OLIVE Platform Version 1.3.0
Status: Security release. Superseded by 1.3.1, which is the current release. This manual documents OLIVE 1.3.
This page records what 1.3.0 introduced, and it remains the substantive description of the 1.3 release line — the breaking changes, the security work and the migration from 1.2 all belong to it.
The current release is 1.3.1, a hotfix that rebuilds every container image on patched base images and fixes two defects. Deploy 1.3.1 rather than 1.3.0, and take the service, chart and image versions from the 1.3.1 page — the tables below are the 1.3.0 ones.
Overview
OLIVE 1.3.0 is primarily a security release. The platform was systematically reviewed for security weaknesses across its microservices, Helm charts, container images, configuration defaults and build pipeline. The serious findings are closed, and — beyond fixing individual defects — the way the platform is built and deployed changed: every component is pinned to an exact, verified version; the containers OLIVE produces are cryptographically signed; each tenant runs in its own isolated compartment; secrets are held in a dedicated vault with narrowly scoped access; and the platform now records the security-relevant events an operator needs in order to notice an attack.
Alongside the security work, the deployment model was simplified: the common assets tenant has been removed. Two operational additions round the release out: the workloads can now expose Prometheus metrics endpoints, and the volumes that hold tenant state are now labelled for backup so a backup tool can select them.
- No common assets tenant. A production installation is now one shared runtime pool plus one deployment per tenant.
- Internal services are no longer published by default. Services that only need to be reachable inside the cluster are not mapped to the ingress unless the deployment asks for it.
- The Temporal web UI is disabled by default.
- OpenSearch is no longer part of the platform.
- Editors can no longer import bundles, edit
core-tagged entries, or deploy pilets. These are admin operations in 1.3. - The AWS connectivity concepts are no longer supported.
AWSQuery/AWSQueryDatapointandAWSv4Authcannot be executed in 1.3 — see Deprecated connectivity concepts. - Upgrade in place is not supported for this release. Export from 1.2, redeploy, import. See Migration 1.2.0 to 1.3.0.
Deployment changes
Common assets tenant removed
OLIVE 1.3 has no common-assets deployment and no commonLibraryName value pointing at one. Reusable scenario and microfrontend assets are distributed with the normal import/export mechanism instead. The oliveCore flag remains for compatibility and should be left false.
NGINX Ingress support removed
NGINX Ingress was deprecated in 1.2 and is no longer supported in 1.3. Traefik is the only supported ingress controller: OLIVE renders Traefik router annotations and a chain of Middleware resources, and there is no NGINX equivalent in the charts. global.ingress.className is expected to name a Traefik IngressClass — RKE2's bundled Traefik and the upstream Traefik Helm chart installed as release traefik both answer to traefik. A cluster that runs NGINX Ingress for other workloads has to add Traefik for the OLIVE hosts. See Cluster Operator → Ingress Requirements.
The handbooks ship inside the foundational bundle
The Connectivity Developer Handbook and the Platform Administrator and Operator Manual are now packaged within the foundational OLIVE Connectivity Bundle. Setting up a tenant is one import, not three — 1.2.0 required the two handbooks as separate bundles afterwards. They can still be imported individually when a handbook has been updated since the foundational bundle you installed. See Import Required Foundational OLIVE Connectivity Bundle.
OpenSearch removed
OpenSearch is no longer part of the platform. A tenant's stateful components are now CouchDB and SeaweedFS. The services.stateful.opensearch key still exists in values.yaml, but no chart is behind it — it deploys nothing and can be removed from your values files. If you are coming from 1.2, the tenant's OpenSearch volume and its Vault credentials are left behind by the teardown; see Uninstall and Teardown.
Deprecated connectivity concepts
The AWS-specific metamodel concepts are no longer supported in 1.3:
AWSQuery/AWSQueryDatapoint— the AWS service query datapointAWSv4Auth— AWS Signature Version 4 authentication
The aws-query-caller extension service that executed them is not part of the release, so instances of these types can no longer be invoked.
See the Connectivity Developer Handbook, chapter Metamodel, for the current list of supported types.
Namespace and tenant name are now independent
In 1.2, global.namespace had to carry the same value as global.tenant. In 1.3 the two are separate: the tenant keeps its short, permanent identifier — which builds its DNS subdomain, appears in log records, and names its Vault and Temporal accounts — while the namespace can carry a prefix of your choosing, for example oli-<tenant>, to give the cluster more structure.
