Deploying with SSL

Deploying with SSL

TripleO supports deploying with SSL on the public OpenStack endpoints as well as deploying SSL in the internal network for most services.

This document will focus on deployments using network isolation. For more details on deploying that way, see Configuring Network Isolation

Undercloud SSL

To enable SSL with an automatically generated certificate, you must set the generate_service_certificate option in undercloud.conf to True. This will generate a certificate in /etc/pki/tls/certs with a file name that follows the following pattern:


This will be a PEM file in a format that HAProxy can understand (see the HAProxy documentation for more information on this).

This option for auto-generating certificates uses Certmonger to request and keep track of the certificate. So you will see a certificate with the ID of undercloud-haproxy-public-cert in certmonger (you can check this by using the sudo getcert list command). Note that this also implies that certmonger will manage the certificate’s lifecycle, so when it needs renewing, certmonger will do that for you.

The default is to use Certmonger’s local CA. So using this option has the side-effect of extracting Certmonger’s local CA to a PEM file that is located in the following path:


This certificate will then be added to the trusted CA chain, since this is needed to be able to use the undercloud’s endpoints with that certificate.

However, it is possible to not use certmonger’s local CA. For instance, one can use FreeIPA as the CA by setting the option certificate_generation_ca in undercloud.conf to have ‘IPA’ as the value. This requires the undercloud host to be enrolled as a FreeIPA client, and to define a haproxy/<undercloud FQDN>@<KERBEROS DOMAIN> service in FreeIPA. We also need to set the option service_principal to the relevant value in undercloud.conf. Finally, we need to set the public endpoints to use FQDNs instead of IP addresses, which will also then use an FQDN for the certificate.

To enable an FQDN for the certificate we set the undercloud_public_vip to the desired hostname in undercloud.conf. This will in turn also set the keystone endpoints to relevant values.

Note that the generate_service_certificate option doesn’t take into account the undercloud_service_certificate option and will have precedence over it.

To enable SSL on the undercloud with a pre-created certificate, you must set the undercloud_service_certificate option in undercloud.conf to an appropriate certificate file. Important: The certificate file’s Common Name must be set to the value of undercloud_public_vip in undercloud.conf.

If you do not have a trusted CA signed certificate file, you can alternatively generate a self-signed certificate file using the following command:

openssl genrsa -out privkey.pem 2048

The next command will prompt for some identification details. Most of these don’t matter, but make sure the Common Name entered matches the value of undercloud_public_vip in undercloud.conf:

openssl req -new -x509 -key privkey.pem -out cacert.pem -days 365

Combine the two files into one for HAProxy to use. The order of the files in this command matters, so do not change it:

cat cacert.pem privkey.pem > undercloud.pem

Move the file to a more appropriate location and set the SELinux context:

sudo mkdir /etc/pki/instack-certs
sudo cp undercloud.pem /etc/pki/instack-certs
sudo semanage fcontext -a -t etc_t "/etc/pki/instack-certs(/.*)?"
sudo restorecon -R /etc/pki/instack-certs

undercloud_service_certificate should then be set to /etc/pki/instack-certs/undercloud.pem.

Add the self-signed CA certificate to the undercloud system’s trusted certificate store:

sudo cp cacert.pem /etc/pki/ca-trust/source/anchors/
sudo update-ca-trust extract

Overcloud SSL

Certificate and Public VIP Configuration

The public VIP of the deployed overcloud needs to be predictable in order for the SSL certificate to be configured properly. There are two options for configuring the certificate:

  1. The certificate’s Common Name can be set to the IP of the public VIP. In this case, the Common Name must match exactly. If the public VIP is, the certificate’s Common Name must also be Wild cards will not work.
  2. The overcloud endpoints can be configured to point at a DNS name. In this case, the certificate’s Common Name must be valid for the FQDN of the overcloud endpoints. Wild cards should work fine. Note that this option also requires pre-configuration of the specified DNS server with the appropriate FQDN and public VIP.

In either case, the public VIP must be explicitly specified as part of the deployment configuration. This can be done by passing an environment file like the following:

    PublicVirtualFixedIPs: [{'ip_address':''}]


If network isolation is not in use, the ControlFixedIPs parameter should be set instead.

The selected IP should fall in the specified allocation range for the public network.

Certificate Details

Self-Signed SSL

It is not recommended that the self-signed certificate is trusted; So for this purpose, having a self-signed CA certificate is a better choice. In this case we will trust the self-signed CA certificate, and not the leaf certificate that will be used for the public VIP; This leaf certificate, however, will be signed by the self-signed CA.

For the self-signed case, just the predictable public VIP method will be documented, as DNS configuration is outside the scope of this document.

