Reset password flow in PHP

reset password

My email account is a skeleton key to anything online I’ve signed up for. If I forget a password, I can reset it. Implementing this feature for a web app takes just a few steps.

When users enter an incorrect password, I prompt them to reset it.

incorrect password warning

Clicking the reset link calls a “forgot password” back-end service.

$(document).on("click",".reset-pw-cta", function(){
	var email = $(this).attr("data");
	$.ajax({
		url:"/service-layer/user-service.php?method=forgotPw&email="+email,
		complete:function(response){
			console.log(response.responseText)
			window.showStatusMessage("A password reset email as been sent to " + email);
		}
	})
});

A token is created in our ‘password recovery’ database table. That token is related back to an account record.

password recovery database table

As a security practice, recovery tokens are deleted nightly by a cron job.

An email is then sent containing a “reset password” link embedded with the token. AWS SES and PHPMailer is used to send that message.

function forgotPw(){
	$email = $this->email;
	$row = $this->row;
	$number_of_rows = $this->number_of_rows;
	$conn = $this->connection;
	if($number_of_rows > 0){
		$this->emailFound = 1;
		$userid = $row['ID'];
		$this->userid = $userid;

		//create reset token
		$timestamp = time();
		$expire_date = time() + 24*60*60;
		$token_key = md5($timestamp.md5($email));
		$statement = $conn->prepare("INSERT INTO `passwordrecovery` (userid, token, expire_date) VALUES (:userid, :token, :expire_date)");
		$statement->bindParam(':userid', $userid);
		$statement->bindParam(':token', $token_key);
		$statement->bindParam(':expire_date', $expire_date);
		$statement->execute();

		//send email via amazon ses
		include 'send-email-service.php';	
		$SendEmailService = new SendEmailService();

		$reset_url = 'https://www.bjjtracker.com/reset-pw.php?token='.$token_key;
	        $subject = 'Reset your password.';
	        $body    = 'Click here to reset your password: <a href="'.$reset_url.'">'. $reset_url .'</a>';
	        $altBody = 'Click here to reset your password: ' . $reset_url;
	        $this->status = $SendEmailService -> sendEmail($subject, $body, $altBody, $email);


	}else{
		$this->emailFound = 0;
	}
}

That link navigates to a page with a “reset password” form.

reset password form

Upon submission the new password and embedded token are passed along to the server.

$(document).ready(function() {
    $(".reset-button").click(function(){
      var newPassword = $(".password-reset-input").val();
      if(newPassword.length < 1){
        var notifications = new UINotifications();
        notifications.showStatusMessage("Please don't leave that blank :( ");
        return;
      }
      var data = $(".resetpw-form").serialize();
      $.ajax({
        url: "/service-layer/user-service.php?method=resetPw&token=<?php echo $_GET['token']; ?>",
        method: "POST",
        data: data,
        complete: function(response){
          // console.log(response);
          window.location = "/";
        }
      });
    });
    $("input").keypress(function(e) {
      if(e.which == 13) {
        e.preventDefault();
          $(".reset-button").click();
      }
  });
  

});

The correct recovery record is selected by using the token value. That provides the user ID of the account that we want to update. The token should be deleted once the database is updated.

function resetPw(){
	$conn = $this->connection;
	$token = $_GET['token'];
	$password = $_POST['password'];
	$passwordHash = password_hash($password, PASSWORD_DEFAULT);
	$statement = $conn->prepare("SELECT * FROM `passwordrecovery` where token = ?");
	$statement->execute(array($token));
	$row = $statement->fetch(PDO::FETCH_ASSOC);
	$userid = $row['userid'];

	$update_statement = $conn->prepare("UPDATE `users` SET password = ? where ID = ?");
	$update_statement->execute(array($passwordHash, $userid));

	$delete_statement = $conn->prepare("DELETE FROM `passwordrecovery` where token = ?");
	$delete_statement->execute(array($token));
}

This is a secure and user-friendly workflow to allow users to reset their passwords.

Custom UI notifications

UI feedback alerts

Showing brief notifications to website visitors is an important UI/UX component. They’re useful for providing feedback. They can communicate success, failure, or warnings.

Don Norman (The Design of Everyday Things) mentions that “Feedback is essential, but not when it gets in the way of other things, including a calm and relaxing environment” and goes on to say “Feedback is essential, but it has to be done correctly”.

A common use-case is data validation. Specifically, when logging in or signing up. If the user enters an invalid email address, or wrong login credentials, we need to let them know. The built in browser alert() is clunky and unsophisticated. Plugins are bloated and over-engineered. I wrote some basic HTML, CSS, and JavaScript that gets the job done and looks great.

My code provides two versions of the alert. The first is a basic sticky bar that fades in and out at the top of the page.

example of alert message for an invalid email address

The other flashes in the middle of the screen. I call it “in-your-face” alerts and reserve them for positive success messages.

example of a flashing UI alert to provide positive feedback to users

The CSS adds styles for both versions. Both utilize ‘position: fixed’ to stay in a set location on the page. The “in-your-face” example uses a pulse animation to achieve its effect.

