As a developer, you've probably at least heard something about CI - Continuous integration. And if you haven't - you better fix it ASAP, because that's something awesome to have on your skill list and can get extremely helpful in your everyday work. This post will focus on CI for PHP devs, and specifically, on CI implementation from Gitlab. I will suppose you know the basics of Git, PHP, PHPUnit, Docker and unix shell. Intended audience - intermediate PHP devs.
Adding something to your workflow must serve a purpose. In this case the goal is to automate routine tasks and achieve better quality control. Even a basic PHP project IMO needs the following:
* linter) checks (cannot merge changes that are invalid on the syntax level)
* Code style checks
* Unit and integration tests
All of those can be just run eventually, of course. But I prefer an automated CI approach even in my personal projects because it leads to a higher level of discipline, you simply can't avoid following a set of rules that you've developed. Also, it reduces a risk of releasing a bug or regression, thus improving quality.
Gitlab is as generous as giving you their CI for free, even for your private repos. At this point it is starting to look as advertising, therefore a quick comparison table for Gitlab, Github, Bitbucket. AFAIK, Github does not have a built-in solution, instead it is easily integrated with third parties, of which Travis CI seems to be the most popular - I will therefore mention Travis here.

Public repositories (OSS projects). All 3 providers have a free offer for the open-source community!

| Provider | Limits |
|---|---|
| Gitlab | 2,000 CI pipeline minutes per group per month, shared runners |
| Travis | Apparently unlimited |
| Bitbucket| 50 min/month, max 5 users, File storage <= 1Gb/month |

Private repositories

| Provider | Price | Limits |
|---|---|---|
| Gitlab | Free | 2,000 CI pipeline minutes per group per month, shared runners |
| Travis | $69/month | Unlimited builds, 1 job at a time |
| Bitbucket| Free | 50 min/month, max 5 users, File storage <= 1Gb/month |

Getting started

I made a small project based on Laravel framework and called it "ci-showcase". I work in Linux environment, and the commands I use in the examples, are for linux shell. They should be pretty much the same on Mac and nearly the same on Windows though.

composer create-project laravel/laravel ci-showcase

Next, I went to gitlab website and created a new public project: https://gitlab.com/crocodile2u/ci-showcase. Cloned the repo and copied all files and folders from the newly created project - the the new git repo. In the root folder, I placed a .gitignore file:

.idea
vendor
.env

Then the .env file:

APP_ENV=development

Then I generated the application encryption key: php artisan key:generate, and then I wanted to verify that the primary setup works as expected: ./vendor/bin/phpunit, which produced the output OK (2 tests, 2 assertions). Nice, time to commit this: git commit && git push

At this point, we don't yet have any CI, let's do something about it!

Adding .gitlab-ci.yml

Everyone going to implement CI with Gitlab, is strongly encouraged to bookmark this page: https://docs.gitlab.com/ee/ci/README.html. I will simply provide a short introduction course here plus a bit of boilerplate code to get you started easier.
First QA check that we're going to add is PHP syntax check. PHP has a built-in linter, which you can invoke like this: php -l my-file.php. This is what we're going to use. Because the php -l command doesn't support multiple files as arguments, I've written a small wrapper shell script and saved it to ci/linter.sh:

#!/bin/sh
files=<code>sh ci/get-changed-php-files.sh | xargs</code>last_status=0
status=0
# Loop through changed PHP files and run php -l on each
for f in "$files" ; do message=<code>php -l $f</code> last_status="$?" if [ "$last_status" -ne "0" ]; then # Anything fails -> the whole thing fails echo "PHP Linter is not happy about $f: $message" status="$last_status" fi
done
if [ "$status" -ne "0" ]; then echo "PHP syntax validation failed!"
fi
exit $status

Most of the time, you don't actually want to check each and every PHP file that you have. Instead, it's better to check only those files that have been changed. The Gitlab pipeline runs on every push to the repository, and there is a way to know which PHP files have been changed. Here's a simple script, meet ci/get-changed-php-files.sh:

