Debugging a 150k-Request 404 Storm: How I Saved My WordPress Server from High Load

Have you ever looked at your server’s top and seen mysqld and php-fpm battling for 40% of your CPU, even though your traffic shouldn’t be that high?

That was exactly the situation I found myself in this week on my Ubuntu server. Here’s the story of how a small missing CSS file turned into a massive performance nightmare—and how I fixed it.

---

🕵️ The Investigation: Following the CPU Spikes

It started with a slow-loading site. A quick check of the process list revealed multiple php-fpm processes consuming 12-15% CPU each, while MySQL was hovering around 36%.

Step 1: Identify the Culprit Site

Using the command line, I audited the active processes and log sizes:

ps -eo user,pcpu,pmem,comm --sort=-pcpu | head -n 10

One site stood out: culprit.vengalavinay.com. It was breathing heavy.

Step 2: The “Smoking Gun” in the Logs

I dove into the Apache access logs for that specific site. What I found was shocking:

sudo grep " 404 " /var/log/apache2/culprit.vengalavinay.com_access.log | awk '{print $7}' | sort | uniq -c | sort -nr | head -n 5

Results:

• 151,966 requests for /wp-content/themes/jobaway/assets/css/flaticon-one.css
• 722 requests for /wp-content/themes/jobaway/assets/css/icomoon.css

That’s over 150,000 requests in just 5 minutes from the server’s own internal IP!

---

💡 The Problem: Why 404s via PHP are “Expensive”

In a standard WordPress setup, when a static file (like a .css or .png) is missing, Apache doesn’t just say 404. Instead, it hands the request off to index.php. This means every time that missing CSS file was requested:

1. PHP-FPM was invoked.
2. MySQL was queried to check for slugs/redirects.
3. WordPress Internal Logic ran just to tell the user “File Not Found.”

Multiply that by 150,000 requests, and you have a recipe for server exhaustion.

---

🛠️ The Solution: From Mitigation to Optimization

1. The Quick Fix (The “Dummy File” Trick)

To stop the expensive PHP execution immediately, I created an empty, zero-byte file at the expected paths:

sudo touch /home/culpritvengala/public_html/wp-content/themes/jobaway/assets/css/flaticon-one.css
sudo touch /home/culpritvengala/public_html/wp-content/themes/jobaway/assets/css/icomoon.css

Suddenly, those 150,000 requests became cheap static asset calls. CPU usage for PHP dropped from 12% to under 1% instantly.

2. Tuning for the Future (LCP Optimization)

To ensure the site doesn’t just survive but thrives, I implemented three core performance upgrades:

• HTTP/2 Multiplexing: Switched to the h2 protocol to allow simultaneous asset downloads.
• Aggressive Browser Caching: Used mod_expires to cache images, fonts, and scripts for 1 year.
• PHP Opcache Tuning: Increased Opcache memory to 256MB for higher throughput.

---

📊 The Results: A Healthier Server

Metric	Before	After
PHP CPU Usage	~12.5% per process	< 1%
MySQL CPU Load	~36%	~33% (Stabilizing)
Asset Delivery	HTTP/1.1 (Sequential)	HTTP/2 (Parallel)

Scaling isn’t always about bigger servers—sometimes, it’s just about finding that one missing file.