<!-- UI-notifications.css -->
<style>
body{
  margin: 0px;
}
.status-message{
  display: none;
  color: white;
  text-align: center;
  font-size: 16px;
  padding: 8px;
  border-top: 1px solid white;
  border-bottom: 1px solid white;
  position: fixed;
  width: 100%;
  top: 0px;
  padding: 28px 8px;
  background-color: #b12650;
  z-index: 1000;
}
.status-message-inner{
  margin: 0px;
}

.status-message-close{
  cursor: pointer;
  position: fixed;
  right: 10px;
}
.in-your-face{
  display: none;
  position: fixed;
  top: 45%;
  width: 100%;
  text-align: center;
  font-size: 48px;
  color: white;
  z-index: 2;
}
.in-your-face-inner{
    background: #005b96;
    width: 80%;
    margin: 0 auto;
    opacity: .85;
    padding: 10px;
}
@keyframes pulse{
  50%  {transform: scale(1.2);}

}
.pulse{
  animation: pulse 0.5s ease-in infinite;
}
</style>
<!-- end UI-notifications.css -->

The javascript relies on jQuery as a dependency. It is written as a class, with a constructor and two methods. Each method takes message text as a parameter.

class UINotifications {
	constructor() {
		window.jQuery || document.write('<script src="js/vendor/jquery-1.11.2.min.js"><\/script>');
		var statusMessageHtml = '<div class="status-message"><p class="status-message-inner"><span class="status-message-text">Welcome to My App</span><span class="status-message-close">X</span></p></div>';
		var inYourFaceHtml = '<div class="in-your-face pulse"><p class="in-your-face-inner"><span class="in-your-face-text">Great Job!</span></p></div>';

		$(document).on("click", ".status-message-close", function(){
			$(".status-message").fadeOut();
		});

		this.statusMessage = $("<div/>").html(statusMessageHtml);
		this.inYourFace = $("<div/>").html(inYourFaceHtml);
		
		$('body').prepend(this.inYourFace);
		$('body').prepend(this.statusMessage);

	}

 	showStatusMessage(message){
 		var notifications = this;
	  	var message = message || "Default Message"
	  	var statusMessageTimeout;
	  	
		if(notifications.statusMessage.find(".status-message").is(':visible')){
	     clearTimeout(statusMessageTimeout);
	    }

		notifications.statusMessage.find(".status-message .status-message-text").html(message);
		notifications.statusMessage.find(".status-message").fadeIn();
		
	    statusMessageTimeout = setTimeout(function(){
	       notifications.statusMessage.find(".status-message").fadeOut(); 
	    }, 5000)
		
	}
	showInYourFace(message, callback){
		var notifications = this;
		var inYourFaceTimeout;
		var inYourFaceRandoms = ["Good work!", "Hard work!", "Nice job!", "Hustle!"]

		var message = message || inYourFaceRandoms[Math.floor(Math.random()*inYourFaceRandoms.length)];;
		var callback = callback || function(){};

		if(notifications.inYourFace.find(".in-your-face").is(':visible')){
	     clearTimeout(inYourFaceTimeout);
	    }

		notifications.inYourFace.find(".in-your-face .in-your-face-text").html(message);
		notifications.inYourFace.find(".in-your-face").show();
		
	    inYourFaceTimeout = setTimeout(function(){
	       notifications.inYourFace.find(".in-your-face").fadeOut(function(){
	       	callback();
	       }); 

	    }, 1000)
	}
}

This is a simple and lightweight solution to showing web app visitors informative alerts without using a plugin. Please, checkout the code and use it in your next project.

You can find the code on GitHub.

Error establishing connection to database – WordPress solution

solutions for wordpress database errors

A crashed database is a problem I’ve encountered across multiple WordPress websites. When trying to load the site you’re faced with a dreaded “Error establishing a database connection” message. Restarting the DB service usually clears things up. But, sometimes it won’t restart at all – which is why I started automating nightly data dumps to an S3 bucket.

Recently, one particular site kept going down unusually often. I assumed it was happening due to low computing resources on the EC2 t3.micro instance. I decide to spin up a a new box with more RAM (t3.small) and migrate the entire WordPress setup.

Since I couldn’t be sure of what was causing the issue, I needed a way to monitor the health of my WordPress websites. I decided to write code that would periodically ping the site, and if it is down send an email alert and attempt to restart the database.

warning message when a website can't connect to the database

The first challenge was determining the status of the database. Even if it crashed, my site would still return a 200 OK response. I figured I could use cURL to get the homepage content, and then strip out any HTML tags to check the text output. If the text did match the error message, I could take further action.

