Add troubleshooting docs for API request errors

docs/summary.md

@@ -46,6 +46,7 @@
 ## Troubleshooting
 
 * [API is stuck updating](troubleshooting/stuck-updating.md)
+* [404/503 API responses](troubleshooting/api-request-errors.md)
 * [NVIDIA runtime not found](troubleshooting/nvidia-container-runtime-not-found.md)
 * [TF session in predict()](troubleshooting/tf-session-in-predict.md)
 

docs/troubleshooting/api-request-errors.md

@@ -0,0 +1,9 @@
+# 404 or 503 error responses from API requests
+
+_WARNING: you are on the master branch, please refer to the docs on the branch that matches your `cortex version`_
+
+When making prediction requests to your API, it's possible to get a `{"message":"Not Found"}` error message (with HTTP status code `404`), or a `no healthy upstream` error message (with HTTP status code `503`). This means that there are currently no live replicas running for your API. This could happen for a few reasons:
+
+1. It's possible that your API is simply not ready yet. You can check the status of your API with `cortex get API_NAME`, and stream the logs with `cortex logs API_NAME`.
+1. Your API may have errored during initialization or while responding to a previous request. `cortex get API_NAME` will show the status of your API, and you can view the logs with `cortex logs API_NAME`.
+1. If `cortex get API_NAME` shows your API's status as "updating" for a while and if `cortex logs API_NAME` doesn't shed any light onto what may be wrong, please see the [API is stuck updating](stuck-updating.md) troubleshooting guide.