Skip to main content
Version: v5.3

Release 5.2.0

Migrating from Ingress to the Gateway API​

Release 5.2 replaces all Kubernetes Ingress resources with the Gateway API. Every per-service ingress template is now an HTTPRoute, and inbound traffic is served through a shared gateway instead of an nginx/traefik ingress controller.

This introduces a new prerequisite. Gateway API is not part of Kubernetes; it is a set of CRDs installed separately and versioned independently of the cluster, so the Kubernetes version alone does not determine whether the chart will install.

Release 5.2 requires:

  • Gateway API CRDs at v1.5.0 or later, from the standard release channel. The experimental channel is not required. Install the bundle with kubectl apply --server-side -f https://github.com/kubernetes-sigs/gateway-api/releases/download/v1.5.1/standard-install.yaml, adjusting the version as needed.
  • A Gateway API implementation that supports both the ListenerSet resource and the HTTPRoute CORS filter. Support for these two features varies by implementation and lags the upstream release, so confirm them against the Gateway API conformance reports rather than assuming Gateway API support is sufficient. Roughly half of the conformant implementations do not support ListenerSet.

The reference deployment uses Istio 1.30.2 with Gateway API v1.5.1 CRDs. Istio 1.30 or later is conformant for both required features.

Both features reached the Gateway API standard channel in v1.5.0. Earlier CRD bundles, including v1.4.x, do not serve them, and installing against those versions will fail.

The routing changes themselves are:

  1. Each HTTPRoute attaches to a per-namespace ListenerSet (named <namespace>-listeners) that contributes the TLS listeners for the release's hostnames. The ListenerSet in turn attaches to the shared gateway, so each release owns its own listeners and TLS certificates.
  2. A new gatewayAPI block has been added to values.yaml. It sets the parent reference for all routes; listenerSetName defaults to <namespace>-listeners when left empty.
  3. The ingress.<service>.className and ingress.<service>.secretName settings are no longer used for routing. The ingress.<service>.enabled flags are retained and continue to control whether each route is created.
  4. TLS is now terminated by the listener using a certificate in the release namespace; the per-service TLS secrets referenced by the old ingresses are no longer required.

CORS Moved off Ingress Annotations​

CORS was previously configured through nginx ingress annotations, which no longer apply under the Gateway API. Two replacements have been added:

  1. Browser-facing CORS for the web application host is now applied through the standard Gateway API HTTPRoute CORS filter, configured under a new ingress.staticwebsvc.cors block in values.yaml (enabled, allowOrigins, allowMethods, allowHeaders, allowCredentials, maxAge).

    Origins are matched as full origins, not regular expressions. Each entry in allowOrigins must be a bare * or <scheme>://<host>[:port] with at most one leading *. wildcard label, for example https://*.example.com. The wildcard is greedy to the left, so https://*.example.com matches both app.example.com and app.eu.example.com. A * entry cannot be combined with any other origin. Values that are not valid origins are rejected by the chart's values schema at template time.

    maxAge sets Access-Control-Max-Age in seconds and defaults to 600. Browsers cap this value independently (Chromium at 600 seconds, Firefox at 86400), so larger values are not honored everywhere.

  2. CORS for file downloads is applied at the CDN. Allowed origins default to the application host and can be extended for additional origins (for example, local development).

Restricting Platform IAM Administrative Paths​

The Platform IAM route now blocks external access to the IAM administrative console and the master realm. Requests to /admin and /realms/master are routed to a dedicated no-op service (platformiamsvc-blackhole), while all other paths continue to reach platformiamsvc. The blackhole service is created automatically with the Platform IAM route.

New Required Secret for Platform Notification Service​

A new environment variable has been added to the platform-notificationsvc-external secret. It has no default and must be generated and provisioned before deploying.

  • POSTGRES_PLATFORM_NOTIFICATIONSVC_PASSWORD - Password for the platform notification service user in PostgreSQL (also included in platform-kafka-connect-external)

Secret Definition Changes​

  1. SISENSE_SECRET_KEY has been renamed to SISENSE_SHARED_KEY. Customers using Sisense SSO must rename this variable in the datasourcesvc-external secret.
  2. SMTP_DEFAULT_PASSWORD is now also required by the notificationsvc-external and platform-notificationsvc-external secrets, in addition to platformiamsvc-external.

Workflow Service Logging Configuration​

A required logging.level setting has been added to the workflowsvc-api, workflowsvc-backend, and workflowwkr-backend configuration blocks in values.yaml, defaulting to INFO. Existing overrides for these services must include a logging.level value.

Script Worker Execution Timeout​

A new config.internal.scriptworker.script_execution_timeout_ms setting has been added, defaulting to 7200000 (2 hours), to bound the run time of individual script executions.

Cloud-Specific Topology Constraint Made Optional​

The global topology spread constraints included a third constraint on the OCI fault domain label (oci.oraclecloud.com/fault-domain), applied to every multi-replica workload regardless of where the Platform was installed. On non-OCI clusters the label does not exist, so the constraint was ignored, but it was still emitted.

A new topologySpreadConstraints.faultDomainKey setting replaces the hardcoded label and defaults to empty, omitting the constraint. Set it to oci.oraclecloud.com/fault-domain to restore the previous behavior on OKE. The constraints on kubernetes.io/hostname and topology.kubernetes.io/zone are unchanged.

Optional Creation of Shared Volume Claims​

The staticwebsvc and graphicssvc deployments mount shared ReadWriteMany volumes, and the chart has always expected those claims to exist before installation. This was not documented and produced pods stuck in Pending when they were missing.

A new persistentVolumeClaims block can create the claims from the chart. It is disabled by default, so the existing arrangement of creating them externally is unchanged. Each entry accepts create, size, accessModes, storageClassName, volumeName, and annotations.

Removed Unused Ingress Settings​

The ingress.<service>.className and ingress.<service>.secretName settings have been removed. They were left over from the nginx Ingress resources replaced in this release and were read by no template. The ingress.<service>.enabled, annotations, and labels settings are unchanged.

ServiceMonitor Settings Moved and Made Optional​

The ServiceMonitor resources created for the scriptworker and workflowwkr-backend autoscalers use the prometheus-operator ServiceMonitor CRD, which is not part of Kubernetes. They were previously created whenever the corresponding autoscaler was enabled, with no way to turn them off, so enabling autoscaling on a cluster without prometheus-operator installed caused the entire install to fail.

  1. A new top-level serviceMonitor block has been added to values.yaml, with enabled, namespace, labels, and interval. Set serviceMonitor.enabled to false on clusters without prometheus-operator, or where metrics are collected another way. Autoscaling is unaffected by this setting.
  2. The per-service horizontalpodautoscaler.<service>.prometheusNamespace, prometheusRelease, and prometheusInterval settings have been removed and replaced by the serviceMonitor block, which applies to both resources.
  3. Prometheus discovery labels are now an arbitrary map at serviceMonitor.labels rather than a single hardcoded release label, so the chart can target Prometheus deployments that select on other labels. Note that Helm merges maps, so removing the default label requires setting release: null in an override file.
  4. serviceMonitor.namespace may now be left empty to create the resources in the release namespace, which avoids requiring write access to another namespace.

Defaults are unchanged from 5.1, so an upgrade that does not set these values produces the same resources as before.

Pod Labels​

A podLabels setting has been added to every deployment, statefulset, and job in the chart, allowing custom labels to be applied to pods independently of resource labels. It defaults to empty and requires no action.