Heroku Local with PHP on Mac OS X - php

Question

Currently I just use additional terminal tabs to manually start worker and clock processes in addition to an always-on apache proxying to php-fpm.
I tried heroku local when I started with heroku but its setup defeated me.
Now I want to give it another try.
I'm on High Sierra with homebrew php but have stuck with the mac os built in apache till now. It appears brew's apache may be a better choice but hopefully we'll find out if so in the following.
I realise from the answer provided here (by the top contributor to the buildpack) that the apache used by heroku local must be stopped at the time heroku local is started. There's a similar apparent quote from heroku support about this in an answer here.
I also note, my own discovery, that one should install the buildpack locally with * as in composer require-dev heroku/heroku-buildpack-php "*" to ensure the latest version.
Right now I get the following when I issue heroku local, using mac osx's built-in apache (I have it listening on port 8080 to serve my php dev environments, but stopped it for the purpose of this, with sudo apachectl stop).
m$ heroku local
[OKAY] Loaded ENV .env File as KEY=VALUE Format
12:21:03 PM horizon.1 | Horizon started successfully.
12:21:03 PM clock.1 | [2019-09-20 11:21:03] Calling scheduler
12:21:03 PM clock.1 | No scheduled commands are ready to run.
12:21:03 PM web.1 | DOCUMENT_ROOT changed to 'public/'
12:21:04 PM web.1 | 4 processes at 128MB memory limit.
12:21:04 PM web.1 | Starting php-fpm...
12:21:06 PM web.1 | Starting httpd...
12:21:06 PM web.1 | Application ready for connections on port 5000.
12:21:06 PM web.1 | [Fri Sep 20 12:21:06.155117 2019] [core:emerg] [pid 25867] (2)No such file or directory: AH00023: Couldn't create the mpm-accept mutex (file /private/var/run/mpm-accept-0.25867)
12:21:06 PM web.1 | (2)No such file or directory: could not create accept mutex
12:21:06 PM web.1 | AH00015: Unable to open logs
12:21:06 PM web.1 | Process exited unexpectedly: httpd
12:21:06 PM web.1 | Going down, terminating child processes...
[DONE] Killing all processes with signal SIGINT
12:21:06 PM horizon.1 | Shutting down...
12:21:06 PM clock.1 Exited with exit code SIGINT
12:21:06 PM web.1 Exited with exit code null
12:21:07 PM horizon.1 | [2019-09-20 11:21:06][1033] Processing: Laravel\Scout\Jobs\MakeSearchable
12:21:07 PM horizon.1 | [2019-09-20 11:21:06][1032] Processing: Laravel\Scout\Jobs\MakeSearchable
12:21:07 PM horizon.1 | [2019-09-20 11:21:06][1034] Processing: Laravel\Scout\Jobs\MakeSearchable
12:21:08 PM horizon.1 | [2019-09-20 11:21:06][1033] Processed: Laravel\Scout\Jobs\MakeSearchable
12:21:08 PM horizon.1 | [2019-09-20 11:21:06][1032] Processed: Laravel\Scout\Jobs\MakeSearchable
12:21:08 PM horizon.1 | [2019-09-20 11:21:06][1034] Processed: Laravel\Scout\Jobs\MakeSearchable
12:21:08 PM horizon.1 Exited Successfully
My mac os apache vhosts forward to /tmp/php72-fpm.sock. The permissions there are ok as apache in local browser reaches php-fpm fine.
I see that the actual error is (it scrolls off to the right in the code above due to no linebreak): Could not create the mpm-accept mutex. I now know there are a few different apache multi-processing modules, not sure if I need to know more about that.
But is that likely to just be permissions? I note mac osx apache needs sudo, as in sudo apachectl start and I am running heroku local without sudo. I could try sudo heroku local but would rather not until I know what it would do.
