The recovery mode section does not say which failures count #50

What happens

The Cloud Request Engine's recovery mode starts a recovery period when a significant number of request failures occur within a brief period, and during the period no request reaches the cloud service. The section gives the purpose as protecting a site when the cloud service becomes slow or fails, but does not say which failures count towards FailuresToEnterRecovery (lines 428 to 430).

The recovery period belongs to the engine, so every request shares it. The .NET pipeline counts every failure, including a 400 to 499 answer about one request's own content (CloudRequestEngine.cs lines 580 to 593). A few requests from one visitor that the cloud service refuses can then start a recovery period that stops every visitor's requests reaching the cloud service, and keep it going. The Rust SDK counted every failure too.

What is needed

The recovery mode section to say that a failure counts only when it says the cloud service is in trouble, being:

• a request that did not complete, such as a timeout or a connection failure
• a response with a status of 500 or more
• a response with a status of 429
• a response with a success status whose body could not be used

and that any other status from 400 to 499 does not count.

The change is #54. The Rust SDK's change is 51Degrees/51degrees-rust#198.