Service Is Normal, but the Page May Not Be: A WordPress Front-End Troubleshooting Case
Article Abstract
During a visit to the WordPress blog via a cellular network, the page loaded abnormally and triggered a browser permission prompt. Troubleshooting revealed that the WP Editor.md plugin was mistakenly configured with a home intranet IP as the static resource path. Because of this configuration, when public network users accessed the site, resources such as KaTeX formula rendering and emoji displays failed to load as they referenced internal network addresses, triggering browser security policy blocks. The issue was resolved after switching to a CDN resource path, demonstrating the limitation that server-side monitoring cannot cover front-end resource loading anomalies. This incident reflects that when a WordPress system relies on plugins, themes, and the core software working in concert, page integrity is highly dependent on the accuracy of resource references. Resource loading logic must be continuously verified across multiple network environments, and the health check system should integrate end-to-end testing with real-user access scenarios to avoid front-end anomalies caused by misconfigured static resource paths.
Qwen3-14B · 2026-11-13

1 Evolution of the Blog Availability Monitoring System

Following multiple upgrades to my blog's architecture, its underlying operating environment has also grown increasingly complex: in addition to WordPress itself, it involves read/write splitting (2 read-only nodes and 1 primary write node) along with a series of services like Cloudflare. Together, these components constitute the blog's operational system, and an anomaly in any single link can compromise the final visiting experience. Relying solely on manually opening the site occasionally to check makes it difficult to detect promptly whether blog services are functioning properly. Consequently, I started building automated health monitoring and incident notification mechanisms for the blog quite early on.

Initially, I adopted a self-hosted Uptime monitoring system. Once an anomaly in blog services was detected, it would send notifications to my phone and computer via the Bark service (see article: "Building a Real-time Application Health Monitoring and Alerting System Based on Uptime and Bark"). This approach was relatively simple to implement and met basic availability monitoring needs.


Bark is a free, open-source push notification service tailored for the Apple ecosystem, mainly consisting of a client and a server. The client is a free app installed on Apple devices (iPhone, iPad; it can also be used on Apple Silicon Macs that support running iOS/iPadOS apps), responsible for receiving and displaying push notifications. The server is responsible for receiving HTTP requests sent by external apps or monitoring systems (providing an interface similar to a Webhook) and forwarding notifications to the corresponding Apple devices via Apple's APNs (Apple Push Notification service).

The server can use the free public service, or you can self-host it (friends who want to build their own can refer to the article: "Building a Bark Server-based message push server》)。

Bark is very simple to use: you only need to send an HTTP GET or POST request to the designated API address to push custom titles and content to the target device. Therefore, whether it is an Uptime downtime alert or runtime status notifications from other scripts and applications, they can all be easily sent via Bark.


However, as the blog architecture grew more complex, I gradually realized it suffered from two notable limitations:

Monitoring requests must pass through Cloudflare's security policy checks. Although monitoring requests can be allowed through via custom security rules, this means having to maintain an extra set of security rules specifically for monitoring traffic, and one must constantly consider the interplay between these rules and existing security policies.

A self-hosted Uptime essentially reflects only the network perspective of the VPS it resides on: while it can verify whether the path from that VPS to the Cloudflare edge network is clear, it cannot reflect whether users in other regions and on other carriers can access the blog normally. In other words, even if the monitoring results appear completely normal, it does not fully guarantee that the blog maintains good accessibility globally.

To resolve these issues, I later adjusted the monitoring architecture: first, I developed a dedicated heartbeat/liveness probe plugin for WordPress, optimizing the health check interface to return probe results using the HTTP 204 status code (see article: "WordPress Health Check API Optimization: Building an Active Probe Plugin with 204 Responses"), avoiding the return of unnecessary response bodies and reducing response overhead for individual health checks.

Then, combined with Cloudflare's own Health Checks and event notification mechanisms, Cloudflare executes monitoring tasks in the cloud and initiates health checks from multiple regions. Through this approach, I can verify the availability of the blog's public ingress and each WordPress node from diverse network locations, no longer confined to the single network perspective of one VPS. For the implementation process, see the article: "Cloudflare Monitoring and Alerting Combination in Practice: Lightweight O&M Experience Brought by Health Checks + Event Alerts》。

Finally, combining Bark mobile notifications with email alerts, I gradually established a relatively complete blog health monitoring system: whether an anomaly occurs at the public network access ingress or a failure arises on an underlying WordPress node, it can be promptly detected through the corresponding monitoring and notification mechanisms.