So in absence of more understanding here (which would be nice), I may trying installing brew apache (httpd24). It looks like heroku local will just call httpd, so, the one first in the path will be picked up.
Partial answer
I realised that when heroku local starts system apache, system apache will still have, of course, all of its current config. That means writing to sudo locations, error logs, vhosts which I have added. Of course, that's going to error without sudo. First step to fix the above was remove the listen directive from httpd.conf, which gave a new error:
(13)Permission denied: AH00091: httpd: could not open error log file /private/var/log/apache2/error_log.
So then I commented out errorlog to fix that one, which gave another:
(2)No such file or directory: AH02291: Cannot access directory '/usr/logs/' for main error log
2:17:25 PM web.1 | AH00014: Configuration check failed
2:17:25 PM web.1 | This program requires Apache 2.4.10 or newer with mod_proxy and mod_proxy_fcgi enabled; check your 'httpd' command.
Can see where this is going.
Basically, I would need to remove almost all of my mac osx apaches configuration for it not to error (when started without sudo).
So, let's consider brew apache instead... (below).
Of note, examining /tmp, each time I run heroku local I see the following files like heroku.xxxxx written as below of zero bytes. I note that the apache log files below can be found as Log directives in the default vhost include in the buildpack, in <buildpack>/conf/apache2/heroku.conf, hence their existence here.
mbp:tmp m$ ll
total 8
drwxrwxrwt# 16 root wheel 512B 20 Sep 08:54 ./
drwxr-xr-x 6 root wheel 192B 31 Dec 2017 ../
srwxrwxrwx 1 root wheel 0B 18 Sep 22:02 .dbfseventsd=
srwxrwxrwx 1 m wheel 0B 20 Sep 07:59 .s.PGSQL.5432=
-rw------- 1 m wheel 49B 20 Sep 07:59 .s.PGSQL.5432.lock
srwxr-xr-x 1 m wheel 0B 8 Sep 21:05 OSL_PIPE_501_SingleOfficeIPC_48607cb6b283d6f2d9ab5973acdb43c=
drwx------ 3 m wheel 96B 27 Aug 17:28 com.apple.launchd.9wuyYAXuof/
drwx------ 3 m wheel 96B 27 Aug 17:28 com.apple.launchd.v9lh33yWhI/
-rw-r--r-- 1 m wheel 0B 20 Sep 08:54 heroku.apache2_access.5000.log
-rw-r--r-- 1 m wheel 0B 20 Sep 08:54 heroku.apache2_error.5000.log
-rw-r--r-- 1 m wheel 0B 20 Sep 08:54 heroku.php-fpm.5000.log
-rw-r--r-- 1 m wheel 0B 20 Sep 08:54 heroku.php-fpm.5000.www.slowlog
-rw-r--r-- 1 m wheel 0B 20 Sep 08:54 heroku.php-fpm.www.5000.log
drwxr-xr-x 3 m wheel 96B 19 Sep 20:23 pear/
srwxrwxrwx 1 m staff 0B 19 Sep 22:11 php72-fpm.sock=
drwxr-xr-x 2 root wheel 64B 19 Sep 08:29 powerlog/
What else can I do to help the web process spin up?
Update - how to install brew apache for exclusive use of heroku local?
Perhaps what could be really useful is to know steps to install brew apache given that it'll be only used for heroku local. Is it a few simple brew commands and remove the listen directive? I'm cautious here as I would like ideally not to harm my current use of built in apache; these should be able to run alongside one another I think, just need to be reasonably sure before I do it.
Also
Homebrew now no longer supports options on formulae, so the brew install homebrew/apache/httpd24 --with-mpm-event as given at the link above
doesn't seem an option anymore. It is possible to do brew edit httpd24 to edit the formula directly; is that necessary? What are the correct install steps?

Answer 1

I'm posting a partial answer here which I'll update as I find out more rather than append this further information to the question itself.