Generate a private key:

openssl genrsa -out overcloud-ca-privkey.pem 2048

Generate a self-signed CA certificate. This command will prompt for some identifying information. Most of the fields don’t matter, and the CN should not be the same as the one we’ll give the leaf certificate. You can choose a CN for this such as “TripleO CA”:

openssl req -new -x509 -key overcloud-ca-privkey.pem \
     -out overcloud-cacert.pem -days 365

Add the self-signed CA certificate to the undercloud’s trusted certificate store. Adding this file to the overcloud nodes will be discussed later:

sudo cp overcloud-cacert.pem /etc/pki/ca-trust/source/anchors/
sudo update-ca-trust extract

Generate the leaf certificate request and key that will be used for the public VIP. Again, Most of the fields don’t matter, but this is where the Common Name must be set to the fixed IP in the external network allocation pool:

openssl req -newkey rsa:2048 -days 365 \
     -nodes -keyout server-key.pem -out server-req.pem

Process the server RSA key:

openssl rsa -in server-key.pem -out server-key.pem

Sign the leaf certificate with the CA certificate and generate the certificate:

openssl x509 -req -in server-req.pem -days 365 \
      -CA overcloud-cacert.pem -CAkey overcloud-ca-privkey.pem \
      -set_serial 01 -out server-cert.pem

The contents of the private key and certificate files must be provided to Heat as part of the deployment command. To do this, there is a sample environment file in tripleo-heat-templates with fields for the file contents.

It is generally recommended that the original copy of tripleo-heat-templates in /usr/share/openstack-tripleo-heat-templates not be altered, since it could be overwritten by a package update at any time. Instead, make a copy of the templates:

cp -r /usr/share/openstack-tripleo-heat-templates ~/ssl-heat-templates

Then edit the enable-tls.yaml environment file. If using the location from the previous command, the correct file would be in ~/ssl-heat-templates/environments/enable-tls.yaml. Insert the contents of the private key and certificate files in their respective locations.


The certificate and key will be multi-line values, and all of the lines must be indented to the same level.


In Newton the certificate parameters have been decoupled from the other parameters required to enable SSL, so the certificate parameters can be set in any environment file, not just enable-tls.yaml.

An abbreviated version of how the file should look:

    SSLCertificate: |
      -----BEGIN CERTIFICATE-----
      -----END CERTIFICATE-----
[rest of file snipped]

SSLKey should look similar, except with the value of the private key.

SSLIntermediateCertificate can be set in the same way if the certificate signer uses an intermediate certificate. Note that the | character must be added as in the other values to indicate that this is a multi-line value.

Stable Branch

From Newton, the EndpointMap and SSL certificates have been decoupled.

When the certificate’s common name is set to the public VIP, all instances of CLOUDNAME in enable-tls.yaml must be replaced with IP_ADDRESS. This is not necessary when using a DNS name for the overcloud endpoints


This command should be run exactly as shown below. Do not replace IP_ADDRESS with an actual address. Heat will insert the appropriate value at deploy time.

sed -i 's/CLOUDNAME/IP_ADDRESS/' ~/ssl-heat-templates/environments/enable-tls.yaml

When using a self-signed certificate or a signer whose certificate is not in the default trust store on the overcloud image it will be necessary to inject the certificate as part of the deploy process. This can be done with the environment file ~/ssl-heat-templates/environments/inject-trust-anchor.yaml. Insert the contents of the signer’s root CA certificate in the appropriate location, in a similar fashion to what was done for the certificate and key above.

Self-Signed SSL

Injecting the root CA certificate is required for self-signed SSL. The correct value to use is the contents of the overcloud-cacert.pem file.

DNS Endpoint Configuration

When deploying with DNS endpoint addresses, two additional parameters must be passed in a Heat environment file. These are CloudName and DnsServers. To do so, create a new file named something like cloudname.yaml:


Replace the values with ones appropriate for the target environment. Note that the configured DNS server(s) must have an entry for the configured CloudName that matches the public VIP.

In addition, when a DNS endpoint is being used, make sure to pass the tls-endpoints-public-dns.yaml environment to your deploy command. See the examples below.

Deploying an SSL Environment

The enable-tls.yaml file must always be passed to use SSL on the public endpoints. Depending on the specific configuration, additional files will also be needed. Examples of the necessary parameters for different scenarios follow.

