Comment 16 for bug 777672

Brian,

I wanted to mention another failure scenario that seems relevant to this issue.

We've been experimenting against this test daemon https://gist.github.com/1158479 to simulate a situation where connections were being accepted but no clients are receiving any response.

From my perusal of the code it seems like the problem is that server_failure_limit is zero'd upon a successful connection, my feeling is that it should only be cleared upon a successful response (ie. at the end of io_read()).

Thoughts?

-Matt