JuHyung Son
JuHyung Son
@VaibhavMalik4187 sure
@sivanantha321 oh i just edited. it was not logger, it was annotation
yes, prev revision's desired replicas should be to 0, but it does not
follow this, isvc.yaml ```yaml apiVersion: "serving.kserve.io/v1beta1" kind: "InferenceService" metadata: name: inferece-service-70 namespace: default annotations: prometheus.io/scrape: "true" # queue.sidecar.serving.knative.dev/resourcePercentage: "10" spec: predictor: minReplicas: 1 scaleTarget: 2 canaryTrafficPercent: 100 logger: mode: all...
as i tested. in 0.10, isvc become ready=True and prev revision does not set to 0. and in 0.11 even isvc ready does not to True, it is False
what is your k8s env? i'm using minikube v1.24.10 on mac m2
i checked on my dev cluster which kserve 0.10 production env. and it works well. so maybe this is problem on minikube or kserve quick install version
strange... i update minikube k8s to 1.28.3. but on my env, same problem remains.. below screenshot, first redbox is isvc with annotation and all revision's desired replica is 1 (0001,...
one more thing, both of them are minReplicas to 0, so they are serverless. but only inference-service-13 actually working as serverless and inference-service-12 does not scale to zero, even message...
i tested versions same as you. and there is no splitting traffic.. nothing deployed except istio knative kserve cert manager