Handling of queued clients and timeout

[v6ops]

Hi,

First thanks so much for sharing your code. There really are very few usable examples of multithreaded libevent that can be used as a tutorial.

I see something that I perceive as an issue. Correct me if I'm wrong.

I start a server with 2 threads on a server with 2 CPU.
I connect 3 independent telnet clients a,b,c close in time to each other to create congestion.

client a connects and echos immediately. There are messages from thread [10] on the server
client b connects and echos immediately. There are messages from thread [14] on the server
client c connects immediately (process is listening) but does not echo (event is not triggering).
I'd expect this as there are #define CONNECTION_BACKLOG 8 on the TCP socket.

I now leave client b to time out after SOCKET_READ_TIMEOUT_SECONDS 100.
Client a continues working normally. Client c is never serviced.

Now what I think is abnormal. Client b times out expires. Client c times out simultaneously and never gets serviced.

Is this expected behaviour, a bug, or a limitation in libevent?

Shouldn't the code be using something like evconnlistener_new_bind rather than listening and binding independently of libevent?

I can debug further on request if you're still maintaining this code.

best regards.