Metrics endpoints
OLIVE's workloads can now expose Prometheus metrics endpoints — the Express services, CouchDB, SeaweedFS, Temporal and the CloudNativePG database. They are off by default and switched on per namespace with global.metrics.enabled.
OLIVE's part is deliberately narrow: it exposes the port, names it, and renders the NetworkPolicy rule that lets a scraper reach it. It does not scrape, store, visualise or alert, and it does not create the ServiceMonitor or PodMonitor objects that tell an agent what to collect — those belong to whoever runs the observability stack.
global.metrics.scrape is mandatory once metrics are on. The charts fail the Helm render rather than open a port with an empty namespaceSelector, which would mean every namespace in the cluster.
They carry no TLS, no token and no authorization check: any workload that can reach the port can read them. The NetworkPolicy rendered alongside the port is the only access control there is.
Do not enable metrics in a namespace where global.networkPolicy.enabled is false.
See Metrics Endpoints for the values and the per-component scrape details an operator needs.
Backup labels on the data volumes
The couchdb and seaweedfs volume claims now carry a configurable label, so a backup tool can select the volumes that hold state without anyone maintaining a list of claim names.
OLIVE marks what is worth backing up; which tool performs the backup is your decision. global.volumeLabels controls it: unset gives the default olive-boc/backup: velero, a map replaces that default entirely, and {} means no labels at all.
The labels are part of the StatefulSet's volumeClaimTemplates and cannot be changed afterwards — a release has to keep rendering the set it was installed with, or every sync fails with spec: Forbidden. Changing it later is a data migration, not a values change.
What to Back Up covers which volumes actually matter and which of them have to be captured together.
New and changed values
| Value | Purpose |
|---|---|
global.urlScheme | Scheme the platform uses when it builds its own URLs |
global.keycloakBaseURL | Origin of the Keycloak instance |
global.keycloakVaultPath | Path of the Keycloak admin credential in Vault, relative to the KV mount |
global.networkPolicy.* | Master switch and ingress-controller selector for default-deny network policies |
global.serviceMesh.* | none / sidecar / ambient, plus vaultMeshed and the ambient-specific settings |
global.serviceAccounts.mode | split (default) or legacy |
global.keycloakIpBlocks | Narrows the Keycloak JWKS egress rule. Empty (the default) allows any public host on 443 with private ranges carved out; set it to the VPC CIDR when Keycloak is behind an internal load balancer |
global.nonRoutableCidrs | Ranges carved out of every internet egress rule |
global.vault.* | Where the platform finds HashiCorp Vault |
tenant-setup.imageVerification.mode | disabled / audit / enforce — image signature verification |
tenant-setup.imageRegistry.mode | disabled / audit / enforce — registry allow-list |
couchdb.credentials.rotationId, seaweedfs.credentials.rotationId | Request new generated storage credentials |
global.metrics.enabled, global.metrics.scrape.* | Expose the Prometheus metrics endpoints, and which scrape agent may reach them |
global.volumeLabels | Labels on the data volume claims, for selecting them with a backup tool |
global.storageClassName | StorageClass for every volume in the namespace |
Full examples: Chart Values Reference.
Security changes
Reduced attack surface
- Internal services are no longer published on the ingress unless the deployment explicitly asks for it. The Temporal web UI is disabled by default; the SeaweedFS filer ports were reduced; the cloud-function registry is no longer publicly accessible.
- A central egress guard rejects outbound requests to internal address ranges, Kubernetes-internal service names and link-local metadata addresses before the request is made. It applies to Data Orchestrator datapoints, OpenAPI import, image fetching and OAuth/OIDC configuration, and emits
OLIFETCH1when it blocks one. - Placeholder injection prevention: strict character allow lists per HTTP position (URL host, port, method, query key and value, header value, form body). Header names are no longer substitutable.
- Input validation and explicit limits throughout: length and format constraints, recursion depth, promise fan-out, uncompressed archive size, entry counts, request sizes, and constrained JSONata expressions.