#!/bin/sh
# What's happening here?
#
# 1. We get names and statuses of files that differ in current branch from their state in origin/master.
# These come in form (multiline)
# 2. The output from git diff is filtered by unix grep utility, we only need files with names ending in .php
# 3. One more filter: filter *out* (grep -v) all lines starting with R or D.
# D means "deleted", R means "renamed"
# 4. The filtered status-name list is passed on to awk command, which is instructed to take only the 2nd part
# of every line, thus just the filename
git diff --name-status origin/master | grep '\.php$' | grep -v "^[RD]" | awk '{ print }'

These scripts can easily be tested in your local environment ( at least if you have a Linux machine, that is ;-) ).
Now, as we have our first check, we'll finally create our .gitlab-ci.yml. This is where your pipeline is declared using YAML notation:

# we're using this beautiful tool for our pipeline: https://github.com/jakzal/phpqa
image: jakzal/phpqa:alpine
# For this sample pipeline, we'll only have 1 stage, in real-world you would like to also add at least "deploy"
stages: - QA
linter:
stage: QA
# this is the main part: what is actually executed
script:	- sh ci/get-changed-php-files.sh | xargs sh ci/linter.sh

The first line is image: jakzal/phpqa:alpine and it's telling Gitlab that we want to run our pipeline using a PHP-QA utility by jakzal. It is a docker image containing PHP and a huge variety of QA-tools. We declare one stage - QA, and this stage by now has just a single job named linter. Every job can have it's own docker image, but we don't need that for the purpose of this tutorial. Our project reaches Step 2. Once I had pushed these changes, I immediately went to the project's CI/CD page. Aaaand.... the pipeline was already running! I clicked on the linter job and saw the following happy green output:

Running with gitlab-runner 11.9.0-rc2 (227934c0) on docker-auto-scale ed2dce3a
Using Docker executor with image jakzal/phpqa:alpine ...
Pulling docker image jakzal/phpqa:alpine ...
Using docker image sha256:12bab06185e59387a4bf9f6054e0de9e0d5394ef6400718332c272be8956218f for jakzal/phpqa:alpine ...
Running on runner-ed2dce3a-project-11318734-concurrent-0 via runner-ed2dce3a-srm-1552606379-07370f92...
Initialized empty Git repository in /builds/crocodile2u/ci-showcase/.git/
Fetching changes...
Created fresh repository.
From https://gitlab.com/crocodile2u/ci-showcase * [new branch] master -> origin/master * [new branch] step-1 -> origin/step-1 * [new branch] step-2 -> origin/step-2
Checking out 1651a4e3 as step-2...
Skipping Git submodules setup
$ sh ci/get-changed-php-files.sh | xargs sh ci/linter.sh
Job succeeded

It means that our pipeline was successfully created and run!

PHP Code Sniffer.

PHP Code Sniffer is a tool for keeping app of your PHP files in one uniform code style. It has a hell of customizations and settings, but here we will only perform simple check for compatibilty with PSR-2 standard. A good practice is to create a configuration XML file in your project. I will put it in the root folder. Code sniffer can use a few file names, of which I prefer phpcs.xml:

<?xml version="1.0"?>
/resources

I also will append another section to .gitlab-ci.yml:

code-style:	stage: QA	script:	# Variable $files will contain the list of PHP files that have changes	- files=<code>sh ci/get-changed-php-files.sh</code> # If this list is not empty, we execute the phpcs command on all of them - if [ ! -z "$files" ]; then echo $files | xargs phpcs; fi

Again, we check only those PHP files that differ from master branch, and pass their names to phpcs utility. That's it, Step 3 is finished! If you go to see the pipeline now, you will notice that linter and code-style jobs run in parallel.

Adding PHPUnit

Unit and integration tests are essential for a successful and maintaiable modern software project. In PHP world, PHPUnit is de facto standard for these purposes. The PHPQA docker image already has PHPUnit, but that's not enough. Our project is based on Laravel, which means it depends on a bunch of third-party libraries, Laravel itself being one of them. Those are installed into vendor folder with composer. You might have noticed that our .gitignore file has vendor folder as one of it entries, which means that it is not managed by the Version Control System. Some prefer their dependencies to be part of their Git repository, I prefer to have only the composer.json declarations in Git. Makes the repo much much smaller than the other way round, also makes it easy to avoid bloating your production builds with libraries only needed for development.
Composer is also included into PHPQA docker image, and we can enrich our .gitlab-ci.yml:

test:	stage: QA	cache:	key: dependencies-including-dev	paths: - vendor/	script:	- composer install	- ./vendor/bin/phpunit

PHPUnit requires some configuration, but in the very beginning we used composer create-project to create our project boilerplate.laravel/laravel package has a lot of things included in it, and phpunit.xml is also one of them. All I had to do was to add another line to it:

xml

APP_KEY enironment variable is essential for Laravel to run, so I generated a key with php artisan key:generate.
git commit & git push, and we have all three jobs on theQA stage!

