CPS-chained subrequests with I/O interceptions no longer work in nginx 0.8.21 (Was Re: Custom event + timer regressions caused by the new release)

[Maxim Dounin]

Hello!

On Tue, Oct 27, 2009 at 06:42:06PM +0800, agentzh wrote:

> On Tue, Oct 27, 2009 at 5:41 PM, agentzh <agentzh at gmail.com> wrote:
> >
> > Sorry, the failures should be rephrased a bit:
> >
> > "All the regression tests for my "echo" module's echo_sleep directive
> > that is used in subrequests chained by "post_subrequest" callback
> > accepted by ngx_http_subrequest."
> >
> 
> Okay, more investigation shows that the real problem has nothing to do
> with "sleep", custom events, nor timers. It's the chained subrequest
> model with "I/O interceptions" that no longer works in nginx 0.8.21.
> 
> The problem is much easier to phrase in terms of the "echo" module's directives:
> 
>     location /main {
>         echo_location /foo;
>         echo_location /sub1;
>     }
>     location /sub1 {
>         proxy_pass 'http://127.0.0.1:$server_port/foo';
>     }
>     location /foo {
>         echo $echo_request_uri;
>     }
> 
> This is the minimal test case that I can produce. Basically, the
> "post_subrequest" handler for the /foo subrequest issued directly by
> /main issues a subrequest to /sub1. Then /sub1 in turn calls the
> standard proxy module's handler to do some I/O operation. Then the
> whole connection never terminates properly and hangs forever.
> 
> If no I/O operation is involved, then the whole request completes
> normally in 0.8.21, no matter it's "echo_sleep" or "proxy_pass" that
> do the actual I/O interception.
> 
> The "echo_location" directive and its async friend are defined in the
> following .c file (for total 130+ lines of code only):
> 
>     http://github.com/agentzh/echo-nginx-module/blob/master/src/location.c
> 
> Could anyone give me a handle? :)

Could you please try to produce reduced code that triggers the issue?

I don't really have much time to investigate your module code, 
but I believe the problem is that you use subrequests and recall 
content handlers in mostly arbitrary way.  E.g. it looks like you 
call content handler again from post_subrequest handler.  This 
isn't supposed to work - they are normally called in quite a 
different contexts and have very different expectations.  Probably 
some changes in 0.8.21 rendered your code to be completely 
invalid.

Maxim Dounin