- Less information given away: normalised error messages, no stack traces in client responses, no internal cluster URLs in debug output, the
exportAllroute removed, and Data Orchestrator diagnostics disabled in production deployments. - Browser-side protections reviewed per service: CSP for the Editor and documentation routes, corrected CORS, CSRF protection by default, per-popup nonces, and production webpack builds for the Editor.
Identity, access and tenant boundaries
- Roles are genuinely separated.
user,editorandadminpreviously mapped to the same underlying Vault permissions. They no longer do: an administrator can perform administrative functions but not ordinary editing functions, andrequireRoleCheckis the single enforcement point, emitting events on both success and denial for privileged operations. - Three operations moved from
editortoadmin. Importing an OLIVE Connectivity Bundle; editing connectivity entries taggedcore(the foundational-package entries, visible in the editor only withdebug=true, which govern how OLIVE's own services behave); and deploying or changing a pilet. Development deployments have no roles and are unaffected. - A logout path was added.
- Per-user authorization on addressable resources. Job polling,
setResponse, pause, cancel and resume now verify an HMAC-derived token identifying the triggering user. Session memory generates a secure random session ID server-side instead of trusting a client-supplied one, and sensitive entries are bound to the creating user. File storage separates read and write paths so system files cannot be overwritten, and validates file names, extensions and MIME types. Uploads now carry a category —core,staticortemp— which decides who may overwrite the file:staticrequires the editor role,corethe admin role, and an upload with no category is stored astemp, writable by any user of the tenant. - Microfrontends can no longer call every service directly. A pilet rendered by the microfrontend-interpreter may reach the local paths
/data-orchestrator/and/repository/get, plus the publicfile-storage-serviceAPI with GET only; uploads and every other method go through the orchestrator, which applies the caller's permissions. Existing UI components that called services directly stop working — expect this to surface as a user-reported breakage rather than as a deployment error. - Tenant isolation in four independent layers: the JWT claim check in every service, the per-tenant Vault role, the per-tenant namespace and storage, and default-deny network policies.
- Tenant ID is now mandatory log metadata on every record.
- Tenant context propagation through
AsyncLocalStoragewas rewritten so it can no longer mix or overwrite context under concurrency.
Secrets
- Four purpose-scoped ServiceAccounts per tenant, each bound to exactly one Vault role, replacing the single shared identity whose four role bindings collapsed the Vault policy separation into its union.
- The Keycloak master credential is reachable only by the realm-import job.
- Workload identities have no Kubernetes API rights;
automountServiceAccountToken: falseon all four. - Main tenant credential TTL reduced from 1 h / 24 h to 15 min / 1 h.
- Infrastructure and application secrets no longer share a path, role or policy.
- Root tokens and unseal keys are no longer written to logs.
- Credentials, tokens, secrets and request or response bodies are not logged in production; a separate developer log stream covers development environments only. Security events do record the caller's source IP address, user name and tenant — deliberately, since an audit trail that cannot attribute an action is of no use.
Platform hardening
- Distroless base images where practical; slim or officially maintained images otherwise.
- Every chart hardened: AppArmor,
seccomp: RuntimeDefault, all capabilities dropped, read-only root filesystem,allowPrivilegeEscalation: false, non-root execution, resource limits. - No package installation at runtime — the jobs that used to install
curl,jq,coreutilsandgettextat startup now run from purpose-built images. - Pod Security Admission labels, enforcing
baselinewithrestrictedaudited and warned. - Kyverno support via the
kyverno-infrachart, with image signature verification (disabled/audit/enforce), an example registry allow-list, and support for operator-written policies applied selectively by namespace label.
Supply chain
- All npm packages pinned to exact versions with lockfile integrity; a seven-day cooldown before adopting a newly published third-party version.
- All container images — first- and third-party — pinned by digest rather than tag.
- Images OLIVE builds are signed with cosign, with signatures and attestations stored as OCI artefacts.
- SBOM per release, covering the npm dependency tree of every service and the contents of every container image.
- The release build is gated on dependency and image scanning.
Known gap: Helm chart dependencies cannot currently be pinned by OCI digest.
Security observability
- A defined catalogue of security detection points with stable event codes (
OLIAUTHN*,OLIAUTHZ*,OLIFETCH*,OLIERROR*,OLIFILE*,OLIORCH*,OLISEC*,OLIMISC*), so alert rules can be written against codes rather than message text. See Log Events and Automated Reactions. - Each event carries event code, source IP, user, tenant ID, service name and version, description and request ID, with a traceable identifier propagated across services.
