[08:40:30] Hello team, could I ask someone to chime in on this one https://wikitech.wikimedia.org/wiki/Helm/Upstream_Charts/gateway-api-crds ? [08:54:13] dpogorzelski: could you elaborate/link somewhere why this is required and can't be done with the existion istio CRDs? [08:59:14] for example Inference Extension which istio can implement only via gateway api and InferencePool is also specific to gateway api. kserve uses these primitives in order to provide functions like LLM specific routing (like k/v cache aware routing) [08:59:33] this is specific to serving LLM models on ml-serve [09:00:13] what this enables is also a different autoscaling mode that is built on top of vllm specific primitives rather than typical cpu/mem/requests metrics [09:00:26] but that's a second step [09:04:32] might be worth it adding this to the task as rn it's pretty unclear for a reader why this should be the only option [09:09:52] Agreed. It's a bit impenetrable for me, too. Would love a bit more of a high-level overview of the solution. [09:12:21] roger [09:13:03] jayme: you mentioned ST6.4.6 - I should know, but I'm finding APP docs a bit hard to navigate this year, for some reason. What is that project? [09:15:13] btullis: it's about research/design of autoscaling systems for k8s [09:15:28] where KEDA will most likely play a role - at least in evaluation [09:26:37] Ooh, interesting. Glad I asked. Thanks. [12:04:02] i provided some additional clarification in that task