Introducing KubeVault v2026.11.15

KubeVault is a Kubernetes operator for OpenBao . OpenBao is an open source tool for secrets management, encryption as a service, and privileged access management. The KubeVault operator makes it easy to deploy, maintain and manage OpenBao servers in Kubernetes. It also manages secret engines, auth methods and policies in a Kubernetes-native way.

We are excited to announce the release of KubeVault v2026.11.15 . The main change in this release is nine new storage backends for VaultServer. It also adds support for Sigilr 2.6.5.1. You can find the full commit-by-commit changelog here .


Quick highlights

  • VaultServer now supports nine more storage backends: Aerospike, Alibaba Cloud OSS, Cassandra, CockroachDB, CouchDB, Google Cloud Spanner, Microsoft SQL Server (MSSQL), OCI Object Storage and ZooKeeper.
  • Most of the database backends can point at a KubeDB-managed database through databaseRef. KubeVault reads the address, and where available the TLS material, from the database’s AppBinding.
  • Credentials and TLS material always come from Kubernetes Secrets. They are never written into the VaultServer spec.
  • The admission webhook checks HA capability for every new backend, so spec.exposePrimary is only accepted on a backend that can actually elect a leader.
  • Added support for Sigilr 2.6.5.1.

New Version Support

KubeVault now supports Sigilr 2.6.5.1, which is built on OpenBao 2.6.5. To deploy a VaultServer with it, set spec.version to sigilr-2.6.5.1:

apiVersion: kubevault.com/v1alpha2
kind: VaultServer
metadata:
  name: vault
  namespace: demo
spec:
  replicas: 3
  version: sigilr-2.6.5.1
  backend:
    raft:
      storage:
        storageClassName: "standard"
        resources:
          requests:
            storage: 1Gi
  unsealer:
    secretShares: 5
    secretThreshold: 3
    mode:
      kubernetesSecret:
        secretName: vault-keys
  terminationPolicy: WipeOut

Nine New Storage Backends

Before this release, a VaultServer could store its data in Raft, Consul, etcd, PostgreSQL, MySQL, DynamoDB, GCS, S3, Azure, Swift, the local filesystem or in memory. This release adds nine more under spec.backend:

Backendspec.backend fieldHigh availability
AerospikeaerospikeNo
Alibaba Cloud OSSalicloudossNo
Apache CassandracassandraNo
Apache CouchDBcouchdbNo
Apache ZooKeeperzookeeperAlways
CockroachDBcockroachdbWith haEnabled: true
Google Cloud SpannerspannerWith haEnabled: true
Microsoft SQL ServermssqlNo
OCI Object StorageociWith haEnabled: true and lockBucketName

A few things apply to all of them:

  • Secrets stay in Secrets. Usernames, passwords, service account keys and API keys are read from the Secret named in credentialSecretRef (or authInfoSecretRef for ZooKeeper). Depending on how the underlying storage backend reads its configuration, KubeVault either passes them as environment variables (Alibaba Cloud OSS, CouchDB, OCI, Spanner) or writes them into the generated storage config (Aerospike, Cassandra, CockroachDB, MSSQL, ZooKeeper).
  • KubeDB integration. When databaseRef points at a KubeDB AppBinding, KubeVault generates the host address from it. For Cassandra, CockroachDB and MSSQL it can also take the CA bundle (and client certificate, where supported) from the AppBinding when you don’t set tlsSecretRef.
  • Pre-created storage. Some backends expect the database, keyspace, namespace or bucket to exist already. The notes under each example say which.

Below is a short example for each backend. Each one shows only spec.backend; the rest of the VaultServer stays the same. All examples use the demo namespace and Sigilr 2.6.5.1. Here is a complete manifest, using CockroachDB as the backend:

apiVersion: kubevault.com/v1alpha2
kind: VaultServer
metadata:
  name: vault
  namespace: demo
spec:
  replicas: 3
  version: sigilr-2.6.5.1
  backend:
    cockroachdb:
      databaseRef:
        name: cockroachdb
        namespace: demo
      database: vault
      sslMode: verify-full
      haEnabled: true
  unsealer:
    secretShares: 5
    secretThreshold: 3
    mode:
      kubernetesSecret:
        secretName: vault-keys
  terminationPolicy: WipeOut

Aerospike

The Aerospike namespace must already exist on the server.

  backend:
    aerospike:
      hostList: aerospike.demo.svc:3000
      namespace: vault
      set: vault-data
      credentialSecretRef:
        name: aerospike-cred   # username, password
      authMode: INTERNAL

Use hostname/port for a single seed node, or databaseRef for a KubeDB-managed Aerospike. clusterName, timeout and idleTimeout are also available.

Alibaba Cloud OSS

The bucket must already exist, and credentialSecretRef is required.

  backend:
    alicloudoss:
      endpoint: https://oss-us-east-1.aliyuncs.com
      bucket: vault-data
      credentialSecretRef:
        name: alicloudoss-cred   # access_key, secret_key

The Secret holds access_key and secret_key.

Cassandra

The keyspace and table must already exist. protocolVersion must be "2" or higher for username/password authentication.

  backend:
    cassandra:
      databaseRef:
        name: cassandra
        namespace: demo
      keyspace: vault
      table: entries
      consistency: LOCAL_QUORUM
      protocolVersion: "4"
      credentialSecretRef:
        name: cassandra-cred   # username, password
      tlsEnabled: true

Without a databaseRef, list the nodes in hosts as a comma-separated string. For TLS, tlsSecretRef points at a Secret with a pem_bundle key; tlsServerName lets you verify nodes against a shared DNS name instead of their IPs.

