Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:96491
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Received-SPF: pass (pb1.pair.com: domain lerdorf.com designates 209.85.223.181 as permitted sender)
MIME-Version: 1.0
In-Reply-To: <CAPZMMAtTV8bQm69Dub2uRLDewQxxVKWjyHCfDv6ND+-dJy5+bg@mail.gmail.com>
References: <CAPZMMAsZHO+=XAFUWQsE5GNc8c16+fEZk_LTyn8y-X17QRFzhA@mail.gmail.com>
 <CACXBjuiKuYp_OCWVPnoXdnY38vC5We7-QhiN8qv8Wd60TDVD4Q@mail.gmail.com> <CAPZMMAtTV8bQm69Dub2uRLDewQxxVKWjyHCfDv6ND+-dJy5+bg@mail.gmail.com>
Date: Wed, 19 Oct 2016 07:37:40 -0700
Message-ID: <CACXBjujRh1S2cgcfQ=fu2NjUmbLdr5brqEGmnxuyMt-yOpohFg@mail.gmail.com>
To: =?UTF-8?B?SsOpcsOpbWllIEJvcmRpZXI=?= <jeremie.bordier@gmail.com>
Cc: PHP internals <internals@lists.php.net>
Content-Type: multipart/alternative; boundary=001a113f9772bc6186053f38c22b
Subject: Re: [PHP-DEV] Performance drops after some time, PHP7-FPM + Docker
From: rasmus@lerdorf.com (Rasmus Lerdorf)

--001a113f9772bc6186053f38c22b
Content-Type: text/plain; charset=UTF-8

>
> The output of the perf diff is quite poor I think, here's the mainline :
>     35.90%  +44.94%  php-fpm             [.] 0x0000000000042412
>     10.72%   -6.05%  libc-2.19.so        [.] 0x0000000000079030
>      9.71%   -9.34%  newrelic.so         [.] 0x0000000000030980
>      3.81%   -3.47%  [kernel.kallsyms]   [k] ipt_do_table
>      2.56%   -2.02%  [kernel.kallsyms]   [k] nf_iterate
>      2.32%   -1.99%  opcache.so          [.] 0x0000000000008cec
>      1.44%           [kernel.kallsyms]   [k] update_cfs_shares
>      1.42%           [kernel.kallsyms]   [k] __bpf_prog_run
>      1.28%           [vdso]              [.] __vdso_gettimeofday
>
> How could I provide more info ?
>

Yes, not much to go on there. But newrelic.so stands out to me. Can you try
removing newrelic and see if it still happens?

If it does, your next step is to look at the timing of a request. I usually
do that with something like this:

    sudo strace -Ff -ttt -T -p <pid> 2>&1 | tee strace_output_file

Do that on a fast process and on a slow process and let it run for a little
while to the point where you think you have captured similar requests in
both runs, then have a look at the timing between syscalls in both and see
if you can spot if it is an isolated delay in a certain part of the request
or if it is just an overall slowdown.

I would also run:

    vmstat 1

both before and after. This gives you a picture of the overall health of
your server. The si/so columns, for example, will tell you if you are
swapping. If something has eaten a ton of memory and your system has
started swapping more than before then that could also explain this.

-Rasmus

--001a113f9772bc6186053f38c22b--