Next, I needed to programmatically restart MySql. This is the command I run to do it manually: sudo service mariadb restart 

After doing some research, I found that I could use shell_exec() to run it from my PHP code. Unfortunately, Apache wouldn’t let the (non-password using) web server user execute that without special authorization. I moved that command to its own restart-db.sh file, and allowed my code to run it by adding this to the visudo file: apache ALL=NOPASSWD: /var/www/html/restart-db.sh

My visudo file was located at /usr/sbin/visudo. It is a tool found on most Linux systems to safely update the /etc/sudoers file, which is the configuration file for the sudo command. To edit this file, I don’t open it in vim like I would with other editable files. Instead, I run the file as its own command: sudo visudo. Once it is open, you can press the i key to enter “insert” mode. It is considered “safe” because it edits the sudoers file following a strict procedure.

edit the visduo file

I also needed to make the file executable by adjusting permissions: sudo chmod +x /var/www/html/restart-db.sh

Once those pieces were configured, my code would work:

<?php

$url = "https://www.antpace.com/blog/";
$curl_connection = curl_init();

curl_setopt($curl_connection, CURLOPT_URL, $url);

curl_setopt($curl_connection, CURLOPT_RETURNTRANSFER, true);
$curl_response = curl_exec($curl_connection);
$plain_text = strip_tags($curl_response);

if(strpos($plain_text, "Error establishing a database connection") !== false){
	echo "The DB is down.";
        
        //restart the database
        shell_exec('sudo /var/www/html/restart-db.sh');
        
        //send notification email
        include 'send-email.php';
        send_email();
}else{
	echo "The DB is healthy.";
}

?>

You can read more about how to send a notification email in another post that I wrote on this blog.

The contents of restart-db.sh looks like this:

#!/bin/bash

sudo service mariadb restart

Create the cron job

A cron job is a scheduled task in Linux that runs at set times. For my PHP code to effectively monitor the health of the database, it needs to run often. I decided to execute it every five minutes. Below are three shell commands to create a cron job.

The first creates the cron file for the root user:

sudo touch /var/spool/cron/root

The next appends my cron command to that file:

echo "*/5 * * * * sudo wget -q https://www.antpace.com/check-db-health.php" | sudo tee -a /var/spool/cron/root

And, the last sets the cron software to listen for that file:

sudo crontab /var/spool/cron/root

Alternatively, you can create, edit, and set the cron file directly by running sudo crontab -e . The contents of the cron file can be confirmed by running sudo crontab -l .

Pro-tip: If your WordPress site does continually crash, you probably do need to upgrade to an instance with more RAM. Also, consider using RDS for the database.

Update

I previously used the localhost loop back address in my cron file: */5 * * * * sudo wget -q 127.0.0.1/check-db-health.php. After setting up 301 redirects to prevent traffic from hitting my public DNS, that stopped working. It is more reliable to use an explicit domain name URL: */5 * * * * sudo wget -q https://www.antpace.com/check-db-health.php

Migrate a WordPress Site to AWS

Migrate a WordPress site to AWS

In a previous article I discussed launching a website on AWS. The project was framed as transferring a static site from another hosting provider. This post will extend that to migrating a dynamic WordPress site with existing content.

Install WordPress

After following the steps to launch your website to a new AWS EC2 instance, you’ll be able to connect via sFTP. I use FileZilla as my client. You’ll need the hostname (public DNS), username (ec2-user in this example), and key file for access. The latest version of WordPress can be downloaded from wordpress.org. Once connected to the server, I copy those files to the root web directory for my setup: /var/www/html

Make sure the wp-config.php file has the correct details (username, password) for your database. You should use the same database name from the previous hosting environment.

Data backup and import

It is crucial to be sure we don’t lose any data. I make a MySql dump of the current database and copy the entire wp-content folder to my local machine. I’m careful to not delete or cancel the old server until I am sure the new one is working identically.

Install phpMyAdmin

After configuring my EC2 instance, I install phpMyAdmin so that I can easily import the sql file.

sudo yum install php-mbstring -y
sudo systemctl restart httpd
sudo systemctl restart php-fpm
cd /var/www/html
wget https://www.phpmyadmin.net/downloads/phpMyAdmin-latest-all-languages.tar.gz
mkdir phpMyAdmin && tar -xvzf phpMyAdmin-latest-all-languages.tar.gz -C phpMyAdmin --strip-components 1
rm phpMyAdmin-latest-all-languages.tar.gz
sudo systemctl start mariadb

The above Linux commands installs the database management software on the root directory of the new web server. It is accessible from a browser via yourdomainname.com/phpMyAdmin. This tool is used to upload the data to the new environment.

phpMyAdmin import screen

Create the database and make sure the name matches what’s in wp-config.php from the last step. Now you’ll be able to upload your .sql file.

Next, I take the wp-content folder that I stored on my computer, and copy it over to the new remote. At this point, the site homepage will load correctly. You might notice other pages won’t resolve, and will produce a 404 “not found” response. That error has to do with certain Apache settings, and can be fixed by tweaking some options.

Server settings

With my setup, I encountered the above issue with page permalinks . WordPress relies on the .htaccess file to route pages/posts with their correct URL slugs. By default, this Apache setup does not allow its settings to be overridden by .htaccess directives. To fix this issue, the httpd.conf file needs to be edited. Mine was located in this directory: /etc/httpd/conf

You’ll need to find (or create) a section that corresponds to the default document root: <Directory “/var/www/html”></Directory>. In that block, they’ll be a AllowOverride command that is set to “None”. That needs to be changed to “All” for our configuration file to work.

apache config settings found in the HTTPD conf file

Final steps

After all the data and content has been transferred, do some smoke-testing. Try out as many pages and features as you can to make sure the new site is working as it should. Make sure you keep a back-up of everything some place secure (I use an S3 bucket). Once satisfied, you can switch your domain’s A records to point at the new box. Since the old and new servers will appear identical, I add a console.log(“new server”) to the header file. That allows me tell when the DNS update has finally resolved. Afterwards, I can safely cancel/decommission the old web hosting package.