Checking that our checks work

In this branch I intentionally added changes that should fail all three job in our pipeline, take a look at git diff. And we have this out from the pipeline stages:Linter:

$ ci/linter.sh
PHP Linter is not happy about app/User.php:
Parse error: syntax error, unexpected 'syntax' (T_STRING), expecting function (T_FUNCTION) or const (T_CONST) in app/User.php on line 11
Errors parsing app/User.php
PHP syntax validation failed!
ERROR: Job failed: exit code 255

**Code-style**:

$ if [ ! -z "$files" ]; then echo $files | xargs phpcs; fi
FILE: ...ilds/crocodile2u/ci-showcase/app/Http/Controllers/Controller.php
----------------------------------------------------------------------
FOUND 0 ERRORS AND 1 WARNING AFFECTING 1 LINE
---------------------------------------------------------------------- 13 | WARNING | Line exceeds 120 characters; contains 129 characters
----------------------------------------------------------------------
Time: 39ms; Memory: 6MB
ERROR: Job failed: exit code 123

**test**:

$ ./vendor/bin/phpunit
PHPUnit 7.5.6 by Sebastian Bergmann and contributors.
F. 2 / 2 (100%)
Time: 102 ms, Memory: 14.00 MB
There was 1 failure:
1) Tests\Unit\ExampleTest::testBasicTest
This test is now failing
Failed asserting that false is true.
/builds/crocodile2u/ci-showcase/tests/Unit/ExampleTest.php:17
FAILURES!
Tests: 2, Assertions: 2, Failures: 1.
ERROR: Job failed: exit code 1

Congratulations, our pipeline is running, and we now have much less chance of messing up the result of our work.

Conclusion

Now you know how to set up a basic QA pipeline for your PHP project. There's still a lot to learn. Pipeline is a powerful tool. For instance, it can make deployments to different environments for you. Or it can build docker images, store artifacts and more! Sounds cool? Then spend 5 minutes of your time and leave a comment, you can also tell me if there is a pipeline topic you would like to be covered in next posts.

INTRO

If you are working as a PHP software developer, there is an extremely high chance that all of your application, you’re currently working on, using frameworks of any kind.
PHP community developers of all levels worship frameworks since there are big historical and practical reasons for that.

Historical reasons

Since early PHP versions, developers were disreputable because not everybody considered PHP as a programming language, similar to JavaScript a couple of years ago. While strong type language existed decades ago, PHP continues to be soft type since now, only in version 7 basic types were introduced. There is also a matter of the fact that you can script in PHP without using a single object.

But that opened a space for frameworks to step in and introduce themselves as a tool or standard which will shape projects, give them right and order, introduce structure and rules.
And finally, they did. Frameworks are good examples of nice structures, using all available new features PHP offers with every version, enforcing some good practice, etc.

Practical reasons

The framework offers a lot of common problems already solved. They offer a nice ecosystem for other developers to contribute and plug their components. There is a lot of online resources for learning and stay updated about any particular framework. Also, what every framework community tries very hard, is to make setup and usage easy.

WHEN PHP FRAMEWORKS SUCKS

I recently had the opportunity to give a talk on a conference and one meetup about why PHP framework sometimes sucks. Sometimes things we see in framework tutorials does not seem to be very much aligned with some object-oriented standards we are striving to enforce, and with basic clean code guidelines. On the other hand, there is nothing wrong with using a framework, if you use it right.

This article is the first "pilot" article in this series. In every new blog in this series, we will go more in-depth about every specific topic I covered during my presentation.
I'm very excited to share this knowledge, as I saw many developers suffer from bonded-to-framework disease.

https://twitter.com/damnjan/status/1058306144458956800

I won’t spend much time here on any particular framework discussion. This series will be just a guide on how to unbind yourself from frameworks and use them as a tool, instead of being independent.

