CRD ownership
Use the Envoy Gateway version and Gateway API bundle required by your selected Formbricks chart. Check that release’sChart.yaml and CRD ownership guide; the links below show the current source,
so select your chart’s publication commit before copying examples.
Choose who owns the two cluster-wide CRD sets before enabling Envoy:
For a chart-managed Envoy controller, configure a dedicated Redis/Valkey rate-limit backend:
maxmemory-policy to noeviction and configure
authentication/TLS as needed.
This keeps rate-limiting state separate from the application’s own Redis traffic.
envoy.crds.enabled=false disables both Gateway API and Envoy Gateway CRDs, plus the bundled Gateway
API safe-upgrade admission resources. A platform that manages only Gateway API must install matching Envoy
Gateway CRDs separately at the chart-required version. After the platform has established compatible
Gateway API CRDs, install only the Envoy CRDs:
crds.gatewayAPI.enabled=true, crds.gatewayAPI.channel=standard, and
crds.envoyGateway.enabled=true for a fresh cluster. During a handoff from bundled Experimental CRDs,
keep crds.gatewayAPI.channel=experimental until clients using Experimental-only versions, such as
TLSRoute v1alpha2, have migrated; a Standard bundle may not serve them. Do not apply the
both-sets variant over provider-owned Gateway API CRDs. The Gateway API bundle must serve the types Envoy
needs, including BackendTLSPolicy and ListenerSet. The OCI install does not
place chart scripts in your working directory. Use the immutable source commit from the
chart publication run
for your installed chart version, then run the optional, read-only preflight after the CRD owner provisions the CRDs:
my-cluster with the target kubectl context. The script checks Gateway API version/channel, served
APIs, and Envoy CRD schemas against the selected release without changing the cluster. Envoy CRDs have no release
annotation, so their owner must record and pin the matching package. It is separate from Helm rendering, so offline and GitOps
workflows do not need Kubernetes access to render the chart.
For Helm, apply externally owned CRDs first, wait for Established, run preflight, then install or
upgrade Formbricks using the matching values example. Upgrade CRDs through their owner before a later
Envoy controller upgrade; Helm does not upgrade or delete CRDs
in crds/.
For Argo CD, keep external CRDs in a platform Application, wait for Established, run preflight,
then sync Formbricks with envoy.crds.enabled=false. For Flux, use a separate CRD Kustomization
or release, wait for Established, run preflight, then reconcile Formbricks after its Ready condition.
A global Argo CD spec.source.helm.skipCrds or Flux spec.install.crds: Skip or
spec.upgrade.crds: Skip also skips the bundled SpiceDB operator CRD; use these only when all required
CRDs are managed separately. Apply the same ownership order during upgrades and keep externally owned
resources outside Formbricks pruning.
A shared controller needs a compatible GatewayClass whose controllerName belongs to that controller and
whose proxy settings and Redis rate-limit backend support the Formbricks routes and policies. Configure
envoy.formbricks.ingress or your ingress/load balancer to send public traffic through Envoy. Other Gateway
API implementations do not reconcile Envoy-specific SecurityPolicy, HTTPRouteFilter, or EnvoyProxy
resources. See the chart ownership guide
for the installation matrix and the rate-limiting guide for
route coverage.
When handing off to a shared controller, its proxy Service may live in the controller namespace. Update
the public ingress target to the new proxy after the Gateway is Programmed and a request succeeds. Keep
the shared GatewayClass and any class-level EnvoyProxy under platform ownership so they survive a
Formbricks uninstall.