@@ -96,7 +96,56 @@ payload processing is not standardized in Kubernetes today.
9696
9797# How?
9898
99- TODO in a later PR.
99+ This proposal is meant to drive towards multiple outcomes and engagements with
100+ the SIGs:
101+
102+ 1 . A proposal for "Payload Processors" in SIG Network, via [ Gateway API] .
103+ 2 . Development of example "GuardRails" extensions using Payload Processing
104+ (e.g. prompt guards). SIG Security (and potentially some LLM security
105+ specialists) will be engaged for review to get any concerns or insights they
106+ can provide.
107+ 3 . (possibly, to be discussed further) a proposal to re-implement the existing
108+ Body-Based Router (BBR) in the [ Gateway API Inference Extension (GIE)] using
109+ the standardized mechanisms.
110+ 4 . (possibly, to be discussed further) a connection with the existing Gateway
111+ API [ Firewall GEP] .
112+
113+ [ Gateway API ] :https://github.com/kubernetes-sigs/gateway-api
114+ [ Gateway API Inference Extension (GIE) ] :https://github.com/kubernetes-sigs/gateway-api-inference-extension
115+ [ Firewall GEP ] :https://github.com/kubernetes-sigs/gateway-api/issues/3614
116+
117+ ## Gateway API - Payload Processors
118+
119+ The following is a proposal for [ Gateway API] to enable "Payload Processing"
120+ within its API surface (` Gateway ` , ` HTTPRoute ` , etc). The overview includes
121+ high level concepts and technical requirements as derviced from our user
122+ stories and the implementation entation details include examples to illustrate
123+ how this would work (but are not intended to be new APIs that we actually
124+ propose).
125+
126+ [ Gateway API ] :https://github.com/kubernetes-sigs/gateway-api
127+
128+ ### Overview
129+
130+ > ** WIP** : this will be updated iteratively across multiple PRs.
131+
132+ Requirements:
133+
134+ * ** Declarative** : The administrator of a ` HTTPRoute ` needs to be able to add
135+ payload processors declaratively, as part of the route definition.
136+ * ** Ordered** : Each specified payload processor needs to be executed by the
137+ Gateway API implementation in the order it was configured.
138+ * ** Abundancy** : A route needs to be able to configure a potentially very large
139+ number of payload processors.
140+ * ** Routing** : Some payload processors may trigger routing to a specific
141+ backend (e.g. semantic routing, semantic caching). This optional routing may
142+ only trigger conditionally.
143+ * ** Failure Modes** : Payload processors need to have configurable failure modes
144+ which determine what happens if the extension fails during processing.
145+
146+ ### Implementation Details
147+
148+ TODO: in a later PR.
100149
101150> This should be left blank until the "What?" and "Why?" are agreed upon,
102151> as defining "How?" the goals are accomplished is not important unless we can
0 commit comments