Grid Site Route

# The grid_site_route filter as it appears under filters: in praxis.yaml, with the
# load_balancer block that always follows it. grid_site_route decides the cluster; load_balancer
# dials it and runs the health checks selection demotes on. The chart renders both from
# gridServing. Nothing in availability or prefix_affinity needs setting: every field has a
# default, listed in README.md, and shedding is the one switch, off unless set as here.
- filter: grid_site_route
  model_header: X-Gateway-Model-Name
  availability:
    shedding: true
  prefix_affinity:
    capacity: 16384
    ttl_secs: 600
- filter: load_balancer
  clusters:
    # One cluster per candidate, named as the serving config names it.
    - name: hq-east
      tls:
        ca:
          ca_path: /etc/praxis/backend-ca/0/service-ca.crt
        sni: grid-internal-openshift-ai-inference.ai-tenant-site-a.svc.cluster.local
        verify: true
      endpoints:
        - grid-internal-openshift-ai-inference.ai-tenant-site-a.svc.cluster.local.:443
      trusted_private_endpoints:
        - grid-internal-openshift-ai-inference.ai-tenant-site-a.svc.cluster.local
      health_check:
        type: tcp
        interval_ms: 5000
        timeout_ms: 2000
        healthy_threshold: 1
        unhealthy_threshold: 3
    - name: hq-west
      tls:
        ca:
          ca_path: /etc/praxis/backend-ca/1/service-ca.crt
        sni: grid-internal-openshift-ai-inference.ai-tenant-site-b.svc.cluster.local
        verify: true
      endpoints:
        - grid-internal-openshift-ai-inference.ai-tenant-site-b.svc.cluster.local.:443
      trusted_private_endpoints:
        - grid-internal-openshift-ai-inference.ai-tenant-site-b.svc.cluster.local
      health_check:
        type: tcp
        interval_ms: 5000
        timeout_ms: 2000
        healthy_threshold: 1
        unhealthy_threshold: 3