Re: proxy_fcgi - force flush to client (Apache Users)

2018-02-02 12:20 GMT+01:00 Hajo Locke <Hajo.Locke@xxxxxx>:

Am 02.02.2018 um 07:05 schrieb Luca Toscano:

Hello Hajo,

2018-02-01 13:20 GMT+01:00 Hajo Locke <Hajo.Locke@xxxxxx>:

Hello Luca,

Am 01.02.2018 um 09:10 schrieb Hajo Locke:

Hello Luca,

Am 01.02.2018 um 04:46 schrieb Luca Toscano:

Hi Hajo,

2018-01-31 1:27 GMT-08:00 Hajo Locke <Hajo.Locke@xxxxxx>:

Hello List,

currently i compare features and behaviour of proxy_fcgi to classical methods like mod_fastcgi/mod_php.

mod_php/fastcgi have options to send every output from backend immediately to client. So it is possible to see progressing output in browser and not complete websiteoutput at once.

Here is an example script:
https://pastebin.com/4drpgBMq

if you ran this with php-cli or adjusted mod_php/mod_fastcgi you see progress in browser and numbers 0 1 2 appear one after another.
If you run this with proxy_fcgi you will see no progress, but complete output at once.

mod_proxy knows about worker parameter flushpackets, but the docs say this is in effect only for AJP. I can confirm that this and related options have no effect.
There are some workarounds posted in the web, but only one worked for me. If i add following line to the script, i also see a progress with proxy_fcgi in browser:

header('Content-Encoding: none');

Somebody knows a working workaround which works without scriptediting? some workarounds tell about using "SetEnv no-gzip 1". This was not working for me and iam not please to disable content-compression.
Is it planned to support >>flushpackets<< also to proxy_fcgi?

May be this is not important for typical website but some service/monitoring scripts.

The functionality is committed to trunk but never backported to 2.4.x because I was not sure about its importance, it looks like some users might benefit from it :)

The trunk patch is http://svn.apache.org/r1802040, it should apply to 2.4.x if you want to test it and give me some feedback.

Thanks!

I tried this and it works great. I see same behaviour as expected with other methods. I think some users might benefit from this. I saw some discussion related to this topic and people just ended up by ungainly workaround.
Great news!

Unfortunately i spoke too soon. I was too euphoric when reading your answer ;)
Behaviour is definitively more then expected, but it seems there is still a minimum-limit for the buffer to flush. I suppose this limit is 4096 bytes.
you can comprehend this with pastebinexample above.
Change line 2 from "$string_length = 14096;" to "$string_length = 1331;"
When calling this php-file you will see no progress. All output appears at once.
Change scriptline to "$string_length = 1332;", you will see at least 2 steps of output, because first step seems to break this 4096 bufferlimit. increasing $string_length more and more results in more steps of output.
So current mod_proxy_fcgi.c from svn with configured "flushpackets=On" seems to work exaktly like "flushpackets=auto iobuffersize=4096".
setting iobuffersize to lower numbers has no effect.
What do you think? Is there still a hard-coded limit or do i have a problem in my configuration?
I would be really glad, if you could take a look at this issue.

I am far from being an expert in PHP, but I added "ob_flush();" right before "flush()" in your script and the 1331 use case seems flushing correctly. Do you mind to check and let me know what do you get on your testing environment? As far as I can see in the mod_proxy_fcgi's code the iobuffersize variable is taken into account..

It seems that i was additional mocked by my browser. There is no need to edit this script, just using the right browser ;)
I think your new mod_proxy_fcgi.c did it and my testing was incorrect. I think we can go into weekend..

Full list of commits is: svn merge -c 1802040,1807876,1808014,1805490 ^/httpd/httpd/trunk .

mod_proxy_fcgi.c only patch: http://people.apache.org/~elukey/httpd_2.4.x-mod_proxy_fcgi-force_flush.patch

If you want to give it another round of test it would be really appreciated, in case everything is fine I'll propose it for backport to 2.4.x :)