[APM] Investigate whether we handle errors appropriately and use the right status codes

[gbamparop]

One of the metrics to identify the health of the APM app is the request success rate for internal and external APIs as it should be accessible whenever customers use it through the browser or the Kibana REST API. Monitoring client errors is also important as they could deteriorate user experience.

Investigating whether we handle errors appropriately and using the right status codes could help us with monitoring errors and identifying underlying issues.

Use of APM data and Kibana logs in cloud could help surfacing some other issues.

Some examples we might want to have a look at are:

• For serverless, the sourcemap endpoints throw a 501 Not Implemented error while fleet endpoints and agent configuration endpoints throw a 404 Not Found error.
• Use the appropriate status code for client and server errors
• Use descriptive error messages without compromises in security
• Errors that are not handled appropriately and end up being caught by platform code
• Places where APIs are called without the right permissions and where we could consider checking for user permissions first and displaying an appropriate message instead
  • ML feature might be an example
• Investigate whether returning a 499 status code is the best approach for aborted requests

Links

• Pitch (internal)

[elasticmachine]

Pinging @elastic/apm-ui (Team:APM)