Hey everyone,

I've been keeping an eye on my server logs recently and noticed something I hate to see: the gunicorn worker for my **Market-Viewr** app was periodically dying and being restarted. The application was unstable under load, and it was time to investigate why.


![It works so much faster!](https://files.peakd.com/file/peakd-hive/thecrazygm/EoEwMZiUTxMXEjeDZ2CgouAkHfQqfFdmZjv7ZNAqPGt39mmPTfi21u34Q2DhsZKBhjH.png)


After digging into the logs and doing some performance profiling, the culprit became clear. The application was making frequent, expensive API calls to fetch token richlists from Hive-Engine, especially when a user requested a CSV export. I realized the CSV export was making a _separate, fresh pull_ of the entire richlist, even if the web page had just loaded it. This was inefficient and, for very large richlists, was causing the worker to time out or run out of resources.

To fix this and make the whole application more stable and performant, I've implemented two key backend improvements: **caching** and **retry logic**.

#### 1. Smart Caching with a Fallback

The biggest improvement was adding a caching layer. In simple terms, caching stores the result of an expensive operation (like an API call) for a set period. If another request for the same data comes in during that period, the app serves the stored result instead of hitting the API again. This is massively more efficient.

I used the `flask-caching` library and set it up with a clever fallback system:

- It first tries to connect to a **Redis** server, which is a fast, in-memory database perfect for caching.
- If Redis is unavailable for any reason, it automatically falls back to a **SimpleCache** that just stores the data in the application's memory.

This ensures the app still benefits from caching even if the Redis service isn't running. I've applied different cache timeouts to different pages – for example, the homepage with the list of all tokens is cached for longer, while a specific token's market page has a shorter cache time to keep the data fresher.

#### 2. Resilient API Calls with Retry Logic

The second issue is that sometimes API nodes can have a temporary blip or a network connection can fail. To make the app more resilient to these transient errors, I've configured the `requests` library to automatically retry failed requests.

If a connection to the Hive-Engine API fails, the app will now automatically try again a few times with a short delay between attempts. This means a temporary network hiccup is much less likely to result in an error page for the user.

#### The Result: A Faster, More Stable App

These backend changes should result in a noticeably better experience for you:

- **Faster Load Times:** Many pages will now load almost instantly from the cache.
- **Increased Reliability:** The site should be much more stable and less prone to errors or timeouts, especially when viewing tokens with large richlists.

While I was in there, I also made a small quality-of-life tweak to sort the richlist by the total holdings (staked + balance) by default, which gives a more accurate picture of a user's position.

These kinds of infrastructure improvements aren't as flashy as a new feature, but they are crucial for a good user experience. I hope you find the updated Market-Viewr to be a smoother and more reliable tool!

As always,
Michael Garcia a.k.a. TheCrazyGM