Judging from the actual results at the time, this system was already capable of meeting my requirements for monitoring blog service availability. I originally assumed that as long as the public ingress and backend nodes remained normal, visitors would essentially be guaranteed smooth access to the blog.

Yet a recent access anomaly made me realize: a monitoring system showing normal services does not necessarily mean the access paths for all users are sound.

2 An Abnormal Access Experience

While out and about recently, I noticed a rather peculiar issue: after switching my phone to a cellular network and visiting my blog via a browser, the page consistently failed to load properly, with the progress bar always stalling halfway:

image.png

However, this issue was not persistent. When I was at home, whether using a desktop computer, an iPad, or a mobile browser, the blog opened without issue—meaning the problem seemed to appear only when accessing the blog from a phone on a cellular network. At first, I thought I was the only one encountering it, but other friends in the TG group reported it as well:
image.png

I first checked Cloudflare's health monitoring logs and found no anomaly alerts; the monitoring status for both the public ingress and backend nodes was normal. Then, I checked the DNS resolution results under home broadband and mobile cellular networks respectively, finding that the blog domain properly resolved to Cloudflare's IP address under both network environments:

image.png

image.png

Based on the information available at the time, neither server anomalies nor obvious DNS resolution issues were detected. Considering the complex factors in the domestic network environment and the fact that access from home was always normal, I did not pay it much mind, thinking it might just be a temporary glitch on the cellular network side that would resolve on its own after a while.

Later, however, another group member raised an issue: recently, when visiting my site, the browser would suddenly display a permission prompt requesting the website to ”access other devices on your local network.” Hearing this, I felt something was off, so I tested it using the browser's incognito mode, and sure enough, I was able to reproduce it:

image.png

This problem could no longer be put off. After all, when a regular visitor visits a blog and a permission prompt suddenly pops up asking to access local network devices, it inevitably raises suspicions. Crucially, I had no idea when the website had started requesting these permissions, let alone why it would want to access a visitor's local network. Thus, I hurried to troubleshoot.

3 From Abnormal Requests to Problem Resolution

Since the browser displayed a permission prompt to ”access other devices on your local network,” it indicated that the page loading process very likely involved attempts to access an intranet IP. The most direct troubleshooting method was to inspect which resources were actually loaded when opening the web page.

I opened incognito mode in the browser, went to the ”Network” tab in Developer Tools, and visited the blog again. Sure enough, I spotted an anomaly:

image.png

All the error-throwing resources pointed to the path ”https://192.168.10.xx:xx/wp-content/plugins/wp-editormd/assets/“, which is the directory where my installed WP Editor.md plugin resides.

Note: WP Editor.md is a plugin that provides Markdown editing and related extension features for WordPress, supporting writing posts in Markdown syntax, code highlighting, mathematical formula rendering, and more.

Seeing this, it suddenly occurred to me that a few weeks ago, I had encountered an error with the wp-editormd plugin in the WordPress dashboard:

image.png

At the time, I suspected an issue with the resources on the CDN, so I directly switched the resource loading method to ”Use local resources”:
image.png


Here, I also need to explain why the local resource address I configured uses a home internal network IP, which is related to the multi-node architecture currently used by the blog: the Mac mini primary write node is normally responsible for writing and modifying posts and does not directly handle external publishing tasks; the nodes that actually provide external blog access services are two other read-only nodes. Therefore, if I access the blog via the public domain name, I cannot determine which read-only node is currently being accessed, nor can I directly access the primary write node this way.

To facilitate daily management and resource access between different nodes, I have always used internal network addresses to access the primary write node and read-only node in my home LAN, while the read-only node on the Chicago VPS is accessed via a Tailscale IP. Therefore, when configuring the local resources for WP Editor.md, I casually used the internal network address of the home primary write node 192.168.10.xx:xx, without considering that this address might ultimately be written into the front-end pages facing public visitors.

