With a managed PostgreSQL,
global.postgresql.authMode: iam reaches every
service: each runtime service’s database.iamAuth inherits it when unset.
Set agentgateway.database.iamAuth or trustguard.database.iamAuth explicitly
only for a mixed-auth install. The token provider (AWS, Azure, or GCP) and the
setup for each are in
IAM database authentication.AWS (EKS)
AWS (EKS)
ACM certificates terminate at the load balancer, so they cover Ingress hosts but
not the layer-4 endpoints a central control
plane publishes — TLS terminates in those
pods and ACM does not export private keys. Use cert-manager or your own PKI
there, with
service.beta.kubernetes.io/aws-load-balancer-type: nlb and
aws-load-balancer-scheme: internal for private callers.Azure (AKS)
Azure (AKS)
An internal load balancer is
service.beta.kubernetes.io/azure-load-balancer-internal: "true".GCP (GKE)
GCP (GKE)
--num-nodes 1 in a regional cluster gives one node per zone, so three in total.For GKE Ingress set
global.platform: gcp,
global.ingress.gcp.staticIpName, and managedCertificates. Google-managed
certificates reject wildcards, so if you rely on the chart’s wildcard hosts
(*.llm.<domain> / *.mcp.<domain>) use cert-manager, or set
agentgateway.config.autoWildcardHosts: false and list exact hosts.A plain LoadBalancer Service already gives a layer-4 passthrough load balancer;
its private form is networking.gke.io/load-balancer-type: "Internal".Conformant Kubernetes
Conformant Kubernetes
For clusters without a dedicated guide. Confirm the capabilities first:You need a default StorageClass (or
global.storageClass), an Ingress or Gateway
implementation, DNS for global.domain, and TLS from cert-manager or
pre-created Secrets.GPU nodes
Only needed if you run Firewall on GPU, which is opt-in. It requires a separate GPU node pool, the vendor device plugin, and matching labels and taints — see GPU Firewall workers. Plan these before creating the pool:- GPU slots. Every Firewall worker pod requests one GPU
(
nvidia.com/gpu: 1), seven with all products and six with TrustGuard only. The pool needs at least as many slots as worker pods: GPUs × MPS clients per GPU. Clients per GPU are bounded bycudaMpsPinnedDeviceMemLimit; the rule is in GPU Firewall workers. - Sharing is set at creation. On GKE, the GPU sharing strategy and
max_shared_clients_per_gpucan only be set when the node pool is created. Changing them means a new pool. - Boot disk of 100 GB or more. The
firewall-gpuimage is about 10 GB compressed and 20–25 GB unpacked. On a smaller disk the node goes intoDiskPressureduring the pull and evicts pods. - First pull time. Pulling that image through a registry proxy or a
pull-through cache can take more than 30 minutes the first time. Mirror it
into your registry ahead of the install, or allow for it in
--timeout.