On Wed, Apr 03, 2013 at 12:11:35AM +0200, Willy Tarreau wrote: > Hi Sander, > > On Tue, Apr 02, 2013 at 11:35:45PM +0200, Willy Tarreau wrote: > > Sounds like I broke the response forward state machine last evening > > then (the last 3 patches that went into 20130402), because all other > > patches are just cosmetic. I'm re-auditing the code now. > > OK I could reproduce using nginx only as well. I still don't know what > the issue is, but I can already confirm that reverting commit d655ffe8 > fixes the issue here.
And here is the fix. I indeed messed up one point. Everything was OK in the code... except that we need to loop one last time before finishing the transfer, in order to flush pending data. This loop is still ugly and I'd really like to attack it but it scares me and you now understand why :-/ I've pushed the fix as commit 2d43e18b, for those who want to pick it. Now I'm back finishing the ACL/sample merge. !hanks for reporting this! Willy