CockroachDB

CockroachDB uses the PostgreSQL wire protocol, so the spec looks a lot like the existing postgresql backend. The database must already exist; OpenBao creates the tables. CockroachDB supports HA, so you can set haEnabled: true and use spec.exposePrimary.

  backend:
    cockroachdb:
      databaseRef:
        name: cockroachdb
        namespace: demo
      database: vault
      sslMode: verify-full
      haEnabled: true

With databaseRef, KubeVault reads the host and the credentials from the database’s AppBinding. Without it, set address to the CockroachDB host and point credentialSecretRef at a Secret holding username and password. That Secret can instead hold a full connection_url, which replaces address, database and sslMode.

CouchDB

endpoint is the full URL of the database, including its name. The database must already exist.

  backend:
    couchdb:
      endpoint: http://couchdb.demo.svc:5984/vault
      credentialSecretRef:
        name: couchdb-cred   # username, password

With databaseRef, KubeVault builds the endpoint as <appbinding url>/<database>, so set database instead of endpoint.

Google Cloud Spanner

database is the full Spanner database name. Credentials come from a service account key stored under sa.json. If you leave out credentialSecretRef, Google’s default credentials are used instead, for example Workload Identity on GKE. The service account needs the roles/spanner.databaseUser role on the database.

  backend:
    spanner:
      database: projects/<project-id>/instances/<instance-id>/databases/vault
      haEnabled: true
      credentialSecretRef:
        name: spanner-cred   # sa.json

OpenBao does not create tables, so create them before deploying. The data table is named Vault by default (set table to change it), and with haEnabled: true OpenBao also needs a leader-election table, named after table with HA appended (set haTable to change it):

CREATE TABLE Vault (
  Key   STRING(MAX) NOT NULL,
  Value BYTES(MAX),
) PRIMARY KEY (Key);

CREATE TABLE VaultHA (
  Key       STRING(MAX) NOT NULL,
  Value     STRING(MAX),
  Identity  STRING(MAX),
  Timestamp TIMESTAMP,
) PRIMARY KEY (Key);

Microsoft SQL Server

OpenBao creates the database, schema and table if they are missing, provided the user has permission to do so.

  backend:
    mssql:
      databaseRef:
        name: mssqlserver
        namespace: demo
      database: vault
      schema: dbo
      table: vault
      credentialSecretRef:
        name: mssql-cred   # username, password
      encrypt: "true"

encrypt accepts true, false, strict (TDS 8.0, SQL Server 2022+) or disable. If you leave it unset, KubeVault turns encryption on whenever a CA is available, either from tlsSecretRef or from the databaseRef AppBinding.

OCI Object Storage

Both buckets must already exist. By default the backend authenticates with instance principals, which works when OpenBao runs on OCI compute and needs no Secret. To use an API key instead, set authTypeAPIKey: true:

  backend:
    oci:
      bucketName: vault-data
      namespaceName: <object-storage-namespace>
      region: us-ashburn-1
      authTypeAPIKey: true
      credentialSecretRef:
        name: oci-cred   # config, oci_api_key.pem
      haEnabled: true
      lockBucketName: vault-lock

Every key of the Secret is mounted at /etc/vault/storage/oci/.oci. The config file is in the usual ~/.oci/config format, and its key_file must point at the full mounted path, for example key_file=/etc/vault/storage/oci/.oci/oci_api_key.pem. HA needs a second bucket, set in lockBucketName.

ZooKeeper

ZooKeeper is always HA-capable, just like Raft and Consul.

  backend:
    zookeeper:
      address: zookeeper.demo.svc:2181
      path: vault/
      znodeOwner: digest:vault:<hashed-password>
      authInfoSecretRef:
        name: zookeeper-auth   # authInfo

The zookeeper-auth Secret holds an authInfo key with a scheme:auth pair (for example digest:vault:<password>). For TLS, set tlsEnabled: true and point tlsSecretRef at a Secret containing ca.crt, tls.crt and tls.key.


HA Validation for exposePrimary

spec.exposePrimary sends traffic only to the active OpenBao node, so it only works on a storage backend with HA support. The admission webhook now knows about all nine new backends:

  • Aerospike, Alibaba Cloud OSS, Cassandra, CouchDB and MSSQL are never HA-capable, so the webhook rejects exposePrimary with any of them.
  • CockroachDB and Spanner are HA-capable when haEnabled: true. OCI Object Storage also needs lockBucketName.
  • ZooKeeper is always HA-capable, like Raft and Consul.

The webhook rejects an invalid combination when you apply the VaultServer, so you don’t find out later from a cluster that never elects a primary.


What’s Next?

Please try the latest release and give us your valuable feedback.

  • To install KubeVault, follow the installation instructions here .
  • To upgrade an existing installation, follow the upgrade guide here .

Support

To speak with us, please leave a message on our website .

To receive product announcements, follow us on Twitter .

If you have found a bug with KubeVault or want to request new features, please file an issue .


TAGS

Ready to Transform Your
Cloud Native Journey?

Join hundreds of companies using AppsCode products and services to build, deploy, and manage production-grade Kubernetes applications.

Enterprise Support

24/7 Monitoring

Expert Consultation

Trusted by Fortune 500 companies and innovative startups worldwide