routing

const (
	// PrefillEndpointHeader is the header name used to indicate Prefill worker <ip:port>
	PrefillEndpointHeader = "x-prefiller-host-port"

	// EncoderEndpointsHeader is the header name used to indicate Encoder workers <ip:port> list
	EncoderEndpointsHeader = "x-encoder-hosts-ports"

	// DataParallelEndpointHeader is the header name used to indicate the worker <ip:port> for Data Parallel
	DataParallelEndpointHeader = "x-data-parallel-host-port"

	// InferencePoolAPIGroup is the default InferencePool API group
	InferencePoolAPIGroup = "inference.networking.k8s.io"

	// PreferHeader is the standard HTTP "Prefer" header (RFC 7240). EPP
	// receives header keys lowercased.
	PreferHeader = "prefer"

	// PreferIfAvailable is the preference token the coordinator sets to mark a
	// request as a speculative early-decode attempt: route to a decode worker
	// only if its KV cache already covers the prompt (at least partially); otherwise EPP surfaces
	// 412 Precondition Failed so the coordinator restarts the pipeline.
	PreferIfAvailable = "if-available"
)