Don’t forget to make sure SSL is setup!

Updates

AWS offers an entire suite of services to help businesses migrate. AWS Application Migration Service is a great choice to “simplify and expedite your migration while reducing costs”.

Upgrade PHP

In 2023, I used this blog post to stand-up a WordPress website. I was using a theme called Balasana. When I would try to set “Site Icon” (favicon) from the “customize” UI I would receive a message stating that “there has been an error cropping your image“. After a few Google searches, and also asking ChatGPT, the answer seemed to be that GD (a graphics library) was either not installed or not working properly. I played with that for almost an hour, but with no success. GD was installed, and so was ImageMagick (a back-up graphics library that WordPress falls back on).

The correct answer was that I needed to upgrade PHP. The AWS Linux 2 image comes with PHP 7.2. Upgrading to version 7.4 did the trick. I was able to make that happen, very painlessly, by following a blog post from Gregg Borodaty . The title of his post is “Amazon Linux 2: Upgrading from PHP 7.2 to PHP 7.4” (thanks Gregg).

Update

My recommendation, as of 2024, is to use a managed WordPress service. I wrote a post about using AWS Lightsail for that purpose: Website Redesign with WordPress Gutenberg via AWS Lightsail

 

Lazy Load Images and Assets on WordPress with IntersectionObserver

wordpress homepage design

I write online a lot. Adding articles to this blog serves to build a catalog of technical solutions for future reference. Updating the homepage has improved user experience and SEO. The new design displays the most recent articles as clickable cards, rather than listing the entire text of each one. The changes for this were added to index.php file, in the child-theme folder. The theme’s original code already used a While() loop to iterate through the post records. My modification removed the article content, and only kept the title and image:

<div class="doc-item-wrap">
	<?php
	while ( have_posts() ) {
		the_post();
		echo "<div class='doc-item'><a href='". get_the_permalink() ."'><img class='lazy' data-src='".get_the_post_thumbnail_url()."'><h2>" . get_the_title() . "</h2></a></div>";
	} ?>
</div> <!-- doc-item-wrap -->

I used custom CSS, leveraging Flexbox, to style and position the cards:

.doc-item-wrap{
    display: flex;
    flex-wrap: wrap;
    justify-content: center;
}
.doc-item{
    width: 30%;
    padding: 20px;
    border: 3px solid #f0503a;
    margin: 15px;
    background: black;
    flex-grow: 1;
    text-align: center;
}
.doc-item:hover{
    background-color: #34495e;
}
.doc-item p{
    margin: 0px;
    line-height: 40px;
    color: white;
}
.doc-item img{
    display: block;
    margin: 0 auto;
}
.doc-item h2{
    font-size: 22px;
    color: white;

}
@media(max-width: 1000px){
	.doc-item{
		width: 45%
	}
}
@media(max-width: 700px){
	.doc-item{
		width: 100%
	}
}

The media queries adjust the size of the cards (and how many are in a row), based on screen size.

Look and feel of the design

Card layout design is a common way to arrange blog content. It gives visitors a visual overview of what’s available. It also stops the homepage from duplicating content that’s already available on the individual post pages.

You can see this pattern throughout the digital world. Card layout translates well across screen sizes and devices. Since I put much effort into writing, making it organized was a priority. This implementation can be extended to add additional content (such as date, description, etc.) and features (share links, animations, expandability). And, it fits nicely with what WordPress already provides.

Lazy loaded images

Image content can often be the biggest drag to site speed. Lazy loading media defers rendering until it is needed. Since this blog’s homepage has an image for each post, this was essential.

While iterating through post records the image URL is assigned to a custom data-src attribute on the image tag, leaving the normal src blank. This assures the image is not immediately retrieved nor loaded. I wrote a JavaScript function to lazy load the images, relying on the IntersectionObserver API. The card’s image does not load until a user scrolls it into view. This improves the speed of the page, which has a positive effect on SEO and UX.

The code creates a IntersectionObserver object.  It observes each of the image elements, checking to see if they are within the browser viewport. Once the image elements come into view, it takes the image URL from the data-src attribute, and assigns it to the tag’s src – causing the image to load.

document.addEventListener("DOMContentLoaded", function() {
  var lazyImages = [].slice.call(document.querySelectorAll("img.lazy"));
  if ("IntersectionObserver" in window) {
    let lazyImageObserver = new IntersectionObserver(function(entries, observer) {
      entries.forEach(function(entry) {
        if (entry.isIntersecting) {
          let lazyImage = entry.target;
          lazyImage.src = lazyImage.dataset.src;
          // lazyImage.srcset = lazyImage.dataset.srcset;
          lazyImage.classList.remove("lazy");
          lazyImageObserver.unobserve(lazyImage);
        }
      });
    });

    lazyImages.forEach(function(lazyImage) {
      lazyImageObserver.observe(lazyImage);
    });
  } 
});

The original JS code was referenced from a web.dev article. Web.dev is a resource created by Google that provides guidance, best practices, and tools to help web developers build better web experiences.

