Security Guide

Vespa Cloud has several security mechanisms it is important for developers to understand. Vespa Cloud has two different interaction paths, Data Plane and Control Plane. Communication with the Vespa application goes through the Data Plane, while the Control Plane is used to manage Vespa tenants and applications.

The Control Plane and the Data Plane has different security mechanisms, described in this guide.

Data Plane

Data plane is protected using mutual TLS or optionally tokens.

Configuring mTLS

Certificates can be created using the Vespa CLI:

$ vespa auth cert --application <tenant>.<app>.<instance>

$ vespa auth cert --application scoober.albums.default
Success: Certificate written to security/clients.pem
Success: Certificate written to $HOME/.vespa/scoober.albums.default/data-plane-public-cert.pem
Success: Private key written to $HOME/.vespa/scoober.albums.default/data-plane-private-key.pem

The certificates can be created regardless of the application existence in Vespa Cloud. One can use this command to generate security/clients.pem for an application package:

$ cp $HOME/.vespa/scoober.albums.default/data-plane-public-cert.pem security/clients.pem

Certificates can also be created using OpenSSL:

$ openssl req -x509 -sha256 -days 1825 -newkey rsa:2048 -keyout key.pem -out security/clients.pem

The certificate is placed inside the application package in security/clients.pem. Make sure clients.pem is placed correctly if the certificate is created with OpenSSL, while the Vespa CLI will handle this automatically.

security/clients.pem files can contain multiple PEM encoded certificates by concatenating them. This allows you to have multiple clients with separate private keys, making it possible to rotate to a new certificate without any downtime.

Permissions

To support different permissions for clients, it is possible to limit the permissions of a client. Only read or write permissions are supported.

Request mapping

The request actions are mapped from HTTP method. The default mapping rule is:

For /search/ this is replaced by:

Example

Create 3 different certificates, for three different use cases:

$ openssl req -x509 -sha256 -days 1825 -newkey rsa:2048 -keyout key.pem -out security/serve.pem
$ openssl req -x509 -sha256 -days 1825 -newkey rsa:2048 -keyout key.pem -out security/ingest.pem
$ openssl req -x509 -sha256 -days 1825 -newkey rsa:2048 -keyout key.pem -out security/full_access.pem

Notes:

Reference the certificate files from services xml using the clients element:

<container version='1.0'>
    ...
    <clients>
        <client id="serve" permissions="read">
            <certificate file="security/serve.pem"/>
        </client>
        <client id="ingest" permissions="write">
            <certificate file="security/ingest.pem"/>
        </client>
        <client id="full_access" permissions="read,write">
            <certificate file="security/full_access.pem"/>
        </client>
    </clients>
  ...
</container>

Custom request mapping

The default mapping can be changed by overriding requestHandlerSpec():

/**
 * Example overriding acl mapping of POST requests to read
 */
public class CustomAclHandler extends ThreadedHttpRequestHandler {

    private final static RequestHandlerSpec REQUEST_HANDLER_SPEC =
            RequestHandlerSpec.builder().withAclMapping(
                    HttpMethodAclMapping.standard()
                            .override(Method.POST, AclMapping.Action.READ)
                            .build())
                    .build();

    @Override
    public RequestHandlerSpec requestHandlerSpec() {
        return REQUEST_HANDLER_SPEC;
    }

Configure tokens

While mTLS continues to be the recommended option, the application can also be configured to consume token based authentication when mTLS is not available for the client (e.g. in case of edge functions). Note that it is still required to define at least one client for mTLS.

Create tokens using the console

Tokens are managed in the console under Account > Tokens. All tokens are identified by a name, and can contain multiple versions to easily support token rotation. To create a new token:

  1. Click Add token
  2. Enter a name for the token, note that this name must also be referenced in the application later.
  3. Select an expiration for the token.
  4. Click add. Remember to copy the token value and store securely. The value is not stored in Vespa Cloud.

To add a new version:

  1. Find the existing token, click Add version
  2. Select expiration and click Add. Copy the token value and store securely.

To revoke a version:

  1. Find the existing token version, click Revoke

To manually rotate a token:

  1. Add a new token version following the above steps
  2. Revoke the old version when no clients use the old version

Application configuration

After creating a token in the console it must be configured for accessing a container cluster, using clients configuration. Below is a simplified example for an application with two container clusters, one for feeding and document access (i.e. read+write), and another for query access (i.e. read) - one token for each:

<container id="documentapi" version="1.0">
    ...
    <clients>
        <client id="mtls" permissions="read,write">
            <certificate file="security/clients.pem"/>
        </client>
        <client id="feed-token-client" permissions="read,write">
            <token id="feed-token"/>
        </client>
    </clients>
    ...
</container>
<container id="query" version="1.0">
    ...
    <clients>
        <client id="mtls" permissions="read">
            <certificate file="security/clients.pem"/>
        </client>
        <client id="query-token-client" permissions="read">
            <token id="query-token"/>
        </client>
    </clients>
    ...
</container>

Notes:

Security recommendations

The cryptographic properties of token authentication vs mTLS are comparable. There are however a few key differences in how they are used:

It is therefore recommended to

Use endpoints

Using mTLS

Once the application is configured and deployed with a certificate in the application package, requests can be sent to the application. Again, the Vespa CLI can help to use the correct certificate.