**Here is the link to the presentation slides.**

Now that the Thanksgiving and Black Friday are left behind, we're all back at our desks, some of us having PHPStorm open for the whole day. In this post, I'll say a few words on this beautiful IDE, PHPUnit and XDebug.
You know that unit tests are essential, don't you? So do the PHPStorm developers. This industry-standard level IDE has tons of capabilities for integrating test frameworks and debuggers into your project. Even if you use VMs or containers to run your development environment, chances are they got you covered!



I often see even experienced PHP programmers debugging their code with var_dump(), which is obviously not the best way to do it. If you see the code for the first time, if you work with legacy code - step-by-step interactive debugging is the way to go. Sometimes it can save you hours of old school var_dumping.

As of unit tests, I often hear that it's good enough to run tests from the terminal. I even know a guy who runs watch phpunit /path/to/test while developing: this way the test is run every 2 seconds, you switch to the terminal whenever you want to see the latest results and that's it. However, there are certain advantages in running tests from the IDE. First, it's super-handy to launch a test method, test class or a whole folder with tests, just by pressing a hotkey. Second, the test results appear right there, in PHPStorm, with failures and their stack traces, every entry clickable and takes you directly to the file:line where a nasty thing happened. I also find the ability to run a debugger for a unit test, extremely attractive. Test fails, you click on a trace entry, get to a problematic line, place a break point, re-run the test in debug mode - and there you go.

For all those integrations, you will first need to setup the PHP interpreter for the project: Configuring PHP Development Environment. You will find both local and remote interpreter setups. "Local" is the PHP that you have on your workstation, the host machine. "Remote" can be pretty much everything: SSH if your Dev environment runs on a shared sandbox for all developers, docker or docker-compose if you run it using docker containers.

Next step - creating PHPUnit configuration. Go toSettings -> Languages and Frameworks -> PHP -> Test Frameworks. Follow this guide, it has much more information which will be more up-to-date than this post.Don't forget to set Path Mappings for your remote environments! That is, you probably have your project in, say, $HOME/projects/cool-project, but inside a docker or on a remote host it might be located at /app or /var/www, then you have to let PHPStorm know about this.

Once you're done with PHPUnit setup, you can finally run your tests! The default shortcut on my Linux machine isCtrl+Shift+F10 (shortcuts are usually different on Mac though). Place a cursor inside a test method, press the shotcut: PHPStorm will launch PHPUnit withthat particular test method! When the cursor in a scope of test class but not inside a test method - the whole test class will be run. And, you also can select a whole folder with tests, in the project tree and run it, ain't that cool?

A small tip for the docker-compose lovers. When I first set PHPStorm integration with docker-compose and ran the tests, I was quite surprised (unpleasantly) to see that myphp-fpm service that I was connecting to, is gone after the test process is finished. Took me some time to figure out that it's PHPStorm's expected behavior. It stops the target service after it's done testing. A workaround I started to use is as follows: I just add another service calledphpunit which uses a php-fpm or php-cli image, and is not needed by anything except unit testing in PHPStorm.

Now to debugging.

Debugging is like being the detective in a crime movie where you are also the murderer. Filipe Fortes a.k.a. @fortes

Obviously, your PHP interpreter in development environment will need a debugger extension in order for you to debug interactively. PHPStorm support the two most widely used options: XDebug and Zend Debugger. When using docker I usually make a separate Dockerfile for development, using production image as base, then add development tools,XDebug being the most important. Honestly, I've never usedZend Debugger, so have little to tell about its' nuances.

Got an extension? Go to Debugging Ultimate Guide! Debugger settings in PHPStorm are atSettings -> Languages and Frameworks -> PHP -> Debug. Most of the time you don't need to change them.Again, a note for docker-compose users. There is an XDebug setting that allow debugger to resolve the client (PHPStorm) IP address:xdebug.remoteconnect_back_. That's a disappointment but those will not work, at least with a default docker-compose setup. Thing is, all containers in a compose stack are running behind a network proxy provided by docker-compose. That is, the REMOTE_ADDR for all the containers will always be the IP of proxy. A workaround:

* disablexdebug.remoteconnect_back_;
* add.user.ini to the application root folder with the following contents:xdebug.remotehost = 192.168.X.X_ (your machine's IP address in the LAN). It's generally a good idea to exclude.user.ini from VCS control.