IP-based certificate:

-e ~/ssl-heat-templates/environments/enable-tls.yaml -e ~/ssl-heat-templates/environments/tls-endpoints-public-ip.yaml

Self-signed IP-based certificate:

-e ~/ssl-heat-templates/environments/enable-tls.yaml -e ~/ssl-heat-templates/environments/tls-endpoints-public-ip.yaml -e ~/ssl-heat-templates/environments/inject-trust-anchor.yaml

DNS-based certificate:

-e ~/ssl-heat-templates/environments/enable-tls.yaml -e ~/ssl-heat-templates/environments/tls-endpoints-public-dns.yaml -e ~/cloudname.yaml

Self-signed DNS-based certificate:

-e ~/ssl-heat-templates/environments/enable-tls.yaml -e ~/ssl-heat-templates/environments/tls-endpoints-public-dns.yaml -e ~/cloudname.yaml -e ~/ssl-heat-templates/environments/inject-trust-anchor.yaml

Stable Branch

In Mitaka and older releases, the EndpointMap was in enable-tls.yaml, so there is no need to pass a tls-endpoints-*.yaml file. However, this means that the enable-tls.yaml file must be rebased when upgrading to reflect any new endpoints that may have been added. Examples of the necessary parameters for different scenarios follow.

The enable-tls.yaml file must be passed to the overcloud deploy command to enable SSL on the public endpoints. Include the following additional parameter in the overcloud deploy command:

-e ~/ssl-heat-templates/environments/enable-tls.yaml

The inject-trust-anchor.yaml file must also be passed if a root certificate needs to be injected. The additional parameters in that case would instead look like:

-e ~/ssl-heat-templates/environments/enable-tls.yaml -e ~/ssl-heat-templates/environments/inject-trust-anchor.yaml

When DNS endpoints are being used, the cloudname.yaml file must also be passed. The additional parameters would be (inject-trust-anchor.yaml may also be used if it is needed for the configured certificate):

-e ~/ssl-heat-templates/environments/enable-tls.yaml -e ~/cloudname.yaml [-e ~/ssl-heat-templates/environments/inject-trust-anchor.yaml]


It is also possible to get the public certificate from a CA. See TLS everywhere for the overcloud

Getting the overcloud to trust CAs

As mentioned above, it is possible to get the overcloud to trust a CA by using the ~/ssl-heat-templates/environments/inject-trust-anchor.yaml environment and adding the necessary details there. However, that environment has the restriction that it will only allow you to inject one CA. However, the file ~/ssl-heat-templates/environments/inject-trust-anchor-hiera.yaml is an alternative that actually supports as many CA certificates as you need.


This is only available since Newton. Older versions of TripleO don’t support this.

This file is a template of how you should fill the CAMap parameter which is passed via parameter defaults. It looks like this:

content: |
The content of the CA cert goes here
content: |
The content of the CA cert goes here

where first-ca-name and second-ca-name will generate the files first-ca-name.pem and second-ca-name.pem respectively. These files will be stored in the /etc/pki/ca-trust/source/anchors/ directory in each node of the overcloud and will be added to the trusted certificate chain of each of the nodes. You must be careful that the content is a block string in yaml and is in PEM format.

TLS everywhere for the overcloud

It is possible to deploy most of the services to use TLS for communications in the internal network as well. This, however, needs several more certificates than the public approach, with the number being dependant on the number of nodes in your deployment. This complicates certificate and key management to the extent where it’s not sustainable to have the deployer inject all the certificates and keys needed and then have to handle all their lifecycles. Then, we have to take into account that a certificate revocation might be needed at some point. So, from both the maintenance and security standpoints this is not sustainable.

For the aforementioned reasons, we decided to rely on certmonger to get the certificates from an actual CA. Certmonger will do the certificate requests and do the certificate renewals when it’s needed, thus reducing the maintenance burden.

FreeIPA has been chosen as the default CA. Certmonger already has a plugin for it, and it has the added value that, besides being able to automatically provide the certificates we need, we can also keep track of the nodes and have an identity for them.


The default CA can be overriden via the CertmongerCA parameter. However, the CA has to be something that certmonger understands, so there are adjustments to be done. For more information on how to change it you can consult the certmonger documentation

Communicating with the CA (FreeIPA) requires the nodes to have proper credentials, and these credentials also need to be transported into the overcloud nodes in a secure manner. To address this, we use a Nova vendordata plugin called novajoin whose purpose is to detect the nodes that are created by nova, register or join them in FreeIPA and provide an OTP that the node can subsequently use to enroll to FreeIPA. The node subsequently enrolls by loading the vendordata-provided JSON via the config-drive, which ends up executing a cloud-init script to do this. With the node enrolled, certificates can be requested securely. Novajoin can also receive extra entries from nova metadata to create extra principals that the services will need. These create service principals for services such as httpd, mysql and haproxy, and are used to requests the certificates for the specific service users with the correct SubjectAltNames.

