CODEWITHSUNDEEP
resthttpamazon-elb
Emanuel George Hategan
Emanuel George Hategan

Reputation: 1263

http status code to signal an instance that is part of an ELB is unhealthy

As per https://docs.aws.amazon.com/elasticbeanstalk/latest/dg/using-features.healthstatus.html the web service should respond with 200 OK HTTP status code to signal a healthy application or any other one for unhealthy.

My questions is which HTTP status is most appropriate for an unhealthy application?

Out of all HTTP status classes below

simple logic suggests it should be a server error other than anything else and out of all the 5xx Server errors codes 503 Service Unavailable fits most naturally however I couldn't find any documentation supporting this.

Upvotes: 0

Views: 169

Answers (1)

VoiceOfUnreason
VoiceOfUnreason

Reputation: 57299

My questions is which HTTP status is most appropriate for an unhealthy application?

503 Service Unavailable, or a lower level TCP RST, are perfectly satisfactory ways for a service to announce that it isn't currently available. You can use Retry-After if you want to suggest a time to wait. I haven't seen any documentation suggesting that the ELB client will respect that header.

GCP and Azure are similar; 200 Healthy, anything else is not. The Azure document suggests a 500 Internal Server Error as one possible candidate, which is of course fine (Y00 is the coarse grained approximation of all Yxx codes).

The protocol for Consul checks is similar -- 2xx for healthy, 429 Too Many Requests for a warning (which is a really odd choice), and anything else for failure.

I'm not a big fan of using a Client Error to describe what is a server problem; but I don't know that it really hurts anything. The semantics of Retry-After is really only clearly defined for 3xx and 429/503, so if you are hoping to leverage that you should restrict yourself to those codes.

Otherwise, you could dig through the status code registry to see if there is a code you think is more suitable.

Upvotes: 1

Related Questions