- Log streams separated by a
log_typelabel —app,security,audit_ai— so that retention and access rights can be set per stream in whatever log platform the operator runs. OLIVE labels the logs; it does not provide or configure an observability stack. - Keycloak audit logs can be emitted in JSON and ingested into the log platform. Successful logins are covered this way, because they complete in Keycloak rather than in OLIVE.
- Log level and truncation changes are restricted and themselves logged.
- Readiness and liveness probes now reflect real dependency health rather than only process liveness.
Encryption in transit
- Optional mutual TLS between services via an Istio ambient mode service mesh, switchable at deployment time. A per-node ztunnel DaemonSet provides workload identity, mutual authentication and encryption without a per-pod sidecar, and stays transparent to the OLIVE charts.
- Kubernetes NetworkPolicies continue to be enforced under ambient mode; the HBONE port (TCP 15008) is accounted for by the charts.
Compatibility Matrix
These are the third-party component versions OLIVE 1.3.0 was tested against. Share this with your cluster operator before the first install — the versions in the deployment have to match what the release was validated on.
| Component | Tested version | Notes |
|---|---|---|
| Kubernetes | 1.31 or later required | securityContext.appArmorProfile is GA from 1.31 |
| RKE2 | v1.33.8+rke2r1 | The distribution OLIVE is tested on. Other CNCF-compliant clusters may work with extra validation |
| HashiCorp Vault | 2.0.4 | Required. Via vault-infra 0.7.14, which wraps upstream chart 0.33.0 |
| Vault Secrets Operator | 1.5.1 | Exactly one instance per cluster |
| Keycloak | 26.7.5 | Required for authenticated deployments. Via keycloak-infra 1.4.9 (codecentric chart 7.2.3) |
| CloudNativePG (CNPG) operator | 1.27.1 | Provides the workflow engine's PostgreSQL, and Keycloak's |
| Temporal | 1.31.2 | Server and admin-tools; UI 2.53.3, development only |
| Traefik | Managed via RKE2, or installed separately | The only supported ingress controller from 1.3 |
| cert-manager | 1.18.2 | OLIVE creates the Certificate; the ClusterIssuer is the cluster operator's |
| OAuth2 Proxy | 7.15.3 | |
| Redis | 7.4.11-bookworm | Session storage for OAuth2 Proxy |
| CouchDB | 3.5.2.1 | Per tenant |
| SeaweedFS | 4.44 | Per tenant |
| Longhorn | 1.8.0 | The storage tested on RKE2. Any RWO StorageClass works |
| Istio (ambient mode) | Cluster-provided | Optional. Configured by service-mesh-setup 0.5.3 |
| Kyverno | 1.19.0 (upstream chart 3.9.0) | Optional. Via kyverno-infra 1.0.10. Uses policies.kyverno.io/v1 types, not the legacy ClusterPolicy |
| Node.js runtime | gcr.io/distroless/nodejs24-debian13 (nonroot) | Base image for the Express services |
| Python runtime | gcr.io/distroless/python3-debian13 (nonroot) | Base image for the auth-proxy session-purge job |
The authoritative image references for any deployment are the digests pinned in that release's chart values, not the versions above — see Container images.
Keycloak, Vault and the Vault Secrets Operator are infrastructure OLIVE integrates with rather than parts of the platform, so their versions are the operator's to manage and may be moved within the same major version — for a published CVE, for instance — without waiting for an OLIVE release. The components that ship inside the OLIVE charts, including Redis, CouchDB, SeaweedFS and Temporal, are tested as the combination above; if one of them needs a newer version, raise it with olive-support@boc-group.com rather than overriding it. See Who owns which version.