As a conclusion: if you still usevardump()_ to debug, stop living in the stone age, upgrade your knowledge and become more productive! If you don't write unit tests, start doing it. If your managers say it's a waste of time, tell them that it's coding without tests that is a waste of time. And, if you find this post of any use, or have an opinion, or a question - please do comment!

What is a Daemon?

The term daemon was coined by the programmers of Project MAC at MIT. It is inspired on Maxwell's demon in charge of sorting molecules in the background. The UNIX systems adopted this terminology for daemon programs.

It also refers to a character from Greek mythology that performs the tasks for which the gods do not want to take. As stated in the "Reference System Administrator UNIX", in ancient Greece, the concept of "personal daemon" was, in part, comparable to the modern concept of "guardian angel." BSD family of operating systems use the image as a demon's logo.

Daemons are usually started at machine boot time. In the technical sense, a demon is considered a process that does not have a controlling terminal, and accordingly there is no user interface. Most often, the ancestor process of the deamon is init - process root on UNIX, although many daemons run from special rcd scripts started from a terminal console.

Richard Stevenson describes the following steps for writing daemons:
    . Resetting the file mode creation mask to 0 function umask(), to mask some bits of access rights from the starting process.
    . Cause fork() and finish the parent process. This is done so that if the process was launched as a group, the shell believes that the group finished at the same time, the child inherits the process group ID of the parent and gets its own process ID. This ensures that it will not become process group leader.
    . Create a new session by calling setsid(). The process becomes a leader of the new session, the leader of a new group of processes and loses the control of the terminal.
    . Make the root directory of the current working directory as the current directory will be mounted.
    . Close all file descriptors.
    . Make redirect descriptors 0,1 and 2 (STDIN, STDOUT and STDERR) to /dev/null or files /var/log/project_name.out because some standard library functions use these descriptors.
    . Record the pid (process ID number) in the pid-file: /var/run/projectname.pid.
    . Correctly process the signals and SigTerm SigHup: end with the destruction of all child processes and pid - files and / or re-configuration.

How to Create Daemons in PHP

To create demons in PHP you need to use the extensions pcntl and posix. To implement the fast communication withing daemon scripts it is recommended to use the extension libevent for asynchronous I/O.

Lets take a closer look at the code to start a daemon:

umask(0);
$pid = pcntl_fork(); 
if ($pid < 0) {
 print('fork failed');
 exit 1; 
}

After a fork, the execution of the program works as if there are two branches of the code, one for the parent process and the second for the child process. What distinguishes these two processes is the result value returned the fork() function call. The parent process ID receives the newly created process number and the child process receives a 0.

if ($pid > 0) { echo "daemon process started
";
 exit;
} 
 
$sid = posix_setsid();
if ($sid < 0) {
 exit 2;
}

chdir('/');
file_put_contents($pidFilename, getmypid() ); 
run_process();

The implementation of step 5 "to close all file descriptors" can be done in two ways. Well, closing all file descriptors is difficult to implement in PHP. You just need to open any file descriptors before fork(). Second, you can override the standard output to an error log file using init_set() or use buffering using ob_start() to a variable and store it in log file:

ob_start();
var_dump($some_object);
$content = ob_get_clean();
fwrite($fd_log, $content); 

Typically, ob_start() is the start of the daemon life cycle and ob_get_clean() and fwrite() calls are the end. However, you can directly override STDIN, STDOUT and STDERR:

ini_set('error_log', $logDir.'/error.log');
fclose(STDIN); 
fclose(STDOUT);
fclose(STDERR);
$STDIN = fopen('/dev/null', 'r'); 
$STDOUT = fopen($logDir.'/application.log', 'ab');
$STDERR = fopen($logDir.'/application.error.log', 'ab');

Now, our process is disconnected from the terminal and the standard output is redirected to a log file.

