sendBinaryData freezes on large (10 MB) payloads when simultaneously listening on the connection

[kustosz]

Hey,

I've just stumbled upon a very strange issue with the WS server. I have found this in my production app, but shaved it down to a small repro. This code:

{-# LANGUAGE BangPatterns #-}
{-# LANGUAGE ScopedTypeVariables #-}

import qualified Network.WebSockets    as WS
import qualified Data.ByteString.Char8 as BS
import Control.Concurrent              (forkIO)
import Data.Traversable
import Control.Monad (forever)

main = WS.runServer "127.0.0.1" 34345 $ \pending -> do
    conn <- WS.acceptRequest pending

    forkIO $ do
        (msg :: BS.ByteString) <- WS.receiveData conn
        return ()

    WS.sendBinaryData conn $ BS.pack "hello"
    let size = 10000000
        !d = BS.pack $ take size $ cycle "hello"
    putStrLn "sending"
    WS.sendBinaryData conn d
    putStrLn "sent"

just freezes on WS.sendBinaryData (i.e. right after printing sending). A few additional remarks:

• for the other side of the test, I'm using the following simple piece of JS pasted in Chrome console: a = new WebSocket("ws://127.0.0.1:34345"); a.binaryType = "arraybuffer"; a.addEventListener("message", function (e) { console.log(e.data.byteLength) });
• removing the WS.receiveData call fixes the problems and allows me to send even larger (up to 1GB) payloads without any trouble.
• decreasing the payload size from 10MB to 1MB fixes the problem and the code works just fine.
• this is reproducible on a macbook pro running on macOS Sierra. I have tested it on a linux (Gentoo) machine, where the code works fine.
• I have tried replacing the discouraged WS.runServer with Snap, that did not change anything.
• It seems like some kind of a deadlock – memory usage jumps to 267M and no CPU is used.

Have you ever seen similar behavior? Is there any more information I can provide for this issue?

[kustosz]

I don't know how this got closed, probably automatically after I've closed a PR in my own project mentioning this one. Just confirming the issue still stands.

[jaspervdj]

Oh, I was wondering why this was closed. I don't have a macbook so unfortunately I can't reproduce this right now, hopefully someone else steps up.

[domenkozar]

Would be great to have a test case, we run CI on macOS so it should reproduce.