Additionally, for those interested in the WordPress multi-active architecture, you can refer to the article:Implementation Scheme of WordPress Multi-Active Architecture (Simplified Version) in Personal Blogs。


After switching, the dashboard error indeed disappeared, so I assumed the problem was resolved. However, at the time, I took it for granted that WP Editor.md only affected the admin post editing functionality and wouldn't affect the front-end pages. Looking back now, that judgment was clearly overly simplistic.

From the browser's request logs, there were actually three resources loaded to the front end:

/wp-content/plugins/wp-editormd/assets/KaTeX/katex.min.css?ver=10.2.1
/wp-content/plugins/wp-editormd/assets/Emojify.js/css/basic/emojify.min.css?ver=10.2.1
/wp-content/plugins/wp-editormd/assets/FrontStyle/frontstyle.min.js?ver=10.2.1

They involve KaTeX mathematical formula rendering, Emojify emoji display, and the plugin's front-end styling functionality, respectively. In other words, these were not resources strictly for the admin editor, but actually appeared on the blog's front-end pages. Since an absolute path to the home internal network had been configured at the time, when a visitor's browser loaded these resources, it triggered access restrictions on local network resources, leading to the permission prompt.

As for my resolution process, it was somewhat unexpected. Originally, I planned to change the static resource path to point to the blog's public domain name https://blog.tangwudi.com/wp-content/plugins/wp-editormd, which would at least prevent the front end from continuing to reference home internal network resources that external visitors cannot access. However, this adjustment would also introduce an issue: when I usually access the WordPress dashboard via an internal network address to edit articles, requests for static resources would still point to the public domain name. Because the request Referer would not match existing Cloudflare WAF rules, it might be blocked, so a minor modification to the WAF rules would also be required.

Unexpectedly, when rechecking the plugin configuration, I tried switching back to “Use CDN Resources” again, and the previous resource loading error had actually disappeared. Since the CDN resources were back to normal, it naturally saved the trouble of modifying the static resource path and adjusting WAF rules. As for why the CDN resources had an issue in the first place, I didn't dig any deeper—I'll look into it if I run into it again next time~.

After saving the configuration, I accessed the blog again using the browser's incognito mode, and the permission prompt requesting access to local network devices was gone. Then I switched to the phone's cellular network for testing, and normal access to the blog was restored as well.

4 From Server-Side Monitoring to Real User Experience

Looking back now, the issue where the mobile browser got stuck halfway through the loading progress bar when using cellular data must have been caused by the same reason:192.168.10.xx:xx It belongs to the private IP addresses used in the home LAN. When accessing within the home LAN, this problem wasn't noticeable; however, after switching to the cellular network, because the device could not access this private IP address at all, the abnormal page loading became much more prominent.

This impact is not only evident when accessing via mobile cellular networks. On computers outside the home local network, when accessing the blog via the public domain name, besides Chromium-based browsers popping up a permission prompt saying “The website is trying to access local network devices,” resource requests in the page pointing to intranet addresses cannot complete normally either. The browser needs to wait for these requests to process, so the overall page loading speed may also be affected. No wonder group members had reported similar issues before:

image.png

It's just that the specific manifestations vary across different browsers and network environments, and they don't necessarily get stuck right in the middle of the loading progress bar like mobile browsers do.

The reason this issue went unnoticed by me for so long is precisely because I usually test the blog within my home local network most of the time. Although the incorrect resource URLs could not be accessed normally in this environment either, their impact on page loading speed was not noticeable. Moreover, since these resources are not necessary for normal blog browsing, and I rarely use plugin front-end features like KaTeX math formula rendering or Emojify emoji display, it never caught my attention until this hidden issue was exposed when accessing from an external network.

This issue also made me realize that normal server-side monitoring does not necessarily mean a normal user access experience. The health check systems we usually build mainly verify whether the service itself responds normally and whether specific endpoints and network links are available, but they cannot fully cover every step involved when visitors open a web page.

For a system like WordPress, which consists of the core program, themes, and plugins, whether a page can render properly also depends on whether JavaScript, CSS, images, and other resources can load correctly. Even if the server responds normally, as long as a resource references an incorrect URL or is affected by browser security policies, user network environments, and other factors, visitors may still encounter abnormal page loading. This WP Editor.md plugin issue is a typical example: the WordPress service itself had no anomalies, and Cloudflare's health monitor raised no alerts, yet the access experience of regular visitors was already affected.

