Hi Again,
I hope I have been testing the right stuff thing; I changed my config to look
like so:
> http-send-name-header Host
> option forwardfor
> option http-server-close
> option httpclose
> option http-pretend-keepalive
> […]
> listen jiffy
> bind *:8181
> # reqidel ^Host:
It won't work correctly if I remove the reqidel Host line; the POST uploads
lead to:
• Status Code:500 could not parse request
Let me know if I should test different options.
Michael
Am 01.11.2012 um 15:03 schrieb Michael Seiferle <[email protected]>:
> Hi Willy,
> hi Cyril,
>
> (wow you are fast!) :-)
> I will compile and try it as soon as I can, think I'll need two or three
> hours though!
>
> Thanks!
> Michael
> Am 01.11.2012 um 14:53 schrieb Willy Tarreau <[email protected]>:
>
>> Hi Cyril,
>>
>> On Thu, Nov 01, 2012 at 02:25:17PM +0100, Cyril Bonté wrote:
>>> OK, I could reproduce the issue, which happens as soon as
>>> http-send-name-header is applied to a header provided in the request.
>>
>> This is interesting, because the same method is used when redispatching
>> a request to another server after a failed connection attempt. I think
>> that if I have not noticed it while testing, it is because all the
>> servers I configured had the same name length.
>>
>>> I'm sending a patch just after this mail (for haproxy 1.4), but I'm not
>>> sure it's the right way to fix it : after headers are removed, I
>>> decrease the buf->send_max value with the length of headers removed.
>>
>> This makes sense indeed.
>>
>>> This bug doesn't appear as soon as reqdel is called on the same headers.
>>
>> And it explains why Michael experienced different behaviours when
>> removing the header by hand or when using http-pretend-keep-alive
>> which leaves headers intact.
>>
>> Michael, could you please try Cyril's patch and confirm that it completely
>> fixes the issue you're experiencing ? If so, I'll merge it.
>>
>> Thanks,
>> Willy
>
