On 2/15/2012 10:00 AM, William A. Rowe Jr. wrote:
When any one of these were incorrectly implemented by a network stack driver, DisableWin32AcceptEx (and now AcceptFilter none) was necessary to work around the broken driver. That feature has always been a gross hack around what should "just work".
It does seem odd that so many drivers out there on the market have incorrectly implemented these mandatory features incorrectly since these days many have been certified by MS, and that they only seem to affect httpd. As far as Win32DisableAcceptEx, hack or not it worked great. What I have now to keep the server answering on https is a bigger hack, scanning error log every 30 minutes and if new AcceptEx error found, signaling a graceful restart. How did "timing issues" not affect the old hack, and why not use the old "known to work" hack for at least AcceptFilter none, up until some better fix is found? If that fix is found and cannot be implemented till 2.next, then 2.next it is.
Sorry to prod, but this will affect adoption of 2.4 on Windows. Gregg