Deployment workflow

The following are instructions assuming the default CA, which is FreeIPA.

CA setup

The undercloud needs to be enrolled to FreeIPA, and we need to create some extra privileges/permissions to be used by the novajoin services. Assuming there’s an already existing FreeIPA installation, we can use a script that comes with the python-novajoin package:

sudo /usr/libexec/novajoin-ipa-setup \
    --principal admin \
    --password < freeipa admin password > \
    --server < freeipa server hostname > \
    --realm < overcloud cloud domain in upper case > \
    --domain < overcloud cloud domain > \
    --hostname < undercloud hostname > \

This command will give us a One-Time Password (OTP) that we can then use for the undercloud enrollment. We can also specify the command to output the OTP into a file by using the --otp-file option.


This can be run from either the undercloud node itself or the FreeIPA node. Just note that the example provided is using the FreeIPA admin credentials. This can be done using another principal if it has the approprite permissions.

Undercloud setup

Now that we have an OTP we can either deploy or update the undercloud. The following settings in undercloud.conf will get the undercloud to enroll to FreeIPA and deploy novajoin:

enable_novajoin = True
ipa_otp = < OTP provided by the novajoin-ipa-setup script >

The undercloud fully-qualified hostname should also be set in undercloud.conf, since this is the host that will be used to enroll to FreeIPA. It should match the one provided in the novajoin-ipa setup script. We can set it like this:

undercloud_hostname = < undercloud FQDN >

It is useful to have FreeIPA set as the DNS server since this will automatically: discover the FreeIPA server hostname, set up the Kerberos realm/domain automatically, and it will set the DNS entries of the overcloud nodes once they’re deployed. We can set it in undercloud.conf with the following setting:

undercloud_nameservers = < FreeIPA IP >


This takes a comma-separated list, so we can set another nameserver with this configuration option.

With these settings, do the following command to set the desired configurations and enable novajoin:

openstack undercloud install

Overcloud deployment

The TLS-everywhere setup only works with FQDNs so we need to set the appropriate entries for the overcloud endpoints as well as setting an appropriate domain for the nodes that matches the one we set for FreeIPA. We can do this by overriding some parameters via parameter_defaults. Assuming that the domain for our cloud is We’ll set the following in a file we’ll call cloud-names.yaml which we’ll include in our overcloud deploy command:


As with our undercloud, we also want the overcloud nodes’ name server to point to FreeIPA. We can do this by setting the DnsServers parameter via parameter_defaults. You can create an environment file for it, however, since you probably are deploying with network isolation, you can already set this parameter in the network-environment.yaml file that’s referenced in Configuring Network Isolation. So that setting would look like this:

  DnsServers: ["< FreeIPA IP >"]

Remembering that optionally we can set other nameservers with this parameter.

To tell the overcloud deployment to deploy the keystone endpoints (and references) using DNS names instead of IPs, we need to add the following environment to our overcloud deployment:


Finally, to enable TLS in the internal network, we need to use the following environment:


This will set the appropriate resources that enable the certificate requests via certmonger and create the appropriate service principals for kerberos (which are used by FreeIPA).


As part of the enrollment, FreeIPA is set as a trusted CA, so we don’t need to do any extra steps for this.

Classic public TLS and certmonger-based internal TLS

enable-internal-tls.yaml will be used for the internal network endpoints. One can still use the enable-tls.yaml environment for the public endpoints if a specific certificate for the public endpoints is needed.

The arguments for a deployment using injected certificates for the public endpoints, and certmonger-provided certificates for the internal endpoints look like the following:

openstack overcloud deploy \
    -e ~/ssl-heat-templates/environments/tls-everywhere-endpoints-dns.yaml \
    -e ~/ssl-heat-templates/environments/enable-tls.yaml \
    -e ~/ssl-heat-templates/environments/enable-internal-tls.yaml \
    -e ~/cloud-names.yaml

Certmonger-based public and Internal TLS

It is also possible to get all your certificates from a CA. For this you need to include the environments/services/haproxy-public-tls-certmonger.yaml environment file.

To do a deployment with both public and internal endpoints using certificates provided by certmonger, we would need to issue a command similar to the following:

openstack overcloud deploy \
    -e ~/ssl-heat-templates/environments/tls-everywhere-endpoints-dns.yaml \
    -e ~/ssl-heat-templates/environments/services/haproxy-public-tls-certmonger.yaml \
    -e ~/ssl-heat-templates/environments/enable-internal-tls.yaml \
    -e ~/cloud-names.yaml
Creative Commons Attribution 3.0 License

Except where otherwise noted, this document is licensed under Creative Commons Attribution 3.0 License. See all OpenStack Legal Documents.