I Thought Redis Would Save My API. Just 500 Requests Proved Me Wrong.

In the author's attempt to create a weather API backend, they initially used Redis to cache API responses and improve performance. However, the system quickly proved to be unstable as only 500 requests were made. The author had anticipated this issue but was taken aback by its early appearance. The post delves into the specifics of the problem encountered below.

Initially, a simple API was constructed to mimic a weather service, avoiding the rate limit imposed by the actual weather API for testing purposes. The `handleWeatherJSON` function read JSON data from a file, set the response header to JSON format, and returned the data upon a successful request. The main function then established a local server to listen on port 3000 for incoming requests.

Moving on, the actual system was built to serve users. In this case, a basic Weather API was used, although any third-party service could be substituted. The strategy was to cache the API response for subsequent requests to the same data, thus avoiding additional API calls. The `callTheAPI` function handled this by first checking if the requested data was available in Redis.

If it was, the cached response was decoded and sent back to the user. If not, the function made a request to the upstream API, stored the response in Redis for future use, and then returned it to the requester. For comprehensive load testing, K6 was used to simulate 500 concurrent requests to the API. The test code imported necessary modules, defined custom metrics for tracking cache hits and misses, and then initiated the load test.