Hung server connection with tokio-rustls

[haata]

I've been working on pycapnp (client) <-(TLS)-> capnproto-rust (server), which is basically working now except for (what I hope) is the final hurdle.

There was some sort of write IO change in tokio-rustls-0.9.1 -> tokio-rustls-0.9.2 that seems to block all capnproto wait requests after the first exchange (amusingly if I request a bunch of promises before waiting I can actually get all of them, but then next time I make a request it also hangs).

Here are 3 examples, the first one works, the 2nd one is the first version not working, the 3rd one is the latest version,
https://github.com/haata/capntls/tree/tokio-rustls-0.9.1

cargo run server localhost:12345 &
cargo run client localhost:12345
    Finished dev [unoptimized + debuginfo] target(s) in 0.24s
     Running `target/debug/capntls client 'localhost:12345'`
[email protected]:hello
[email protected]:helloa

vs.
https://github.com/haata/capntls/tree/tokio-rustls-0.9.2

cargo run server localhost:12345 &
cargo run client localhost:12345
    Finished dev [unoptimized + debuginfo] target(s) in 0.15s
     Running `target/debug/capntls client 'localhost:12345'`
[email protected]:hello
^C

and
https://github.com/haata/capntls/tree/tokio-rustls-0.10.1

cargo run server localhost:12345 &
cargo run client localhost:12345
    Finished dev [unoptimized + debuginfo] target(s) in 0.08s
     Running `target/debug/capntls client 'localhost:12345'`
[email protected]:hello
^C

The suggested "fix" from the tokio-rustls side is detailed here: quininer/tokio-rustls#37 unfortunately I'm not really sure how this change would fit into capnproto-rust (maybe somewhere in capnproto-futures?).

Anyways, I don't mind working on a patch though I might need some guidance on where to look.

[dwrensha]

The "suggested fix" seems to be "call flush()", and capnp-futures is already doing that:

capnproto-rust/capnp-futures/src/serialize.rs

Line 413 in 96f2f78

match writer.flush(){

So, at first glance, this looks to me like a bug somewhere other than capnproto-rust.

[haata]

Thanks for pointing this out!

Hmmm, I'll have keep hunting.

I traced the hung transaction to that particular commit in tokio-rustls so something is amiss around writing of data.

I haven't finished tracing all the way, but it seems like the server stop here:
https://github.com/capnproto/capnproto-rust/blob/master/capnp-futures/src/write_queue.rs#L196

[haata]

Turns out the issue was in tokio-rustls, quininer/tokio-rustls@fe8fafe

Thanks for your help @dwrensha