Fork me on GitHub

Trusted Clusters


The design of trusted clusters allows Teleport users to connect to compute infrastructure located behind firewalls without any open TCP ports. The real-world usage examples of this capability include:

  • Managed service providers (MSP) remotely managing the infrastructure of their clients.
  • Device manufacturers remotely maintaining computing appliances deployed on-premises.
  • Large cloud software vendors manage multiple data centers using a common proxy.

Example of a MSP provider using trusted cluster to obtain access to clients clusters. MSP Example

The Trusted Clusters chapter in the Admin Guide offers an example of a simple configuration which:

  • Uses a static cluster join token defined in a configuration file.
  • Does not cover inter-cluster Role-Based Access Control (RBAC).

This guide's focus is on more in-depth coverage of trusted clusters features and will cover the following topics:

  • How to add and remove trusted clusters using CLI commands.
  • Enable/disable trust between clusters.
  • Establish permissions mapping between clusters using Teleport roles.
Teleport Node Tunneling

If you have a large number of devices on different networks, such as managed IoT devices or a couple of nodes on a different network you can utilize the Teleport Node Tunneling.


As explained in the architecture document, Teleport can partition compute infrastructure into multiple clusters. A cluster is a group of SSH nodes connected to the cluster's auth server acting as a certificate authority (CA) for all users and nodes.

To retrieve an SSH certificate, users must authenticate with a cluster through a proxy server. So, if users want to connect to nodes belonging to different clusters, they would normally have to use different --proxy flags for each cluster. This is not always convenient.

The concept of leaf clusters allows Teleport administrators to connect multiple clusters and establish trust between them. Trusted clusters allow users of one cluster, the root cluster to seamlessly SSH into the nodes of another cluster without having to "hop" between proxy servers. Moreover, users don't even need to have a direct connection to other clusters' proxy servers.

The examples below may include the use of the sudo keyword, token UUIDs, and users with elevated privileges to make following each step easier.

We recommend you follow the best practices to avoid security incidents:

  1. Avoid using sudo in production environments unless it's necessary.
  2. Create new, non-root, users and use test instances for experimenting with Teleport.
  3. You can run many Teleport's services as a non root. For example, auth, proxy, application access, kubernetes access, and database access services can run as a non-root user. Only the SSH/node service requires root access. You will need root permissions (or the CAP_NET_BIND_SERVICE capability) to make Teleport listen on a port numbered < 1024 (e.g. 443)
  4. Follow the "Principle of Least Privilege" (PoLP) and "Zero Admin" best practices. Don't give users permissive roles when giving them more restrictive access,editor roles will do instead.
  5. Save tokens into a file rather than sharing tokens directly as strings.

The user experience looks like this:

Log in using the root "root" cluster credentials:

tsh login

SSH into some host inside the "root" cluster:

tsh ssh host

SSH into the host located in another cluster called "leaf"

The connection is established through

tsh ssh --cluster=leaf host

See what other clusters are available

tsh clusters

Leaf clusters also have their own restrictions on user access, i.e. permissions mapping takes place.

Once a connection has been established it's easy to switch from the "root" root cluster Teleport Cluster Page

Let's take a look at how a connection is established between the "root" cluster and the "leaf" cluster:


This setup works as follows:

  1. The "leaf" creates an outbound reverse SSH tunnel to "root" and keeps the tunnel open.
  2. Accessibility only works in one direction. The "leaf" cluster allows users from "root" to access its nodes but users in the "leaf" cluster can not access the "root" cluster.
  3. When a user tries to connect to a node inside "leaf" using the root's proxy, the reverse tunnel from step 1 is used to establish this connection shown as the green line above.
Load Balancers

The scheme above also works even if the "root" cluster uses multiple proxies behind a load balancer (LB) or a DNS entry with multiple values. This works by "leaf" establishing a tunnel to every proxy in "root". This requires that an LB uses a round-robin or a similar balancing algorithm. Do not use sticky load balancing algorithms (a.k.a. "session affinity" or "sticky sessions") with Teleport proxies.

Join Tokens

Lets start with the diagram of how connection between two clusters is established:


The first step in establishing a secure tunnel between two clusters is for the leaf cluster "leaf" to connect to the root cluster "root". When this happens for the first time, clusters know nothing about each other, thus a shared secret needs to exist for "root" to accept the connection from "leaf".

