Skip to main content
This page covers the GCP-specific bits. The actual Helm install (values, ingress, env vars) lives in Install with Helm.

​
Persistent storage

The shared RWX PVC (used by prismeai-functions) can be backed by either:
  • Filestore with regional redundancy: managed NFS, lowest latency for the functions sandbox. Survives a zonal outage.
  • Cloud Storage FUSE (via the GCS FUSE CSI driver): mounts a GCS bucket as a filesystem. Cheaper at large scale and inherently regional / multi-region, with higher latency than Filestore; fine for the function code volume, less ideal for hot temp files.
Pick Filestore for hot workloads, GCS FUSE when storage cost dominates.

​
Infrastructure as Code

Use Terraform with the Google provider; Google’s documentation is the canonical reference. Prisme.ai doesn’t ship a dedicated GCP IaC bundle.

​
DNS and TLS

Create two records (Cloud DNS or your registrar):
For TLS, use Google-managed certificates declared on the Ingress resource. They auto-renew.

​
Workload Identity

For GCS, Secret Manager and Cloud SQL access without static credentials:
  1. Enable Workload Identity on the GKE cluster.
  2. Create a Kubernetes ServiceAccount per namespace (e.g. prismeai-backends-sa) and bind it to a Google ServiceAccount with the required IAM roles (roles/storage.objectAdmin, roles/secretmanager.secretAccessor, roles/cloudsql.client).
  3. Set serviceAccount.name: prismeai-backends-sa on every backend service in your Helm values.
GCS can also be accessed via HMAC keys for S3-compatible drivers, useful if you want to keep your config identical to an AWS deployment.

​
Ingress annotations

GKE exposes two distinct LB settings:
  • connectionDraining.drainingTimeoutSec (≈ 60 s, under the api-gateway server keep-alive of 70 s): socket reuse between client requests.
  • timeoutSec (≈ 300 s): kill an in-flight request after this much inactivity, so SSE / long LLM streams aren’t dropped.
The GKE HTTP(S) load balancer applies a default timeoutSec of 30 s on backends. SSE calls (e.g. messages) whose processing exceeds 30 s will be cut off before completion unless you raise this timeout via a BackendConfig attached to the prismeai-console and prismeai-api-gateway Services.
Create the BackendConfig through the chart’s extraObjects value:
Then attach it to the relevant Services via the cloud.google.com/backend-config annotation: at minimum prismeai-console and prismeai-api-gateway, since these are the entry points exposed through the LB:
The full annotation reference lives in Helm install, Ingress and load balancer.

​
Next Steps

Install with Helm

Configure values and deploy core + apps namespaces.

Databases

PostgreSQL or MongoDB, Redis, Elasticsearch or OpenSearch.

Install products

Fresh-install walkthrough.

Migration v27

Migrate an existing instance to v27.