You can also use this same method for lazy loading videos, backgrounds, and other assets.

IntersectionObserver to lazily load JavaScript assets

I discovered another application for the IntersectionObserver implementation that I used above to load lazy load images. The Google Lighthouse performance score on my homepage was being dinged due to the “impact of third-party code”.

impact of 3rd party js on performance scores

The largest third-party resource was is used for the reCaptcha on a contact form at the bottom of my homepage. It makes sense to not load it until the user scrolls down to that section – especially because the UX expects them to take time to fill out the form before hitting “submit” anyway.

Using the same pattern as above, I created a new `IntersectionObserver` and passed the contact form section as the target:

function captchaLazyLoad(){
	contactCaptchaTarget = document.getElementById('contactSection')
	if (!contactCaptchaTarget) {
        return;
    }
	let contactCaptchaObserver = new IntersectionObserver(function(entries, observer) {
		if (entries[0].isIntersecting) {
            var script = document.createElement('script');
		    script.src = "https://www.google.com/recaptcha/api.js";
		    document.body.appendChild(script);
            contactCaptchaObserver.unobserve(contactCaptchaTarget);
        }
	})
	contactCaptchaObserver.observe(contactCaptchaTarget);
}

I included this function to the already existing `DOMContentLoaded`  event listener just below the loop to observe lazy image elements:

<script>
function captchaLazyLoad(){
	contactCaptchaTarget = document.getElementById('contactSection')
	if (!contactCaptchaTarget) {
        return;
    }
	let contactCaptchaObserver = new IntersectionObserver(function(entries, observer) {
		if (entries[0].isIntersecting) {
            var script = document.createElement('script');
		    script.src = "https://www.google.com/recaptcha/api.js";
		    document.body.appendChild(script);
            contactCaptchaObserver.unobserve(contactCaptchaTarget);
        }
	})
	contactCaptchaObserver.observe(contactCaptchaTarget);
}

document.addEventListener("DOMContentLoaded", function() {
  var lazyImages = [].slice.call(document.querySelectorAll("img.lazy"));
  if ("IntersectionObserver" in window) {
    let lazyImageObserver = new IntersectionObserver(function(entries, observer) {
      entries.forEach(function(entry) {
        if (entry.isIntersecting) {
          let lazyImage = entry.target;
          lazyImage.src = lazyImage.dataset.src;
          // lazyImage.srcset = lazyImage.dataset.srcset;
          lazyImage.classList.remove("lazy");
          lazyImageObserver.unobserve(lazyImage);
        }
      });
    });

    lazyImages.forEach(function(lazyImage) {
      lazyImageObserver.observe(lazyImage);
    });

    //
    captchaLazyLoad()

  } else {
    // Possibly fall back to a more compatible method here if IntersectionObserver does not exist on the browser
  }

});

</script>

This update raised my Lighthouse performance score by fifteen points!

Statistics for hypothesis testing

statistics

Recently, I had to implement a function into SplitWit to determine statistical significance. Shortly after, I wrote a blog post that explains what statistical significance is, and how it is used.

Here, I’ll attempt to expound the code I wrote and what I learned along the way.

Statistical significance implies that the results of an experiment didn’t happen by random chance. SplitWit specifically deals with running A/B experiments (hypothesis testing) to help increase conversion rates. Getting this right is important.

Probability

Statistical significance is determined by calculating a “probability value”. The lower that value is, the more confident we can be that our results are probably not random. Statisticians refer to it as “p-value”. In this field of study, a p-value less than 0.05 is considered good. That translates to a less than 5% chance that the results of an experiment are due to error. Source.

When the probability of a sampling error is that low, we are said to have rejected the “null hypothesis” and affirmed our “alternative hypothesis”. Our alternative hypothesis, in the context of A/B website testing, refers to the experimental UI change we made successfully increasing our conversion rate. The null hypothesis represents the idea that our changes had no true affect on any improvement we may have seen.

SplitWit experiment results
SplitWit lets you set custom metrics to measure conversion rates

How do we calculate our p-value for an A/B split test on a eCommerce website?

Here are the steps to figure out if a hypothesis test is statistically significant:

  1. Determine the conversion rates for the control and variation
  2. Work out the standard error of the difference of those rates
  3. Derive a z-score using the conversion rates and the standard error
  4. Convert that z-score to a p-value
  5. Check if that p-value is below the desired confidence level (<0.05 for 95% confidence)

This recipe is based on a field of study called Bayesian statistics. It attempts to describe a degree of certainty for our experiments.

