Back to Cube

Kerberos authentication

docs-mintlify/docs/integrations/power-bi/kerberos.mdx

1.7.812.5 KB
Original Source

Kerberos is the most common authentication method for Windows environments. It can be used to authenticate requests to the DAX API and MDX API.

<Note>

Available on Enterprise plan.

</Note>

On the diagram below, Kerberos is used to authenticate requests from Power BI Desktop (step 2):

Authentication flow

Kerberos is the recommended method to authenticate Power BI Desktop requests.

It works as follows:

  • Power BI Desktop is launched normally, under the Windows domain account of the user.
  • When connecting the DAX API, Windows verifies whether its service principal name is registered in the domain.
  • Once verified, the Key Distribution Center issues a Kerberos ticket for the user.
  • This ticket is transmitted to the DAX API in the request authorization header.
  • The DAX API decrypts and verifies the Kerberos ticket.
  • Finally, the user principal name is resolved to a Cube user.

Configuration

The Settings → Power BI page of your Cube deployment is the main entrypoint for configuring XMLA authentication. It shows the XMLA endpoint and your deployment's domain, generates setspn and ktpass command templates, stores the XMLA service account credentials, and accepts the keytab upload. The generated commands use the XMLA service account name — replace it with the keytab account before running them.

Configuring Kerberos authentication includes the following steps:

Obtaining a Windows machine

To perform the next steps, you need a Windows Server virtual machine:

  • It should be joined to the same domain as the organization’s users.
  • It should have the RSAT feature enabled.
  • It should be able to reach the Key Distribution Center (KDC). For example, on Azure, this virtual machine can be created in the aadds-vnet subnet.

You should log in to this Windows Server machine using the account that has AAD DC Administrators group membership.

It is also recommended to create a custom organizational unit (OU) for the service accounts.

On the screenshot below, the mdax-api-svc-account user is created in the MyCustomOU OU in the CUBE domain:

<Frame> </Frame>

Creating the service accounts

Create two separate accounts in your directory:

  • A keytab account (e.g., mdax-api-svc-account) that the SPNs are registered on and the keytab is generated for. Running ktpass rewrites this account's user principal name to the SPN format — this is expected, but it means the account can no longer sign in as a regular user, so don't use it for anything else.
  • An XMLA service account (e.g., cube-pbi-svc-account) that clients — for example, the Power BI on-premises data gateway — use to connect to Cube.

Registering the SPN

A service principal name (SPN) is a unique identifier of a service instance. Kerberos authentication uses SPNs to associate a service instance with a service sign-in account.

First, obtain your deployment’s domain from the Settings → Power BI page.

Then, use the setspn command to register the Service Principal Name for the DAX API against the keytab account.

In the following example, the web service (HTTP) SPN on the redundant-brohman.gcp-us-central1.cubecloudapp.dev domain is registered for the mdax-api-svc-account user in the CUBE domain:

bash
setspn -S HTTP/redundant-brohman.gcp-us-central1.cubecloudapp.dev CUBE\mdax-api-svc-account

If clients connect to the deployment through a different hostname (e.g., a custom domain), register the SPN for the hostname that clients actually connect to.

Load balancer SPN on AWS

On AWS, the deployment’s domain is a CNAME record pointing to the DNS name of an AWS load balancer. When connecting directly (without an HTTP proxy), Windows resolves the CNAME and requests a Kerberos ticket for the load balancer hostname, so a second SPN must be registered on the same account. Without it, clients silently fall back to NTLM. GCP deployments resolve to an A record and don’t need this.

Find the load balancer hostname and register the second SPN:

bash
nslookup redundant-brohman.aws-us-east-1.cubecloudapp.dev
# Name: a1b2c3d4e5f6a7b8c9d0e1f2a3b4c5d6-1234567890abcdef.elb.us-east-1.amazonaws.com

setspn -S HTTP/a1b2c3d4e5f6a7b8c9d0e1f2a3b4c5d6-1234567890abcdef.elb.us-east-1.amazonaws.com CUBE\mdax-api-svc-account

The keytab doesn’t need to be regenerated — both SPNs share the keytab account’s key. If the deployment’s load balancer is ever re-provisioned, its hostname changes and the second SPN must be registered again.

Generating and uploading the keytab

The keytab file contains information needed to decrypt the Kerberos token.

First, use the ktpass command to generate the keytab file. You will be prompted to enter the password for the specified user:

bash
ktpass /out kerberos.keytab /princ HTTP/[email protected] /mapuser mdax-api-svc-account /crypto All /ptype KRB5_NT_PRINCIPAL /pass *

Then, upload the .keytab file on the Settings → Power BI page. It is stored Base64-encoded in the CUBE_XMLA_KRB5_KEYTAB_B64 environment variable. The CUBE_XMLA_SPN and KRB5_KTNAME environment variables are managed by Cube automatically and don’t need to be set.

