early subrequest

bjornar.ness · 2014-07-29T23:06:26+00:00

Hello! On Tue, Jul 29, 2014 at 11:59 AM, rvsw wrote: > Thank you for your prompt response. I didn't quite realize that UTF-8 regex > features were alr...

early subrequest

bjornar.ness

I am trying to hook in on a early phase (connection), where I have c and hc, but

before ngx_http_wait_request_handler is run and r is initialized.

What is the recommended way to accomplish this?

bjornar.ness

On Tuesday, July 29, 2014 5:06:26 PM UTC+2, Bjørnar Ness wrote:

I am trying to hook in on a early phase (connection), where I have c and hc, but
before ngx_http_wait_request_handler is run and r is initialized.

What is the recommended way to accomplish this?

I want to do a ngx_http_subrequest to an internal location (thats what I need ngx_http_request_t for)

agentzh

Hello!

On Tue, Jul 29, 2014 at 8:06 AM, Bjørnar Ness wrote:
> I am trying to hook in on a early phase (connection), where I have c and hc,
> but
> before ngx_http_wait_request_handler is run and r is initialized.
>

No, you can never safely do that in that (early) context and you
should never use subrequests in that context. Even if you hack the
nginx core to achieve that by forcing initializing ngx_http_request_t
prematurely, such patch is much less likely to get merged into the
upstream nginx core. This is the wrong way. Take a look at how
init_worker_by_lua and ngx.timer callback implementations introduce
"fake requests" to get most of the ngx_lua APIs work (but still
subrequests are too tricky to enable even with "fake requests").

Regards,
-agentzh

bjornar.ness

On Tuesday, July 29, 2014 7:16:04 PM UTC+2, agentzh wrote:

Hello!

On Tue, Jul 29, 2014 at 8:06 AM, Bjørnar Ness wrote:
> I am trying to hook in on a early phase (connection), where I have c and hc,
> but
> before ngx_http_wait_request_handler is run and r is initialized.
>

No, you can never safely do that in that (early) context and you
should never use subrequests in that context. Even if you hack the
nginx core to achieve that by forcing initializing ngx_http_request_t
prematurely, such patch is much less likely to get merged into the
upstream nginx core. This is the wrong way. Take a look at how
init_worker_by_lua and ngx.timer callback implementations introduce
"fake requests" to get most of the ngx_lua APIs work (but still
subrequests are too tricky to enable even with "fake requests").

Does that mean I will not be able to, without a unsafe hack, to read a
string from a internal location in connection phase, before ngx_http_request_t
is initialized?

agentzh

Hello!

On Tue, Jul 29, 2014 at 11:15 AM, Bjørnar Ness wrote:
>
> Does that mean I will not be able to, without a unsafe hack, to read a
> string from a internal location in connection phase, before
> ngx_http_request_t
> is initialized?
>

I cannot parse the thing "to read a string from a internal location".

Regards,
-agentzh

bjornar.ness

On Tuesday, July 29, 2014 10:08:46 PM UTC+2, agentzh wrote:

Hello!

On Tue, Jul 29, 2014 at 11:15 AM, Bjørnar Ness wrote:
>
> Does that mean I will not be able to, without a unsafe hack, to read a
> string from a internal location in connection phase, before
> ngx_http_request_t
> is initialized?
>

I cannot parse the thing "to read a string from a internal location".

Hehe.. sorry. What I want to accomplish is:

before request phase, (dont have r), make a internal request to a /location
and use the response (string) to set some values in the original connection.

--
Bjørnar

agentzh

Hello!

On Tue, Jul 29, 2014 at 1:31 PM, Bjørnar Ness wrote:
> Hehe.. sorry. What I want to accomplish is:
>
> before request phase, (dont have r), make a internal request to a /location
> and use the response (string) to set some values in the original connection.
>

There is no such term like "internal request" in NGINX. You really
mean a subrequest to an internal location, right? As I've said, even
with hacks, it is not safe. Because other NGINX C modules serving the
subrequests usually assume a fully-initialized ngx_http_request_t
struct.

The best approach is to use the "fake request" mechanism as in
init_worker_by_lua and ngx.timer.at(), as I've said. And use cosockets
in such context instead of using subrequests. It's too hard to safely
use subrequests because the target NGINX C modules are out of control
and even a "fake request" has limitations that are too expensive to
emulate. And it's not worth it at all.

Regards,
-agentzh