin::new_unchecked(&mut reload_tls) };
> -        let accept_pin = unsafe { Pin::new_unchecked(&mut accept) };
> +        //let accept_pin = unsafe { Pin::new_unchecked(&mut accept) };
> +        let accept = listener.accept();
> +

For the `accept` call this can even work, but not for the notify one
(won't compile), which is a bit weird, since looking up the Notify code,
it *does* have code in the Drop handler to forward wakeups in case it
gets dropped before being received, however, both need to have correct
code to handle this somehow, and I felt like dropping and requerying is
less "nice" and allows for more "weird" errors (similar to the
DuplexStream issue) happening, and I didn't feel comfortable leaving
more room for our main client accept loop to end up in an unexplained
"hanging" state somehow.
(I have trust issues sometimes...)
But sure, we could do this for accept.
(With the patch set I just sent out that would just need the Box::pin()
calls removed)

I wish there was some nice form of a `select_loop!`-like helper...