This shared secret is called a "join token". There are two ways to create join tokens: to statically define them in a configuration file or to create them on the fly using tctl tool.


It's important to note that join tokens are only used to establish the connection for the first time. The clusters will exchange certificates and won't use tokens to re-establish their connection afterward.

Static Join Tokens

To create a static join token, update the configuration file on "root" cluster to look like this:

# fragment of /etc/teleport.yaml:
  enabled: true
  # If using static tokens we recommend using tools like `pwgen -s 32`
  # to generate sufficiently random tokens of 32+ byte length
  - trusted_cluster:mk9JgEVqsgz6pSsHf4kJPAHdVDVtpuE0

This token can be used an unlimited number of times.

Security implications

Consider the security implications when deciding which token method to use. Short-lived tokens decrease the window for an attack but will require any automation which uses these tokens to refresh them regularly.

Dynamic Join Tokens

Creating a token dynamically with a CLI tool offers the advantage of applying a time-to-live (TTL) interval on it, i.e. it will be impossible to re-use such token after a specified time.

To create a token using the CLI tool, execute this command on the auth server of cluster "root":

Generates a trusted cluster token to allow an inbound connection from a leaf cluster:

tctl tokens add --type=trusted_cluster --ttl=5m

Example output:

The cluster invite token: abcd123-insecure-do-not-use-this

This token will expire in 5 minutes

Generates a trusted cluster token with labels:

every cluster joined using this token will inherit env:prod labels.

tctl tokens add --type=trusted_cluster --labels=env=prod

You can also list the outstanding non-expired tokens:

tctl tokens ls

... or delete/revoke an invitation:

tctl tokens rm abcd123-insecure-do-not-use-this

Users of Teleport will recognize that this is the same way you would add any node to a cluster. The token created above can be used multiple times and has an expiration time of 5 minutes.

Now, the administrator of "leaf" must create the following resource file:

# cluster.yaml
kind: trusted_cluster
version: v2
  # The trusted cluster name MUST match the 'cluster_name' setting of the
  # root cluster
  name: root
  # This field allows to create tunnels that are disabled, but can be enabled later.
  enabled: true
  # The token expected by the "root" cluster:
  token: ba4825847f0378bcdfe18113c4998498
  # The address in 'host:port' form of the reverse tunnel listening port on the
  # "root" proxy server:
  # The address in 'host:port' form of the web listening port on the
  # "root" proxy server:
  # The role mapping allows to map user roles from one cluster to another
  # (enterprise editions of Teleport only)
    - remote: "admin"    # users who have "admin" role on "root"
      local: ["auditor"] # will be assigned "auditor" role when logging into "leaf"

Then, use tctl create to add the file:

tctl create cluster.yaml

At this point, the users of the "root" cluster should be able to see "leaf" in the list of available clusters.

HTTPS configuration

If the web_proxy_addr endpoint of the root cluster uses a self-signed or invalid HTTPS certificate, you will get an error: "the trusted cluster uses misconfigured HTTP/TLS certificate". For ease of testing, the Teleport daemon on "leaf" can be started with the --insecure CLI flag to accept self-signed certificates. Make sure to configure HTTPS properly and remove the insecure flag for production use.


When a leaf cluster "leaf" from the diagram above establishes trust with the root cluster "root", it needs a way to configure which users from "root" should be allowed in and what permissions should they have. Teleport offers two methods of limiting access, by using role mapping of cluster labels.

Consider the following:

  • Both clusters "root" and "leaf" have their own locally defined roles.
  • Every user in Teleport Enterprise is assigned a role.
  • When creating a trusted cluster resource, the administrator of "leaf" must define how roles from "root" map to roles on "leaf".
  • To update the role map for an existing trusted cluster delete and re-create the trusted cluster with the updated role map.


