On Tue, 2014-09-16 at 13:38 -0700, Alex Oscherov wrote: > Oleg, > Here is the link for the successful run with classic HC. This is exactly > the same test as I ran previously with async HC > https://org62.my.salesforce.com/sfc/p/000000000062/a/300000009kGJ/3nxDrGpX7Mz72BRyDe2bUT7X0hLvx2g1egCRBCYgG_w= > > Regards, > Alex Oscherov >
I think I know the reason. Short story, this is a server side issue (IIS brokenness). Long story. The problem is caused by IIS responding 'out of sequence' (before the client is fully done transmitting request message content). Classic HttpClient simply cant handle out of sequence responses. Period. It fully writes out request message and only then starts reading response message. Purely coincidentally, in some cases where out sequence response can be fully buffered inside a TCP internal buffer this actually helps work the problem around. HttpAsyncClient on the other hand can deal with out of sequence responses. It stops transmitting request message immediately after receiving an out of sequence response but then shuts the connection as non-persistent (left in a inconsistent state by partially transmitted request) once done reading the response (which makes sense). The funny thing is that NTLM can work with persistent connections only given it involves multiple message exchanges. So, what IIS does is in my opinion plain stupid. Oleg > On Tue, Sep 16, 2014 at 12:22 PM, Oleg Kalnichevski <[email protected]> > wrote: > > > On 16/09/14 18:12 , Alex Oscherov wrote: > > > >> Oleg, > >> of course server could be a culprit here but my test shows that when I > >> replace async client with regular apache HttpClient the same test works > >> without any problems. Could you please recommend something here. > >> > >> Alex Oscherov > >> > >> > > If you can produce two complete wire / context logs of the sessions, one > > generated with the async HC and another one with the classic HC, I can take > > another look. Please do try to make requests and sessions as comparable as > > possible. > > > > Oleg > > > > PS: Please always respond to the mailing list > > > > --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
