Ok, quite clear. ) Just thought that in case of keepalive connection to
FastCGI upstream we can process more than 1 request using one connect.

Unexpected FCGI_END_REQUEST is received in following case:
1) Request 1 sent to backend, results received and sent to client
2) Request 2 sent to backend by nginx and is in queue because php is busy
after fastcgi_finish_request() call
3) php finishes extra work after fastcgi_finish_request() call and sends
FCGI_END_REQUEST
4) Nginx expects STDERR or STDOUT of request 2 but receives FCGI_END_REQUEST
of request 1.

If this situation is normal by design I will just not use keep_conn.

Kind regards,
Dmitry Saprykin


On 4 July 2014 17:23, Maxim Dounin <mdounin@mdounin.ru> wrote:

> Hello!
>
> On Fri, Jul 04, 2014 at 05:18:52PM +0400, Dmitry Saprykin wrote:
>
> > Hello,
> >
> > This changeset adds support for FastCGI FCGI_END_REQUEST record type.
> > Now nginx does not process this type of FastCGI record.
> > In case of usage php fastcgi upstream which finishes FastCGI
> > requests before end of script using fastcgi_finish_request() call
> > it leads to "upstream sent unexpected FastCGI record: 3 while reading
> > response header from upstream" error messages and 502 for clients.
> >
> > Changeset parses FCGI_END_REQUEST FastCGI records and ignores it
> > if keep_conn is enabled and record has type FCGI_REQUEST_COMPLETE.
>
> What makes you think that this is something to be fixed in nginx?
> The FCGI_END_REQUEST record is clearly unexpected if there are no
> requests in flight. You may want to focus on fixing the problem
> in php instead.
>
> --
> Maxim Dounin
> http://nginx.org/
>
> _______________________________________________
> nginx-devel mailing list
> nginx-devel@nginx.org
> http://mailman.nginx.org/mailman/listinfo/nginx-devel
>
_______________________________________________
nginx-devel mailing list
nginx-devel@nginx.org
http://mailman.nginx.org/mailman/listinfo/nginx-devel