Let's make a few assumptions for this example:

  • The cluster "root" has two roles: user for regular users and admin for local administrators.
  • We want administrators from "root" (but not regular users!) to have restricted access to "leaf". We want to deny them access to machines with "environment=production" and any Government cluster labeled "customer=gov"

First, we need to create a special role for root users on "leaf":

# Save this into root-user-role.yaml on the leaf cluster and execute:
# tctl create root-user-role.yaml
kind: role
version: v4
  name: local-admin
      '*': '*'
    # Cluster labels control what clusters user can connect to. The wildcard ('*') means
    # any cluster. If no role in the role set is using labels and the cluster is not labeled,
    # the cluster labels check is not applied. Otherwise, cluster labels are always enforced.
    # This makes the feature backward-compatible.
      'env': 'staging'
    # Cluster labels control what clusters user can connect to. The wildcard ('*') means
    # any cluster. By default none is set in deny rules to preserve backward compatibility
      'customer': 'gov'
      'environment': 'production'

Now, we need to establish trust between roles "root:admin" and "leaf:admin". This is done by creating a trusted cluster resource on "leaf" which looks like this:

# Save this as root-cluster.yaml on the auth server of "leaf" and then execute:
# tctl create root-cluster.yaml
kind: trusted_cluster
version: v1
  name: "name-of-root-cluster"
  enabled: true
    - remote: admin
      # admin <-> admin works for the Open Source Edition. Enterprise users
      # have great control over RBAC.
      local: [access]
  token: "join-token-from-root"

What if we wanted to let any user from "root" to be allowed to connect to nodes on "leaf"? In this case, we can use a wildcard * in the role_map like this:

  - remote: "*"
    local: [access]
   - remote: 'cluster-*'
     local: [clusteradmin]

You can even use regular expressions to map user roles from one cluster to another, you can even capture parts of the remote role name and use reference it to name the local role:

  # In this example, remote users with a remote role called 'remote-one' will be
  # mapped to a local role called 'local-one', and `remote-two` becomes `local-two`, etc:
  - remote: "^remote-(.*)$"
    local: [local-$1]

NOTE: The regexp matching is activated only when the expression starts with ^ and ends with $

Trusted Cluster UI

For customers using Teleport Enterprise, they can easily configure leaf nodes using the Teleport Proxy UI.

Creating Trust from the Leaf node to the root node. Tunnels

Updating Trusted Cluster role map

To update the role map for a trusted cluster, first, we'll need to remove the cluster by executing:

tctl rm tc/root-cluster

Then following updating the role map, we can re-create the cluster by executing:

tctl create root-user-updated-role.yaml

Updating cluster labels

Teleport gives administrators of root clusters the ability to control cluster labels. Allowing leaf clusters to propagate their own labels could create a problem with rogue clusters updating their labels to bad values.

An administrator of a root cluster can control a remote/leaf cluster's labels using the remote cluster API without any fear of override:

tctl get rc

kind: remote_cluster


name: two


connection: online

last_heartbeat: "2020-09-14T03:13:59.35518164Z"

version: v3

Using tctl to update the labels on the remote/leaf cluster:

tctl update rc/two --set-labels=env=prod

Cluster two has been updated

Using tctl to confirm that the updated labels have been set:

tctl get rc

kind: remote_cluster



env: prod

name: two


connection: online

last_heartbeat: "2020-09-14T03:13:59.35518164Z"

Using Trusted Clusters

Now an admin from the "root" cluster can see and access the "leaf" cluster:

Log into the root cluster:

tsh login admin

See the list of available clusters

tsh clusters

Cluster Name Status

------------ ------

root online

leaf online

See the list of machines (nodes) behind the leaf cluster:

tsh ls --cluster=leaf

Node Name Node ID Address Labels

--------- ------------------ -------------- -----------

db1.leaf cf7cc5cd-935e-46f1 role=db-leader

db2.leaf 3879d133-fe81-3212 role=db-follower

SSH into any node in "leaf":

tsh ssh --cluster=leaf [email protected]

Trusted clusters work only one way. So, in the example above users from "leaf" cannot see or connect to the nodes in "root".

Disabling trust

To temporarily disable trust between clusters, i.e. to disconnect the "leaf" cluster from "root", edit the YAML definition of the trusted cluster resource and set enabled to "false", then update it:

tctl create --force cluster.yaml

Remove Leaf Cluster relationship from both sides

Once established, to fully remove a trust relationship between two clusters, do the following:

  • Remove the relationship from the leaf cluster: tctl rm tc/ (tc = trusted cluster)
  • Remove the relationship from the root cluster: tctl rm rc/ (rc = remote cluster)

Remove Leaf Cluster relationship from the root

Remove the relationship from the root cluster: tctl rm rc/


The cluster will continue to try and ping the root cluster, but will not be able to connect. To re-establish the trusted cluster relationship, the trusted cluster has to be created again from the leaf cluster.

