Empty buffer when null buffer

[benaadams]

Resolves #438

dbug: Microsoft.AspNet.Server.Kestrel[1]
      Connection id "1" started.
info: Microsoft.AspNet.Hosting.Internal.HostingEngine[1]
      Request starting HTTP/1.1 GET https://space.ageofascent.com/?&name=test-Ssl
dbug: Microsoft.AspNet.Server.Kestrel[6]
      Connection id "1" received FIN.
info: Microsoft.AspNet.Hosting.Internal.HostingEngine[2]
      Request finished in 340969ms 101
dbug: Microsoft.AspNet.Server.Kestrel[7]
      Connection id "1" sending FIN.
dbug: Microsoft.AspNet.Server.Kestrel[10]
      Connection id "1" disconnected.
dbug: Microsoft.AspNet.Server.Kestrel[8]
      Connection id "1" sent FIN with status "0".
dbug: Microsoft.AspNet.Server.Kestrel[2]
      Connection id "1" stopped.

vs the Exception thrown in #438; Websocket comms then works fine over Ssl; tested for 5mins as shown above Request finished in 340969ms 101

[benaadams]

w00p! https://twitter.com/ben_a_adams/status/671554633467346944

[benaadams]

Currently testing 1.2M application messages on the websockets per second; tested 1,070,637,277 messages so far, seems happy; will leave it running overnight.

[benaadams]

Fell over ticking into 10 billion messages; but that was an overflow on how I was recording, otherwise seems good.

[benaadams]

Hmm, over 10Bn messages on a restart - maybe I shouldn't have tweeted the url... :P

[jonburney]

Nothing like Twitter for free load testing!

[halter73]

Can we get a regression test?

[benaadams]

Can we get a regression test?

Can do a filter that's unhappy with null buffers? I have a love hate relationship with ssl...

[halter73]

Whatever works. Maybe you can unit test StreamSocketOutput with an outputStream that throws on null arrays.

[benaadams]

Was having trouble recreating the error state in a test; but narrowed down; its from a combination of WebSocketMiddleware.UpgradeHandshake and HttpsConnectionFilter/SslStream at the same time which is why existing tests didn't trigger it.

101Bn messages tested so far; some issues; need to look into if its NPC load or WebSocketMiddleware - but working mostly - error rate is low for that number of messages :)

[benaadams]

@halter73 can I pull in the WebSocketMiddeware for the test; and do a proper SSL+WebSocket test? (Build process-wise)

[halter73]

@benaadams I think that would be fine, but may be more complicated than what is necessary.

I was suggesting something simple like:

var streamSO = StreamSocketOutput(throwsOnNullWriteStream, null);
streamSO.Write(default(ArraySegment<byte>), true); // this shouldn't throw

Of course, if you want to add a functional test too, I'm not going to stop you!

[Tratcher]

@benaadams No, this repo can't depend on the WebSocket repo. You could write the test in the WebSocket repo though.

[benaadams]

@halter73 did as you suggested.

Test fails without the change; passes with it.

[halter73]

Awesome. I'll merge this today.

[benaadams]

Giving AppVeyor a force push (rebase)