Conversion rates

The goal of running A/B split tests is to increase your website’s conversion rate. The idea is to make a UI change, show that variation to a portion of users, and then measure conversions against the control version. I calculate conversion rates by dividing the number of conversions by the number of visitors, giving me an average:

$control_conversion_rate = $control_conversions/$control_visitors;
$variation_conversion_rate = $variation_conversions/$variation_visitors;

Which ever version’s conversion rate is higher is the winner of the experiment. I calculate the winner’s uptick in conversion rate using this formula: (WinningConversionRate – LosingConversionRate) / LosingConversionRate

$uptick = 0;
if($control_conversion_rate > $variation_conversion_rate){
	$uptick = (($control_conversion_rate - $variation_conversion_rate) / ($variation_conversion_rate)) * 100;
}

if($control_conversion_rate < $variation_conversion_rate){
	$uptick = (($variation_conversion_rate - $control_conversion_rate) / ($control_conversion_rate)) * 100;
}

 

Calculating p-value

After researching, I determined that I would calculate my p-value from a “z-score”.

$p_value = calculate_p_value($z_score);

A z-score (also known as a standard score) tells us how far a data point is from the mean. Source.

For the purposes of A/B testing, the data points we are interested in is the conversion rates of our control and variation versions. Consider this code snippet for determining our z-score:

$z_score = ($variation_conversion_rate-$control_conversion_rate)/$standard_error;

This formula takes the difference between the two conversion rates, and divides it by their “standard error”. The standard error is meant to tell us how spread out our data is (sampling distribution). Source.

Standard error of two means’ difference

A conversion rate is essentially an average (mean). To properly determine our z-score, we’ll want to use the standard error of their difference.

First, we’d want to get the standard error of each of those rates. Source.

This is the formula to use: ( conversion_rate * ( 1 – conversion_rate ) / visitors )1/2

Translated as PHP code:

$standard_error_control = sqrt($control_conversion_rate * (1-$control_conversion_rate) / $control_visitors;)
$standard_error_variation = sqrt($variation_conversion_rate * (1-$variation_conversion_rate) / $variation_visitors);

Then, we’d use those values to find the standard error of their difference.

This is the formula: ( standard_error_control2 + standard_error_variation2  )1/2

Translated as PHP code:

$x = pow($standard_error_control, 2) + pow($standard_error_variation, 2);
$standard_error_of_difference = sqrt($x);

If we skip squaring our values in the 2nd step, we can also skip getting their square root in the first. Then, the code can be cleaned up, and put into a function:

public function standardErrorOfDifference($control_conversion_rate, $variation_conversion_rate, $control_visitors, $variation_visitors){
		
	$standard_error_1 = $control_conversion_rate * (1-$control_conversion_rate) / $control_visitors;
	$standard_error_2 = $variation_conversion_rate * (1-$variation_conversion_rate) / $variation_visitors;
	$x = $standard_error_1 + $standard_error_2;

	return sqrt($x);

}

This algorithm represents the “difference between proportions” and can be expressed by this formula: sqrt [p1(1-p1)/n1 + p2(1-p2)/n2]

Source.

Simplified even further, as PHP code:

$standard_error = sqrt( ($control_conversion_rate*(1-$control_conversion_rate)/$control_visitors)+($variation_conversion_rate*(1-$variation_conversion_rate)/$variation_visitors) );

Statistics as a service

Having considered all of these steps, we can put together a simple method to determine statistical significance. It takes the number of visitors and conversions for the control and the variation.

public function determineSignificance($controlVisitors, $variationVisitors, $controlHits, $variationHits){
	
	$control_conversion_rate = $control_hits/$control_visitors;
	$variation_conversion_rate = $variation_hits/$variation_visitors;

	$standard_error = sqrt( ($control_conversion_rate*(1-$control_conversion_rate)/$control_visitors)+($variation_conversion_rate*(1-$variation_conversion_rate)/$variation_visitors) );

	$z_score = ($variation_conversion_rate-$control_conversion_rate)/$standard_error;

	$p_value = $this->calculate_p_value($z_score);

	$significant = false;
	
	if($p_value<0.05){
		$significant = true;
	}else{
		$significant = false;
	}
        return $significant;
}

You can find this code on GitHub.

Additional References

https://github.com/markrogoyski/math-php#statistics—experiments

https://www.evanmiller.org/how-not-to-run-an-ab-test.html

https://codereview.stackexchange.com/questions/85508/certainty-of-two-comparative-values-a-b-results-certainty

https://www.mathsisfun.com/data/standard-deviation-formulas.html

https://sites.radford.edu/~biol-web/stats/standarderrorcalc.pdf

https://help.vwo.com/hc/en-us/articles/360033991053-How-Does-VWO-Calculate-the-Chance-to-Beat

https://towardsdatascience.com/the-math-behind-a-b-testing-with-example-code-part-1-of-2-7be752e1d06f

https://abtestguide.com/calc/

Automatic MySQL dump to S3

Automatic MySQL dump to S3

I have had some lousy luck with databases. In 2018, I created a fitness app for martial artists, and quickly gained over a hundred users in the first week. Shortly after, the server stopped resolving and I didn’t know why. I tried restarting it, but that didn’t help. Then, I stopped the EC2 instance from my AWS console. Little did I know, that would wipe the all of the data from that box. Ouch.

Recently, a client let me know that their site wasn’t working. A dreaded “error connecting to the database” message was all that resolved. I’d seen this one before – no sweat. Restarting the database usually does the trick: “sudo service mariadb restart”. The command line barked back at me: “Job for mariadb.service failed because the control process exited with error code.”