Ultimately, automated monitoring is responsible for detecting service anomalies as early as possible, while real-world access testing ensures that users can actually use the service normally. The former improves the timeliness of issue detection, while the latter makes up for gaps in monitoring coverage. For a personal blog, you don't necessarily need to build a complex testing system, but you shouldn't equate a normal monitoring status with everything being fine on the site. Whether a service is truly available must ultimately be judged by the user's actual access experience.

Note: WP Editor.md hasn't been updated for over 6 years. I previously felt it was handy enough and didn't feel like bothering with alternatives, so I kept using it until now. After this issue, it seems time to consider finding a replacement.

View Related Categories · 3 Matches
📌 Content Structure Prompt:
This content is part of the following knowledge map, where you can view the complete content path: WordPress Learning Roadmap。
📎 Related Articles
Share this article
The blog content is original, please indicate the source when reposting! The RSS address of the blog is:https://blog.tangwudi.com/feed, welcome to subscribe; if needed, you can join theTelegram Groupto discuss questions together.

Comments

  1. Linux Edge 157.0.0.0
    1 week ago
    2026-11-13 15:06:27

    Speaking of WP Editor.md, I forked a copy and fixed several security vulnerabilities, ancient dependencies, a few functional bugs, and compatibility with PHP 7.4+. Feel free to give it a try (remember to back up the original one, though I haven't encountered any issues on my own site) athttps://github.com/tjsky/WP-Editor.md

    • Tang Wudi
      Autumn wind on the Wei River
      Macintosh Chrome 157.0.0.0
      1 week ago
      2026-11-13 15:09:36

      Alright, I'll look around first to see if there's any suitable one; if not, I'll come try your version.

      • Tang Wudi
        Linux Edge 157.0.0.0
        1 week ago
        2026-11-13 17:14:35

        Haha, definitely give it a try. I've tested other Markdown editors as well: Markup Markdown is too rudimentary, and posts are directly stored in the database in Markdown format. WP Githuber MD has also been discontinued, leaving behind multiple security vulnerabilities and functional bugs, which would still require fixing if switched over. After testing around, I still found WP Editor.md the best to use, so I took advantage of the National Day holiday to start patching it. (The “plugin asset package is corrupted” issue you mentioned has also been fixed; this problem was actually accompanied by another issue where "whenever you click any page in the admin dashboard, the plugin sends a synchronous request to an external server to check asset versions, causing a 6-second hang while waiting for a timeout if the CDN resources are inaccessible.")

Send Comment Edit Comment


				
|´・ω・)ノ
ヾ(≧∇≦*)ゝ
(☆ω☆)
(╯‵□′)╯︵┴─┴
 ̄﹃ ̄
(/ω\)
∠( ᐛ 」∠)_
(๑•̀ㅁ•́ฅ)
→_→
୧(๑•̀⌄•́๑)૭
٩(ˊᗜˋ*)و
(ノ°ο°)ノ
(´இ皿இ`)
⌇●﹏●⌇
(ฅ´ω`ฅ)
(╯°A°)╯︵○○○
φ( ̄∇ ̄o)
ヾ(´・ ・`。)ノ"
( ง ᵒ̌皿ᵒ̌)ง⁼³₌₃
(ó﹏ò。)
Σ(っ °Д °;)っ
( ,,´・ω・)ノ"(´っω・`。)
╮(╯▽╰)╭
o(*////▽////*)q
>﹏<
( ๑´•ω•) "(ㆆᴗ
😂
😀
😅
😊
🙂
🙃
😌
😍
😘
😜
😝
😏
😒
🙄
😳
😡
😔
😫
😱
😭
💩
👻
🙌
🖕
👍
👫
👬
👭
🌚
🌝
🙈
💊
😶
🙏
🍦
🍉
😣
Source: github.com/k4yt3x/flowerhd
Kaomoji
Emoji
Little Dinosaur
Flower!
Previous
Next
       

👋 Welcome to “Wudi's Personal Blog”

Here, long-term exploration is mainly carried out around the following directions:

🧱 Personal Digital Infrastructure and Blog System Construction
☁️ Cloudflare and Network Architecture Practice
🧠 AI and Knowledge System Exploration
🛡️ Network Security and Access Optimization
🎵 Music and Sound Cognition
👁️ Cognitive Perspectives and Worldviews