Enable L4 retries

[bedis]

L4 retries (option redispatch) is already enabled in defaults section. That said, it can be improve with the following items:

• allow it both globally and per service
• configure a customer number of retries
• enable "interval" to allow retrying the connection immediately on a new server instead of redispatching the last retry to a server

https://cbonte.github.io/haproxy-dconv/2.0/configuration.html#option%20redispatch https://cbonte.github.io/haproxy-dconv/2.0/configuration.html#retries

[NickMRamirez]

It would be useful to have retry-on empty-response included by default at least. During a rolling update of backend pods, new pods are added while old pods are incrementally deleted. If a client is connected to a soon-to-be deleted pod, they will likely receive a 502 response because the server would have disappeared midway through the session. This can be verified by creating a backend application that is slow and then triggering a rolling update after making a request to this application.

[bedis]

Hi @NickMRamirez , well, 502 also means the server "crashed" while processing the request. We don't want to enable such retry by default because we don't know the "idempotency" of the request. From my point of view, such type of retries should be configuration per service / ingress rule. (this includes all L7 retries)