Remove Leaf Cluster relationship from the leaf

Remove the relationship from the leaf cluster: tctl rm tc/

Sharing user traits between Trusted Clusters

You can share user SSH logins, Kubernetes users/groups, and database users/names between Trusted Clusters.

Suppose you have a root cluster with a role named root and the following allow rules:

logins: ["root"]
kubernetes_groups: ["system:masters"]
kubernetes_users: ["alice"]
db_users: ["postgres"]
db_names: ["dev", "metrics"]

When setting up the Trusted Cluster relationship, the leaf cluster can choose to map this root cluster role to its own admin role:

- remote: "root"
  local: ["admin"]

The role admin of the leaf cluster can now be set up to use the root cluster's role logins, Kubernetes groups and other traits using the following variables:

logins: ["{{internal.logins}}"]
kubernetes_groups: ["{{internal.kubernetes_groups}}"]
kubernetes_users: ["{{internal.kubernetes_users}}"]
db_users: ["{{internal.db_users}}"]
db_names: ["{{internal.db_names}}"]

User traits that come from the identity provider (such as OIDC claims or SAML attributes) are also passed to the leaf clusters and can be access in the role templates using external variable prefix:

logins: ["{{internal.logins}}", "{{external.logins_from_okta}}"]
  env: "{{external.env_from_okta}}"

How does it work?

At a first glance, Trusted Clusters in combination with RBAC may seem complicated. However, it is based on certificate-based SSH authentication which is fairly easy to reason about:

One can think of an SSH certificate as a "permit" issued and time-stamped by a certificate authority. A certificate contains four important pieces of data:

  • List of allowed UNIX logins a user can use. They are called "principals" in the certificate.
  • Signature of the certificate authority who issued it (the auth server)
  • Metadata (certificate extensions): additional data protected by the signature above. Teleport uses the metadata to store the list of user roles and SSH options like "permit-agent-forwarding".
  • The expiration date.

Try executing tsh status right after tsh login to see all these fields in the client certificate.

When a user from "root" tries to connect to a node inside "leaf", her certificate is presented to the auth server of "leaf" and it performs the following checks:

  • Checks that the certificate signature matches one of the trusted clusters.
  • Tries to find a local role that maps to the list of principals found in the certificate.
  • Checks if the local role allows the requested identity (UNIX login) to have access.
  • Checks that the certificate has not expired.


There are three common types of problems Teleport administrators can run into when configuring trust between two clusters:

  • HTTPS configuration: when the root cluster uses a self-signed or invalid HTTPS certificate.
  • Connectivity problems: when a leaf cluster "leaf" does not show up in tsh clusters output on "root".
  • Access problems: when users from "root" get "access denied" error messages trying to connect to nodes on "leaf".

HTTPS configuration

If the web_proxy_addr endpoint of the root cluster uses a self-signed or invalid HTTPS certificate, you will get an error: "the trusted cluster uses misconfigured HTTP/TLS certificate". For ease of testing, the teleport daemon on "leaf" can be started with the --insecure CLI flag to accept self-signed certificates. Make sure to configure HTTPS properly and remove the insecure flag for production use.

Connectivity problems

To troubleshoot connectivity problems, enable verbose output for the auth servers on both clusters. Usually this can be done by adding --debug flag to teleport start --debug. You can also do this by updating the configuration file for both auth servers:

# Snippet from /etc/teleport.yaml
    output: stderr
    severity: DEBUG

On systemd-based distributions, you can watch the log output via:

journalctl -fu teleport

Most of the time you will find out that either a join token is mismatched/expired, or the network addresses for tunnel_addr or web_proxy_addr cannot be reached due to pre-existing firewall rules or how your network security groups are configured on AWS.

Access problems

Troubleshooting access denied messages can be challenging. A Teleport administrator should check to see the following:

  • Which roles a user is assigned on "root" when they retrieve their SSH certificate via tsh login. You can inspect the retrieved certificate with tsh status command on the client-side.
  • Which roles a user is assigned on "leaf" when the role mapping takes place. The role mapping result is reflected in the Teleport audit log. By default, it is stored in /var/lib/teleport/log on a auth server of a cluster. Check the audit log messages on both clusters to get answers for the questions above.
Have a suggestion or can’t find something?