Try to add OPcache

[joanhey]

We could try that: https://bugs.php.net/bug.php?id=65078

And if work without problem, create a bug request to add the ngx_php7 sapi.

[rryqszq4]

@joanhey Not startup, I think need a add request~

[rryqszq4]

The name is better with 'php-ngx' https://github.com/rryqszq4/ngx_php7/blob/master/src/php/impl/php_ngx.c#L263

[joanhey]

Had you tried to add "php-ngx" in https://github.com/php/php-src/blob/master/ext/opcache/ZendAccelerator.c#L2544 ??

Copile it to try with php-ngx.

[rryqszq4]

I am trying something else.

[joanhey]

https://github.com/unbit/uwsgi/blob/master/plugins/php/php_plugin.c

The uwsgi, perhaps they initialize the sapi in another way.

Also useful to take ideas.

[joanhey]

In nginx unit it's working without problems. https://github.com/nginx/unit/blob/master/src/nxt_php_sapi.c

They are using "cli-server" :roll_eyes:

We could try with that first.

[rryqszq4]

I think that "cli-server" maybe better.

[joanhey]

With "cli-server"

[joanhey]

I need to test a little more. With different files. This image is using the app.php loaded in the init_worker. So it is loaded only one time.

[joanhey]

I don't expect any exponencial gain. But all the includes outside of the init_worker will be fast now.

[rryqszq4]

@joanhey Test use opcache and not.

1. Result of use opcache

   ./wrk -H 'Host: 127.0.0.1' -H 'Accept: application/json,text/html;q=0.9,application/xhtml+xml;q=0.9,application/xml;q=0.8,*/*;q=0.7' -H 'Connection: keep-alive' --latency -d 15 -c 512 --timeout 8 -t 28 http://127.0.0.1/h2
   Running 15s test @ http://127.0.0.1/h2
   28 threads and 512 connections
   Thread Stats   Avg      Stdev     Max   +/- Stdev
   Latency   109.19ms  538.96ms   6.81s    97.50%
   Req/Sec     1.77k     1.66k   10.61k    81.66%
   Latency Distribution
    50%    2.47ms
    75%    3.53ms
    90%  200.55ms
    99%    3.21s 
   627719 requests in 15.10s, 81.39MB read
   Socket errors: connect 0, read 10, write 0, timeout 4
   Requests/sec:  41570.54
   Transfer/sec:      5.39MB

2. Result of not use opcache

   ./wrk -H 'Host: 127.0.0.1' -H 'Accept: application/json,text/html;q=0.9,application/xhtml+xml;q=0.9,application/xml;q=0.8,*/*;q=0.7' -H 'Connection: keep-alive' --latency -d 15 -c 512 --timeout 8 -t 28 http://127.0.0.1/h2
   Running 15s test @ http://127.0.0.1/h2
   28 threads and 512 connections
   Thread Stats   Avg      Stdev     Max   +/- Stdev
   Latency   508.66ms    1.27s    7.91s    89.15%
   Req/Sec   198.26    158.90     1.46k    61.65%
   Latency Distribution
    50%   49.07ms
    75%   52.63ms
    90%    2.09s 
    99%    5.82s 
   30416 requests in 15.10s, 3.94MB read
   Socket errors: connect 0, read 0, write 0, timeout 3
   Requests/sec:   2014.45
   Transfer/sec:    267.46KB

And cache is hit.

Finally, but still be wary of unexpected mistakes.

[joanhey]

The socket errors happen without opcache too, so I think it is for the 512 concurrent connections in that computer.

The numbers are impressive. What is the code of that test ??

[rryqszq4]

Very simple code~ and h2 for that.

location = /h1 {
        content_by_php '
                echo "hello, world!"
        ';
}

 location = /h2 {
        content_by_php '
                include "/ngx_php7/t/lib/hello.php";
        ';
}

[joanhey]

And the results of /h1 ??

[rryqszq4]

No, it's h2

[joanhey]

yes, but I want to compare /h1 vs /h2 (with opcache)

[rryqszq4]

OK, show you the result at tomorrow.

[rryqszq4]

@joanhey

3. Result h1 of use opcache

   ./wrk -H 'Host: 127.0.0.1' -H 'Accept: application/json,text/html;q=0.9,application/xhtml+xml;q=0.9,application/xml;q=0.8,*/*;q=0.7' -H 'Connection: keep-alive' --latency -d 15 -c 512 --timeout 8 -t 28 http://127.0.0.1/h1
   Running 15s test @ http://127.0.0.1/h1
   28 threads and 512 connections
   Thread Stats   Avg      Stdev     Max   +/- Stdev
   Latency    66.11ms  355.81ms   6.77s    98.54%
   Req/Sec     1.99k     2.11k   43.31k    84.85%
   Latency Distribution
    50%    2.37ms
    75%    3.31ms
    90%  170.52ms
    99%    1.05s 
   679166 requests in 15.10s, 88.06MB read
   Requests/sec:  44976.38
   Transfer/sec:      5.83MB

4. Result h1 of not use opcache

   ./wrk -H 'Host: 127.0.0.1' -H 'Accept: application/json,text/html;q=0.9,application/xhtml+xml;q=0.9,application/xml;q=0.8,*/*;q=0.7' -H 'Connection: keep-alive' --latency -d 15 -c 512 --timeout 8 -t 28 http://127.0.0.1/h1
   Running 15s test @ http://127.0.0.1/h1
   28 threads and 512 connections
   Thread Stats   Avg      Stdev     Max   +/- Stdev
   Latency    60.00ms  305.88ms   6.77s    98.38%
   Req/Sec     1.84k     1.86k   11.82k    86.27%
   Latency Distribution
    50%    2.32ms
    75%    3.71ms
    90%  167.79ms
    99%  779.61ms
   660201 requests in 15.10s, 85.60MB read
   Socket errors: connect 0, read 0, write 0, timeout 27
   Requests/sec:  43725.02
   Transfer/sec:      5.67MB

   

[joanhey]

Could you create a new tag, to tests in the benchmark. And check the results with opcache and the changes in the async mysql.

[rryqszq4]

Could you create a new tag, to tests in the benchmark. And check the results with opcache and the changes in the async mysql.

OK, I will release the version as soon as possible.

[joanhey]

Now the image with the hello world is not correct.

The php file is almost as fast as inline.

[rryqszq4]

Sure that the image was already legacy.