2014-06-26 12:34 GMT+09:00 Eric Covener <[email protected]>:
[cut]
> I think this is only a performance concern.

I think many things and yes, I realize maybe performance can be
worse than before in both socket readable case.

Before:
poll (event MPM)
pollset_remove
poll (proxy_wstunnel)
read/write (1st socket)
read/write (2nd socket)
pollset_add

After:
poll (event MPM)
pollset_remove
read/write (1st socket)
pollset_add
poll (event MPM)
pollset_remove
read/write (2nd socket)
pollset_add

I intend to do either following:

1) revert this.
I also doubt if r1601943 and r1605307 is useful.

2) make event MPM return all woken pollfd to wstunnel
This is not very hard work, but additional code complexity.
I don't think this is worth adding.

Reply via email to