$ vespa curl --application <tenant>.<app>.<instance> /ApplicationStatus

$ curl --key $HOME/.vespa/scoober.albums.default/data-plane-private-key.pem \
       --cert $HOME/.vespa/scoober.albums.default/data-plane-public-key.pem \
       $ENDPOINT

Using tokens

The token endpoint must be used when using tokens. After deployment is complete, the token endpoint will be available in the token endpoint list (marked “Token”). To use the token endpoint, the token should be sent as a bearer authorization header:

$ vespa query \
  --header="Authorization: Bearer $TOKEN" \
  'yql=select * from music where album contains "head"'

curl -H "Authorization: Bearer $TOKEN" $ENDPOINT

Using a browser

In Vespa guides, curl is used in examples, like:

$ curl --cert ./data-plane-public-cert.pem --key ./data-plane-private-key.pem $ENDPOINT

To use a browser, install key/cert pair - this procedure is tested on macOS Catalina / Chrome 80.0.3987.132, assuming Certificate Common Name is “cloud.vespa.example” (as in the guides):

  1. Install key/cert pair: 
    $ security import data-plane-private-key.pem -k ~/Library/Keychains/login.keychain
$ security import data-plane-public-cert.pem -k ~/Library/Keychains/login.keychain
  2. In Keychain Access: With login keychain -> My Certificates category -> cloud.vespa.example, right-click and "New Identity Preference", then add the endpoint url, like https://me.my-app.mytenant1.aws-us-east-1c.dev.z.vespa-app.cloud.
  3. Open the same URL in Chrome, choose the example.com certificate and allow Chrome to read the private key.

Using Postman

Many developers prefer interactive tools like Postman. The Vespa blog has an article on how to use Postman with Vespa.

Using Cloudflare Workers

See Using Cloudflare Workers with Vespa Cloud.

Control Plane

The control plane is used to manage the Vespa applications.

There are two different ways for access the Control Plane, using vespa auth login to log in as a regular user and using Application Keys. vespa auth login is intended for developers deploying manually to dev, while Application Keys are intended for deploying applications to production, typically by a continuous build tool. See more about these two methods below.

Managing users

Tenant administrators manage user access through the Vespa Console.

Vespa Console user management

Users have two different privilege levels

User access to Control Plane

Outside using the Vespa Console, communicating with the Control Plane is easiest with the Vespa CLI.

$ vespa auth login
Your Device Confirmation code is: ****-****

If you prefer, you can open the URL directly for verification
Your Verification URL: https://vespa.auth0.com/activate?user_code=****-****

Press Enter to open the browser to log in or ^C to quit...

Waiting for login to complete in browser ... done

Successfully logged in.

After logging in with the Vespa CLI, the CLI can be used to deploy applications. Users are logged in with the same privilege as the user described in the Vespa Console.

Application Key

If programmatic access to the Control Plane is needed, for example from a CI/CD system like GitHub Actions, the Application Key can be used - see example deploy-vector-search.yaml.

Configuration

The Application Key can be generated in the Console from the Deployment Screen. The key is generated in the browser but the private key appears as a download in the browser. The public key can be downloaded separately from Deployment Screen. The private key is never persisted in Vespa Cloud, so it is important that the private key is kept securely. If lost, the private key is unrecoverable.

Vespa Console application key management

The Application Key can also be generated using the Vespa CLI.

$ vespa auth api-key -a <tenant>.<app>.<instance>

$ vespa auth api-key -a scoober.albums.default
Success: API private key written to $HOME/.vespa/scoober.api-key.pem

This is your public key:
-----BEGIN PUBLIC KEY-----
MFkwEwYHKoZIzj0CAQYIKoZIzj0DAQcDQgAE5fQUq12J/IlQQdE8pWC5596S7x9f
HpPcyxCX2dXBS4aqKxnfN5HEyTkLCNGCo9HQljgLziqW1VFzshAdm3hHQg==
-----END PUBLIC KEY-----

Its fingerprint is:
91:1f:de:e3:9f:d3:21:28:1b:1b:05:40:52:72:81:4f

To use this key in Vespa Cloud click 'Add custom key' at
https://console.vespa-cloud.com/tenant/scoober/keys
and paste the entire public key including the BEGIN and END lines.

Using the application key

The Application Key can be used from the Vespa CLI to run requests again the Control Plane. Action like deploying applications to Vespa Cloud.

$ vespa deploy -z dev.aws-us-east-1c

Resources

Overview deck
Vespa.ai
Roadmap

Contact

info@vespa.ai
Contact Sales
Issues
Mailing list

Community

Slack
Stack Overflow
Contributing
Copyright Vespa.ai — Privacy