Uh-oh.

The database was corrupted. It needed to be deleted and reinstalled. Fortunately, I just happen to have a SQL dump for this site saved on my desktop. This was no way to live – in fear of the whims of servers.

Part of the issue is that I’m running MySQL on the same EC2 instance as the web server.  A more sophisticated architecture would move the database to RDS. This would provide automated backups, patches, and maintenance. It also costs more.

To keep cost low, I decided to automate MySQL dumps and upload to an S3 bucket. S3 storage is cheap ($0.20/GB), and data transfer from EC2 is free.

Deleting and Reinstalling the Database

If your existing database did crash and become corrupt, you’ll need to delete and reinstall it. To reset the database, I SSH’d into my EC2 instance. I navigated to `/var/lib/mysql`

cd /var/lib/mysql

Next, I deleted everything in that folder:

sudo rm -r *

Finally, I ran a command to reinitialize the database directory

mysql_install_db --user=mysql --basedir=/usr --datadir=/var/lib/mysql

Reference: https://serverfault.com/questions/812719/mysql-mariadb-not-starting

Afterwards, you’ll be prompted to reset the root password.

A CLI prompt to reset the root password after installing mariadb

You’ll still need to import your sql dump backups. I used phpMyAdmin to do that.

Scheduled backups

AWS Setup

The first step was to get things configured in my Amazon Web Services (AWS) console. I created a new S3 bucket. I also created a new IAM user, and added it to a group that included the permission policy “AmazonS3FullAccess”.

AWS policy to allow full S3 access
This policy provides full access to all buckets.

I went to the security credentials for that user, and copied down the access key ID and secret. I would use that info to access my S3 bucket programatically. All of the remaining steps take place from the command line, via SSH, against my server. From a Mac terminal, you could use a command like this to connect to an EC2 instance:

ssh -i /Users/antpace/Documents/keys/myKey.pem ec2-user@ec2-XX-XX-XX.us-west-2.compute.amazonaws.com

Once connected, I installed the software to allow programatic access to AWS:

curl "https://awscli.amazonaws.com/awscli-exe-linux-x86_64.zip" -o "awscliv2.zip"
unzip awscliv2.zip
sudo ./aws/install

Here is the reference for installing the AWS CLI on Linux.

Shell script

Shell scripts are programs that can be run directly by Linux. They’re great for automating tasks. To create the file on my server I ran: “nano backup.sh”. This assumes you already have the nano text editor installed. If not: “sudo yum install nano” (or, “sudo apt install nano”, depending on your Linux flavor).

Below is the full code I used. I’ll explain what each part of it does.

Credit: This code was largely inspired by a post from Marcelo Gornstein.

#!/bin/bash
AWS_ACCESS_KEY_ID=XXX \
AWS_SECRET_ACCESS_KEY=XXX \
S3_BUCKET=myBucketsName \
MYSQL_HOST=localhost \
MYSQL_PORT=3306 \
MYSQL_USER=XXX \
MYSQL_PASS=XXX \
MYSQL_DB=XXX \

cd /tmp
file=${MYSQL_DB}-$(date +%a).sql
mysqldump \
  --host ${MYSQL_HOST} \
  --port ${MYSQL_PORT} \
  -u ${MYSQL_USER} \
  --password="${MYSQL_PASS}" \
  ${MYSQL_DB} > ${file}
if [ "${?}" -eq 0 ]; then
  gzip ${file}
  AWS_ACCESS_KEY_ID=${AWS_ACCESS_KEY_ID} AWS_SECRET_ACCESS_KEY=${AWS_SECRET_ACCESS_KEY} aws s3 cp ${file}.gz s3://${S3_BUCKET}
  rm ${file}.gz
else
  echo "sql dump error"
  exit 1
fi

The first line tells the system what interpreter  to use: “#!/bin/bash”. Bash is a variation of the shell scripting language. The next eight lines are variables that contain details about my AWS S3 bucket, and the MySQL database connection.

After switching to a temporary directory, the filename is built. The name of the file is set to the database’s name plus the day of the week. If that file already exists (from the week previous), it’ll be overwritten.  Next, the sql file is created using mysqldump and the database connection variables from above. Once that operation is complete, then we zip the file, upload it to S3, and delete the zip from our temp folder.

If the mysqldump operation fails, we spit out an error message and exit the program. (Exit code 1 is a general catchall for errors. Anything other than 0 is considered an error. Valid error codes range between 1 and 255.)

Before this shell script can be used, we need to change its file permissions so that it is executable: “chmod +x backup.sh”

After all of this, I ran the file manually, and made sure it worked: “./backup.sh”

Sure enough, I received a success message. I also checked the S3 bucket and made sure the file was there.

S3 file dump

Scheduled Cronjob

The last part is to schedule this script to run every night. To do this, we’ll edit the Linux crontab file: “sudo crontab -e”. This file controls cronjobs – which are scheduled tasks that the system will run at set times.

The file opened in my terminal window using the vim text editor – which is notoriously harder to use than the nano editor we used before.

I had to hit ‘i’ to enter insertion mode. Then I right clicked, and pasted in my cronjob code. Then I pressed the escape key to exit insertion mode. Finally, I typed “wq!” to save my changes and quit.