Setting the XMLA service account

On the Settings → Power BI page, set the XMLA service account name and password. They are stored in the CUBE_XMLA_API_USER and CUBE_XMLA_API_PASSWORD environment variables and default to the deployment’s SQL API credentials.

A connection authenticated as this account is allowed to impersonate other users: when Power BI Service passes an end user’s UPN via EffectiveUserName, Cube switches the session to that user. For this to work, the value must exactly match the user name the connection authenticates as — with Kerberos, that is the full principal in user@REALM format, e.g., [email protected]. The match is case-sensitive, so keep the same letter case as the principal — the realm is typically uppercase. A NetBIOS format like CUBE\cube-pbi-svc-account will not match.

Once the deployment is ready, you can test the Kerberos authentication by connecting from Power BI to the DAX API.

Provisioning users with SCIM

The user principal name passed by Kerberos is often not the email a user is provisioned into Cube Cloud with. For example, Power BI sends a UPN like [email protected], while the user exists in Cube as [email protected]. For authentication to succeed, the principal name must resolve to a Cube user.

To bridge this, Cube Cloud's SCIM API exposes an extension that lets your identity provider sync an alternate username alongside each user. During username-based authentication Cube checks the user's email, username, and any aliases — so the Kerberos UPN resolves to the same user.

Extension schema

Map an IdP attribute to the following single-valued, string target attribute:

urn:cube:params:1.0:UserAliases:alternateUserName

The value is lowercased and trimmed on write, and lookups are lowercased too, so matching is fully case-insensitive — the value just needs to be the same principal string Kerberos sends (same realm and separator). It must be unique across the tenant: if it collides with another user's email, username, or an existing alias, the sync is rejected with 409 Conflict.

Setting it up in Microsoft Entra

These steps assume you already have a SCIM provisioning app connected to Cube Cloud. If not, enable SCIM first under Cube → Settings → Authentication & SSO.

  1. In Enterprise applications → [your Cube SCIM app] → Provisioning → Attribute mappings → Provision Microsoft Entra ID Users, tick Show advanced options and click Edit attribute list for customappsso.
  2. Add a row: urn:cube:params:1.0:UserAliases:alternateUserName, type String, not multi-valued, not required. Save.
  3. Click Add New Mapping and configure:
    • Mapping type: Direct
    • Source attribute: the attribute holding the alternate identity — for hybrid AD / Kerberos this is typically onPremisesUserPrincipalName. To build the principal from parts, use an Expression mapping, e.g. Join("@", [samAccountName], "INTERNAL.REALM").
    • Target attribute: urn:cube:params:1.0:UserAliases:alternateUserName
    • Apply this mapping: Always
  4. Save the mapping and the provisioning configuration.
  5. Verify with Provision on demand on a test user, then fetch them via GET /api/scim/v2/Users/:id to confirm the alias appears under the extension.
<Note>

Entra only syncs a user when their source data changes, so existing users won't receive the alias until their next change — use Restart provisioning to force a full re-sync. onPremisesUserPrincipalName is only populated for users synced from on-prem AD; cloud-only users need a different source attribute or an expression with a fallback.

</Note>

Other identity providers

The Cube-side target attribute is always urn:cube:params:1.0:UserAliases:alternateUserName; only the IdP-side mapping UI differs. In Okta, add a custom attribute on the SCIM app via the Profile Editor, then bind your source attribute to it on the Mappings tab. The value Cube receives is just a string — what you map to it is your decision.

Troubleshooting

  • Kerberos works through an HTTP proxy (e.g., Fiddler) but falls back to NTLM on a direct connection. On AWS, this means the load balancer SPN is missing.
  • Unable to fetch Cloud context for XMLA user in logs, Access Denied in the client. The Kerberos principal doesn’t resolve to a Cube user. Check that the user has a matching alternate username; if the user was provisioned before the SCIM mapping was added, force a re-sync.
  • ... is not allowed to switch to '<user>'. The connection didn’t authenticate as the XMLA service account — often because the account is set in NetBIOS format instead of user@REALM, or with a different letter case.
  • Errors when connecting as the keytab account. ktpass rewrites the keytab account’s UPN, so it can no longer sign in. Connect with a separate XMLA service account.

Cube always offers Negotiate, NTLM, and Basic authentication on the XMLA endpoints; the NTLM fallback can’t be disabled. Check the deployment logs to see which method a connection actually used: method=kerberos vs. method=ntlm.