Description
It looks like we run GET _cat/nodes?local in order to check a pod's readiness:
This API call does two things: it gets a list of nodes from the local cluster state, and then it reaches out to all the nodes in the cluster and obtains some extra info from them. If a single node in the cluster fails to respond in this second step within a few seconds then I think this means we consider all the pods to be failing their readiness checks.
I think it's a bug to accept ?local on this API at all given this gotcha (see elastic/elasticsearch#50088) but also think we should be using a different API for our readiness checks. For instance, GET _cluster/health?timeout=0s returns 200 OK iff the node is part of a cluster with a master, and GET / returns (something) if the node is actually alive and responding.
(I say "something" because in certain versions GET / return 503 Service Unavailable if there is no master, but more recent versions always return 200 OK here).