I got it working with brew apache as follows:
brew install httpd
Edit /usr/local/etc/httpd/httpd.conf:
comment out line Listen 8080
comment in line LoadModule proxy_module lib/httpd/modules/mod_proxy.so
comment in line LoadModule proxy_fcgi_module lib/httpd/modules/mod_proxy_fcgi.so
comment in ServerName, I set it to ServerName localhost:5000
(this suppresses a complaint it will make - is that a correct value for ServerName?)
Edit /usr/local/etc/httpd/extra/httpd-ssl.conf as follows:
comment out Listen 8443
Don't do brew services start httpd, as you normally would for a brew service, that isn't relevant to us here.
(Now that we commented out the Listen directives, httpd won't respond anyway,
until heroku local starts it up and injects a Listen directive).
There is one area I don't fully understand yet.
When I run heroku local now, I can access my site at localhost:5000, but only the front page. Clicking on any subpage returns:
<!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML 2.0//EN">
<html><head>
<title>404 Not Found</title>
</head><body>
<h1>Not Found</h1>
<p>The requested URL was not found on this server.</p>
</body></html>
Perhaps we can fix that later. But let's consider php-fpm now.
Firstly, I had one easily fixable error, which is that heroku local was using mac osx's php-fpm (version 7.1 on my High Sierra), which I fixed by ensuring my php7.2 formula (which is keg only; I'll shortly move to php7.3) comes first in path.
Now which php-fpm returns /usr/local/opt/php#7.2/sbin/php-fpm and not /usr/sbin/php-fpm (mac osx built in). So, now we got it using the version we intend. Small but important fix.
However, we now must consider what happens when heroku local calls (i.e. starts up) php-fpm. Traditionally, I do brew services start php#7.2, which is always on (daemon). However, with Activity Monitor open, I see that heroku local spawns a new parent process and workers. Perhaps that's how it should be. It simply ignores the instance which is run by brew services, and creates its own.
The mystery arises though when I exit Heroku local by pressing Ctrl-C. Is that the correct way to exit it? If I do Ctrl-C, I get the command prompt back, but, looking at the spawned php-fpm processes in Activity Monitor, the php-fpm instances are not terminated. Then, if I issue heroku local a second time after stopping it the first, I get this:
5:44:37 PM web.1 | [20-Sep-2019 17:44:37] ERROR: An another FPM instance seems to already listen on /tmp/heroku.fcgi.5000.sock
5:44:37 PM web.1 | [20-Sep-2019 17:44:37] ERROR: FPM initialization failed
... which is quite understandable. That is fixable by manually terminating the process in Activity Monitor. However, I wonder what's happening here; what's the correct way to terminate heroku local?
To be specific, Ctrl-C quits a single php process which has a parent process of node, but does not quit five php-fpm processes, one of those five being the parent of the other four, and that one having a parent process of bash.
So, we've come a long way today; remaining issues seem to be; still to improve upon:
How to correctly terminate the php-fpm process (Ctrl-C doesn't do it).
Also if I have two terminals open, one which previously ran heroku local,
and I run it again in another window, the window which showed a regular command
prompt waiting for an instruction, suddenly jumps back to life, receives new
information, which is a but surprising, I wonder how that's working and if
that's considered normal.
Find out why apache isn't serving any page other than root as noted above.
It works!
Now we are motoring! The one important change I ommited in /usr/local/etc/httpd/httpd.conf was to comment in the line:
LoadModule rewrite_module lib/httpd/modules/mod_rewrite.so
I'm using laravel which serves all pages via index.php. That means you want to rewrite every page request that is not to index.php back to index php. Thus, the default .htaccess in laravel has the line RewriteRule ^ index.php [L] which does just that...
Now the only problem remaining that I'm currently aware of is that Ctrl-C doesn't terminate child processes.... but first I'm gonna set all my local apps to use fpm 7.3, as I saw a reference in the buildpack specifically to 7.3, so wondering if that might help...