Service Versions
Core Services
| Service Name | Version |
|---|---|
| cloud-function-registry | 2.2.0 |
| instantiator | 1.2.0 |
| data-orchestrator | 3.4.0 |
| file-storage-service | 1.3.0 |
| import-export | 1.2.0 |
| landing-service | 0.3.0 |
| ui-component-registry | 2.3.0 |
| microfrontend-interpreter | 2.3.0 |
| editor | 3.4.0 |
| repository | 1.5.0 |
| scheduler | 1.2.0 |
| session-memory | 1.2.0 |
| template-data-retriever | 0.2.0 |
Extension Services
| Service Name | Version |
|---|---|
| agent-caller | 2.2.0 |
| mcp-server-caller | 0.2.0 |
| prompt-caller | 2.2.0 |
| temporal-caller | 0.2.0 |
| parse-openapi | 0.1.0 |
| develop-support | 0.2.0 |
QA Services
| Service Name | Version |
|---|---|
| ping | 1.2.0 |
| simple-adder | 1.2.0 |
| auth-tester | 0.1.0 |
Infrastructure Charts
| Chart | Version |
|---|---|
| auth-proxy | 0.7.9 |
| certificate | 1.3.3 |
| couchdb | 1.10.2 |
| keycloak-setup | 1.7.5 |
| seaweedfs | 1.4.1 |
| tenant-setup | 1.6.5 |
| vault-secrets-setup | 1.8.9 |
| workflow-engine-postgres | 1.4.1 |
| express (shared library chart behind the Express services) | 2.5.1 |
| schedule-trigger-cron-job | 0.5.0 |
| custom-ca | 0.1.7 |
| utilities (shared Helm library) | 1.11.2 |
adoxx-embedder, aws-query-caller, cloud-function-executor, embed-indexer, transcript-to-txt, vector-service and the opensearch chart are not part of 1.3. If your values files still name them, remove those entries.
Known gaps
| Gap | Status |
|---|---|
| Helm chart dependencies cannot be pinned by OCI digest | A limitation of OCI chart references rather than of OLIVE. Mitigated by the charts residing in an OCI registry under OLIVE's control |
| Malware scanning for uploaded files | File type, extension and size validation are in place; no virus scanner is applied |
| Encryption at rest is not provided by OLIVE | OLIVE encrypts nothing at rest itself. Every data store — each tenant's CouchDB, SeaweedFS object storage, Temporal's PostgreSQL and Keycloak's PostgreSQL — writes to PersistentVolumes, so whether that data is encrypted comes from the StorageClass, the CSI driver and the cloud KMS behind them. That is a cluster-operator responsibility, not a chart value. Vault is the exception: it encrypts its own storage backend by design |
| Detection logic (anomaly detection, per-tenant baselines) | Deliberately not built into the product; an operator capability |
OLIFILE3 (path traversal detection point) | Prevention is in place; the detection event is not yet emitted |
| High-availability behaviour is not characterised | The platform runs on a multi-node cluster, but OLIVE does not yet document what a tenant experiences during a node drain or a node loss — which services have a replica elsewhere, which do not, and how long a tenant is unavailable. No PodDisruptionBudget is created. Until this is measured, plan maintenance windows rather than relying on transparent failover |
| Growing a tenant's storage is disruptive | A volume's size lives in the StatefulSet's immutable volumeClaimTemplates, so enlarging one means deleting the StatefulSet with --cascade=orphan, patching the claim, and letting the chart recreate it. That is the supported procedure today, and it needs cluster rights a platform administrator does not have. Size volumes before the first deployment where you can. See Sizing the volumes |
| No sized reference architecture | The hardware requirements give a floor for a non-HA installation, not a capacity model. There is no tested profile stating how many tenants or how much connectivity activity a given cluster carries |
| Volume snapshots are not application-consistent | OLIVE has no hook that pauses writes, flushes state to disk and signals "safe to snapshot now". A snapshot of a running deployment therefore captures the disk as it would look after a power cut, and recovery relies on each data store's own crash handling. The OLIVE Connectivity Bundle export is the consistent-by-construction alternative, for connectivity content only. See What to Back Up |
Upgrade
Do not upgrade in place. Export from 1.2, deploy 1.3 fresh, import. See Migration 1.2.0 to 1.3.0.
Where to read more
- Security Configuration — what a platform administrator sets in the deployment values
- Cluster Operator → Security — what the cluster must provide
- Chart Values Reference — the 1.3 values examples
- Metrics Endpoints — enabling them, and what an operator needs in order to scrape
- Cluster Operator → Monitoring and Observability — the scrape agent and store, which are the cluster operator's, and who owns which dashboards
- What to Back Up — where the state is, what must be consistent with what, and the restore order