Handling Signals

Signal processing is carried out with the handlers that you can use either via the library pcntl (pcntl_signal_dispatch()), or by using libevent. In the first case, you must define a signal handler:

function sig_handler($signo)
{
 global $fd_log;
 switch ($signo) {
 case SIGTERM:
 fclose($fd_log); unlink($pidfile); exit;
 break;
 case SIGHUP:
 init_data(); break;
 default:
 }
}


pcntl_signal(SIGTERM, "sig_handler");
pcntl_signal(SIGHUP, "sig_handler");

Note that signals are only processed when the process is in an active mode. Signals received when the process is waiting for input or in sleep mode will not be processed. Use the wait function pcntl_signal_dispatch(). We can ignore the signal using flag SIG_IGN: pcntl_signal(SIGHUP, SIG_IGN); Or, if necessary, restore the signal handler using the flag SIG_DFL, which was previously installed by default: pcntl_signal(SIGHUP, SIG_DFL);

Asynchronous I/O with Libevent

In the case you use blocking input / output signal processing is not applied. It is recommended to use the library libevent which provides non-blocking as input / output, processing signals, and timers. Libevent library provides a simple mechanism to start the callback functions for events on file descriptor: Write, Read, Timeout, Signal.

Initially, you have to declare one or more events with an handler (callback function) and attach them to the basic context of the events:

$base = event_base_new();
$event = event_new();
$errno = 0;
$errstr = '';
$socket = stream_socket_server("tcp://$IP:$port", $errno, $errstr);
stream_set_blocking($socket, 0);
event_set($event, $socket, EV_READ | EV_PERSIST, 'onAccept', $base);

Function handlers 'onRead', 'onWrite', 'onError' must implement the processing logic. Data is written into the buffer, which is obtained in the non-blocking mode:

function onRead($buffer, $id)
{
 while($read = event_buffer_read($buffer, 256)) {
 var_dump($read);
 }
}

The main event loop runs with the function event_base_loop($base);. With a few lines of code, you can exit the handler only by calling: event_base_loobreak(); or after the specified time (timeout) event_loop_exit();.

Error handling deals with failure Events:

function onError($buffer, $error, $id)
{
 global $id, $buffers, $ctx_connections;
 event_buffer_disable($buffers[$id], EV_READ | EV_WRITE);
 event_buffer_free($buffers[$id]);
 fclose($ctx_connections[$id]);
 unset($buffers[$id], $ctx_connections[$id]);
}

It should be noted the following subtlety: Working with timers is only possible through the file descriptor. The example of official the documentation does not work. Here is an example of processing that runs at regular intervals.

$event2 = event_new();
$tmpfile = tmpfile();
event_set($event2, $tmpfile, 0, 'onTimer', $interval);
$res = event_base_set($event2, $base);
event_add($event2, 1000000 * $interval);

With this code we can have a working timer finishes only once. If we need a "permanent" Timer, using the function onTimer we need create a new event each time, and reassign it to process through a "period of time":

function onTimer($tmpfile, $flag, $interval)
{
 $global $base, $event2;

 if ($event2) {
 event_delete($event2);
 event_free($event2);
 }

 call_user_function(‘process_data’,$args);

 $event2 = event_new();
 event_set($event2, $tmpfile, 0, 'onTimer', $interval);
 $res = event_base_set($event2, $base);
 event_add($event2, 1000000 * $interval);
}

At the end of the daemon we must release all previously allocated resources:

event_delete($event);
event_free($event);
event_base_free($base);

event_base_set($event, $base);
event_add($event);

Also it should be noted that for the signal processing handler is set the flag EV_SIGNAL: event_set($event, SIGHUP, EV_SIGNAL, 'onSignal', $base);

If needed constant signal processing, it is necessary to set a flag EV_PERSIST. Here follows a handler for the event onAccept, which occurs when a new connection is a accepted on a file descriptor:

function onAccept($socket, $flag, $base) {
 global $id, $buffers, $ctx_connections;
 $id++;
 $connection = stream_socket_accept($socket);
 stream_set_blocking($connection, 0);
 $buffer = event_buffer_new($connection, 'onRead', NULL, 'onError', $id);
 event_buffer_base_set($buffer, $base);
 event_buffer_timeout_set($buffer, 30, 30);
 event_buffer_watermark_set($buffer, EV_READ, 0, 0xffffff); event_buffer_priority_set($buffer, 10); event_buffer_enable($buffer, EV_READ | EV_PERSIST); $ctx_connections[$id] = $connection;
 $buffers[$id] = $buffer;
}

Monitoring a Daemon

It is good practice to develop the application so that it was possible to monitor the daemon process. Key indicators for monitoring are the number of items processed / requests in the time interval, the speed of processing with queries, the average time to process a single request or downtime.

