HTTP server connection stability issues

[malthe]

The server stays up, but is generally unable to handle:

weighttp -k -n 100000 -c 1000 http://127.0.0.1:50000

Typically, more than 85% of the requests do succeed, but various failure modes can happen that still causes a large number of requests to fail such as:

error: read() failed: Connection reset by peer (104)
error: connect() failed: Connection refused (111)

I did not look further into these, but it seems to be a socket handling issue.

[sylvanc]

After much investigation and twiddling, this seems to be working on Linux, FreeBSD and Windows now. On OSX there is still an outstanding issue: if a TCPListener accepts connections fast enough, there is some (as yet not identified) OS resource than be exhausted, leading to a KEVENT_READ not happening on the listening socket for ~30 seconds.

Under these circumstances, netstat -mm seems to show mbufs are not exhausted. The connecting socket is shows as SYN_SENT in lsof, and netstat -a -L shows the listening socket has a 0 qlen and a 0 incqlen. This is annoying.

Can you confirm that connections aren't being dropped for you, other than this OSX "lockup" issue?

[sylvanc]

We're looking into a Windows problem as well now.

[sylvanc]

Latest round should have this fixed.