Remember how crontab works:

minute | hour | day-of-month | month | day-of-week

I set the script to run, every day, at 2:30am:

30 2 * * * sudo /home/ec2-user/backup.sh

And that’s it. I made sure to check the next day to make sure my cronjob worked (it did). Hopefully now, I won’t lose production data ever again!

Updates

Request Time Too Skewed (update)

A while after setting this up, I randomly checked my S3 buckets to make sure everything was still working. Although it had been for most of my sites, one had not been backed up in almost 2 months! I shelled into that machine, and tried running the script manually. Sure enough, I received an error: “An error occurred (RequestTimeTooSkewed) when calling the PutObject operation: The difference between the request time and the current time is too large.

I checked the operating system’s current date and time, and it was off by 5 days. I’m not sure how that happened. I fixed it by installing and running “Network Time Protocol”:

sudo yum install ntp
sudo ntpdate ntp.ubuntu.com

After that, I was able to run my backup script successfully, without any S3 errors.

 


Nano text-editor tip I learned along the way:

You can delete chunks of text content using Nano. Use CTRL + Shift + 6 to enter selection mode, move the cursor to expand the block, and press CTRL + K to delete it.

Additional References:

https://www.javatpoint.com/steps-to-write-and-execute-a-shell-script

https://www.sitepoint.com/cron-jobs/

Drop down with CSS arrow

blog post about css menu with arrow

Here’s a quick one about how to create a drop-down UI element with an arrow via CSS. The aim is to create a menu that has drop-down sub-menus. Each drop-down should have an arrow that points up towards the parent element.

Here’s the HTML to structure the menu:

<div class="menu">
  <div class="menu-item">
    <span>Menu Item</span>
    <div class="drop-down-menu">
      <p>Sub-item</p>
      <p>Sub-item</p>
      <p>Sub-item</p>
    </div>
  </div>
  
  <div class="menu-item">
    <span>Menu Item 2</span>
  </div>
  
  <div class="menu-item">
    <span>Menu Item 3</span>
    <div class="drop-down-menu">
      <p>Sub-item</p>
      <p>Sub-item</p>
      <p>Sub-item</p>
    </div>
  </div>
  
</div>

I nest the sub-menu within the parent item, and use CSS to show it when a user mouses-over:

.menu-item:hover .drop-down-menu{
  display: block;
}

I line the menu-items in a row by setting display to ‘inline-block’. This is preferred over just ‘inline’, so that their height property is respected. This is important because I will create space between the parent item and sub-menu. If the two elements don’t actually overlap, then the hover state will be lost, closing the drop-down. See what I mean:

Drop down menu with CSS
The sub-menu element overlaps with the parent item, so that the hover state is not lost.

I also set the parent item position to relative, so that the drop-down will be absolutely positioned respective to it.

.menu-item{
  cursor: pointer;
  height: 50px;
  display: inline-block;
  position: relative;
}

Since the menu items have a height larger than the actual content, I apply borders to a child span within them:

.menu-item:first-child span{
  border:none;
}
.menu-item span{
  border-left: 1px solid black;
  padding: 0 10px;
}

The sub-menu styling is straight-forward. I set it to display: none, set a width, add a border and padding, and position it absolutely. Its top value pushes it off of the parent item a bit. Setting a left value to zero ensures that it will be aligned with its parent. (If you don’t set a left value, multiple sub-menus will all stack under the very first parent item.)

.drop-down-menu{
  display: none;
  width: 100px;
  position: absolute;
  background: white;
  border: 1px solid #301B46;
  padding: 20px;
  top: 40px;
  left: 0px;
}


The next step is building the arrow in the drop-down menu. The challenge is making the arrow’s border blend seamlessly with the container’s border. The illusion is achieved by overlapping the :before and :after pseudo-elements.

The triangle shape that forms the arrow is achieved by giving  a bottom border to an element with no height or width. This code pen animation does a phenomenal job of explaining the idea: https://codepen.io/chriscoyier/pen/lotjh

The drop-down’s :before element creates the white triangle that is the heart of the arrow itself. This also creates the gap in the sub-menu’s actual top border

The :after element creates another triangle that is behind and slightly above the first one – creating the illusion of a border that connects just right with the menu’s.

The illusion can be better revealed by manipulating the position and border-width of these pseudo-elements in the inspector.

Here is the code I used for those pseudo elements:

.drop-down-menu:before {
  content: "";
  position: absolute;
  border-color: rgba(194, 225, 245, 0);
  border: solid transparent;
  border-bottom-color: white;
  border-width: 11px;
  margin-left: -10px;
  top: -21px;
  right: 65px;
  z-index: 1;
} 

.drop-down-menu:after {
    content: "";
    position: absolute;
    right: 66px;
    top: -21px;
    width: 0;
    height: 0;
    border: solid transparent;
    border-width: 10px;
    border-bottom-color: #2B1A41;
    z-index: 0;
}

You can view the whole thing in action here: https://codepen.io/pacea87/pen/OJywqrj

Code generated from CSS Arrow Please helped me a lot when I was first figuring out how to do this right.