With the help of these metrics can be understood workload of our demon, and if it does not cope with the load it gets, you can run another process in parallel, or for running multiple child processes.

To determine these variables need to check these features at regular intervals, such as once per second. For example downtime is calculated as the difference between the measurement interval and total time daemon.

Typically downtime is determined as a percentage of a measurement interval. For example, if in one second were executed 10 cycles with a total processing time of 50ms, the time will be 950ms or 95%.

Query performance wile be 10rps (request per second). Average processing time of one request: the ratio of the total time spent on processing requests to the number of requests processed, will be 5ms.

These characteristics, as well as additional features such as memory stack size queue, number of transactions, the average time to access the database, and so on.

An external monitor can be obtain data through a TCP connection or unix socket, usually in the format of Nagios or zabbix, depending on the monitoring system. To do this, the demon should use an additional system port.

As mentioned above, if one worker process can not handle the load, usually we run in parallel multiple processes. Starting a parallel process should be done by the parent master process that uses fork() to launch a series of child processes.

Why not run processes using exec() or system()? Because, as a rule, you must have direct control over the master and child processes. In this case, we can handle it via interaction signals. If you use the exec command or system, then launch the initial interpreter, and it has already started processes that are not direct descendants of the parent process.

Also, there is a misconception that you can make a demon process through command nohup. Yes, it is possible to issue a command: nohup php mydaemon.php -master >> /var/log/daemon.log 2 >> /var/log/daemon.error.log &

But, in this case, would be difficult to perform log rotation, as nohup "captures" file descriptors for STDOUT / STDERR and release them only at the end of the command, which may overload of the process or the entire server. Overload demon process may affect the integrity of data processing and possibly cause partial loss of some data.

Starting a Daemon

Starting the daemon must happen either automatically at boot time, or with the help of a "boot script."

All startup scripts are usually in the directory /etc/rc.d. The startup script in the directory service is made /etc/init.d/ . Run command start service myapp or start group /etc/init.d/myapp depending on the type of OS.

Here is a sample script text:

#! /bin/sh
#
$appdir = /usr/share/myapp/app.php
$parms = --master –proc=8 --daemon
export $appdir
export $parms
if [ ! -x appdir ]; then
 exit 1
fi

if [ -x /etc/rc.d/init.d/functions ]; then
 . /etc/rc.d/init.d/functions
fi

RETVAL=0

start () {
 echo "Starting app"
 daemon /usr/bin/php $appdir $parms 
 RETVAL=$?
 [ $RETVAL -eq 0 ] && touch /var/lock/subsys/mydaemon
 echo
 return $RETVAL
}

stop () {
 echo -n "Stopping $prog: "
 killproc /usr/bin/fetchmail
 RETVAL=$?
 [ $RETVAL -eq 0 ] && rm -f /var/lock/subsys/mydaemon
 echo
 return $RETVAL
}

case in
 start)
 start
 ;;
 stop)
 stop
 ;;
 restart)
 stop
 start
 ;;
 status)
 status /usr/bin/mydaemon
 ;;
 *)
 echo "Usage: 

php
if (is_file('app.phar')) {
unlink('app.phar');
}
$phar = new Phar('app.phar', 0, 'app.phar');
$phar->compressFiles(Phar::GZ);
$phar->setSignatureAlgorithm (Phar::SHA1);
$files = array();
$files['bootstrap.php'] = './bootstrap.php';
$rd = new RecursiveIteratorIterator(new RecursiveDirectoryIterator('.'));
foreach($rd as $file){
if ($file->getFilename() != '..' && $file->getFilename() != '.' && $file->getFilename() != __FILE__) {
if ( $file->getPath() != './log'&& $file->getPath() != './script'&& $file->getPath() != '.')
$files[substr($file->getPath().DIRECTORY_SEPARATOR.$file->getFilename(),2)] =
$file->getPath().DIRECTORY_SEPARATOR.$file->getFilename();
}
}
if (isset($opt['version'])) {
$version = $opt['version'];
$file = "buildFromIterator(new ArrayIterator($files));
$phar->setStub($phar->createDefaultStub('bootstrap.php'));
$phar = null;
}

 {start|stop|restart|status}"
 ;;
 
RETVAL=$?
exit $RETVAL

Distributing Your PHP Daemon

To distribute a daemon it is better to pack it in a single phar archive module. The assembled module should include all the necessary PHP and .ini files.

