diff --git a/curio/k8s/ingress.yaml b/curio/k8s/ingress.yaml index 648f90e..128cdda 100644 --- a/curio/k8s/ingress.yaml +++ b/curio/k8s/ingress.yaml @@ -8,6 +8,7 @@ metadata: k8s.apisix.apache.org/proxy-rewrite: '{"uri": "/$1", "regex_uri": ["^/api/whisper/(.*)", "/$1"]}' k8s.apisix.apache.org/upstream-read-timeout: "120s" k8s.apisix.apache.org/upstream-send-timeout: "120s" + k8s.apisix.apache.org/client-max-body-size: "50m" spec: ingressClassName: apisix rules: diff --git a/curio/k8s/whisper.yaml b/curio/k8s/whisper.yaml index a5c9e23..a035bb8 100644 --- a/curio/k8s/whisper.yaml +++ b/curio/k8s/whisper.yaml @@ -21,10 +21,16 @@ spec: resources: requests: cpu: "250m" - memory: "512Mi" # Drastically reduced to ensure scheduling + memory: "512Mi" limits: cpu: "1" - memory: "1Gi" + memory: "2Gi" # Increased for safe model loading + readinessProbe: + httpGet: + path: /health + port: 8000 + initialDelaySeconds: 15 + periodSeconds: 10 env: - name: WHISPER_MODEL value: "tiny"