Below is a sample build script:

#php app.phar
myDaemon version 0.1 Debug
usage:
--daemon – run as daemon
--debug – run in debug mode
--settings – print settings
--nofork – not run child processes
--check – check dependency modules
--master – run as master
--proc=[8] – run child processes

Additionally, it may be advisable to make a PEAR package as a standard unix-console utility that when run with no arguments prints its own usage instruction:
[NMD%%CODE%%]

Conclusion

Creating daemons in PHP it is not hard but to make them run correctly it is important to follow the steps described in this article.

Post a comment here if you have questions or comments on how to create daemon services in PHP.

For those of you not familiar with the concept of underclocking: it's the opposite of overclocking, that is, you don't speed up CPU but instead slow it down..

What for?

Ask the underclockers, I'm totally not sure. Actually, hanging around the Web these days leaves a feeling that nearly every website out there must have been underclocked, but most of the time it's about tons of unnecessary images, megabytes of javascript (of which hardly a hundred kilobytes gets actually executed), and all that. In this post I will, however, tell you about a server-side approach to underclocking, with a help of our good old friend - the MySQL Database Server.

Today I had a nice chat in my client's development telegram channel. The two other devs, R** and V**, were making a switch of the old image API app to a new MySQL server. A couple of days before that, we have discussed a plan, it was as dumb as possible, just as I like it. Super-simple clear steps that a five-year-old can make. Switch to readonly mode (stop uploads), dump DB, restore the dump on the new server, update database connection details, turn off readonly mode. What could possibly go wrong?

Nothing. Except that it did go wrong. The app that I'm talking about, is a really ancient piece of what is gently called "legacy". Once the app was back to normal again, we noticed a significant slowdown on every page that made use of images. Before that point, I never got to that app and/or its database. I logged in to the MySQL console, and started investigating, at the same point chatting with colleagues.

Me: Is that really important that the tables are MyISAM? It's 2018, you know.. There are dozens of queries in queue waiting for table-level locks.

R**: Are they MyISAM? Really?

Me: Yes.. Any objection against converting them to InnoDB? With the current state of the website, with all those tons of Gateway Timeouts, it's not going to make it worse if I do it right now..

    . minutes later:

Me: Nah, it didn't help a lot.. But, looking at the SHOW PROCESSLIST output, I see something weird. What, do you think, this query does? SELECT LAST_INSERT_ID() FROM images? 

R**: ehh... Gets you the last AUTO_INCREMENT id from images table?

Me: Let's play another good news bad news joke.. Good news: you're right, it gets you the last AUTO_INCREMENT id. Bad news: it's not for table, it's for the session. Worse news: this query gets you the last AUTO_INCREMENT id and does it exactly as many times as there are rows in the images table. how many are there?

R**: about 8mln. #@%&! It's sending 8mln rows on every image upload, through the network!

Me: Bingo! 8mln rows, with one and the same integer value in all of them.

R**: Ouch... Aaaand... Before today, it was not an issue. Because the database was on the same server as the application..

Me: Exactly, it used the loopback interface, and now it's using ehternet, which, apparently, doesn't have a super good bandwidth. We don't have a gigabit channel between servers, do we?

R**: No, it's 100 Mbit

Me: Are you fixing the query, BTW?

R**: yeah, man, deploying it...

Another 10 minutes later, problem is gone, performance is back to normal.

What conclusions can one make from this story?

I can think of two at least:
First: never underestimate legacy code. The ways it can move to bite you in the ass, are mysterious.
Second: if you're working with MySQL or another RDBMS, learn SQL, learn the specific SQL dialect you're using and learn how to trouble shoot issues. In this case, I did not need to look in the PHP code at all in order to help my fellow colleagues out. You can also generalize this principle as "you have to know the tools you're using".

Happy optimizing, folks! Comments appreciated!