luxolus
/
open-nomad
2
0
Fork 0
Code Issues 1 Pull Requests Packages Releases Wiki Activity
open-nomad/website/source/guides/security/securing-nomad.html.md

18 KiB
Raw Blame History

layout page_title sidebar_current description
guides Securing Nomad with TLS guides-security-tls Securing Nomad's cluster communication with TLS is important for both security and easing operations. Nomad can use mutual TLS (mTLS) for authenticating for all HTTP and RPC communication.

Securing Nomad with TLS

Securing Nomad's cluster communication is not only important for security but can even ease operations by preventing mistakes and misconfigurations. Nomad optionally uses mutual TLS (mTLS) for all HTTP and RPC communication. Nomad's use of mTLS provides the following properties:

  • Prevent unauthorized Nomad access
  • Prevent observing or tampering with Nomad communication
  • Prevent client/server role or region misconfigurations
  • Prevent other services from masquerading as Nomad agents

Preventing region misconfigurations is a property of Nomad's mTLS not commonly found in the TLS implementations on the public Internet. While most uses of TLS verify the identity of the server you are connecting to based on a domain name such as example.com, Nomad verifies the node you are connecting to is in the expected region and configured for the expected role (e.g. client.us-west.nomad). This also prevents other services who may have access to certificates signed by the same private CA from masquerading as Nomad agents. If certificates were identified based on hostname/IP then any other service on a host could masquerade as a Nomad agent.

Correctly configuring TLS can be a complex process, especially given the wide range of deployment methodologies. If you use the sample Vagrantfile from the Getting Started Guide - or have cfssl and Nomad installed - this guide will provide you with a production ready TLS configuration.

~> Note that while Nomad's TLS configuration will be production ready, key management and rotation is a complex subject not covered by this guide. Vault is the suggested solution for key generation and management.

Creating Certificates

The first step to configuring TLS for Nomad is generating certificates. In order to prevent unauthorized cluster access, Nomad requires all certificates be signed by the same Certificate Authority (CA). This should be a private CA and not a public one like Let's Encrypt as any certificate signed by this CA will be allowed to communicate with the cluster.

~> Nomad certificates may be signed by intermediate CAs as long as the root CA is the same. Append all intermediate CAs to the cert_file.

Certificate Authority

There are a variety of tools for managing your own CA, like the PKI secret backend in Vault, but for the sake of simplicity this guide will use cfssl. You can generate a private CA certificate and key with cfssl:

$ # Generate the CA's private key and certificate
$ cfssl print-defaults csr | cfssl gencert -initca - | cfssljson -bare nomad-ca

The CA key (nomad-ca-key.pem) will be used to sign certificates for Nomad nodes and must be kept private. The CA certificate (nomad-ca.pem) contains the public key necessary to validate Nomad certificates and therefore must be distributed to every node that requires access.

Node Certificates

Once you have a CA certificate and key you can generate and sign the certificates Nomad will use directly. TLS certificates commonly use the fully-qualified domain name of the system being identified as the certificate's Common Name (CN). However, hosts (and therefore hostnames and IPs) are often ephemeral in Nomad clusters. Not only would signing a new certificate per Nomad node be difficult, but using a hostname provides no security or functional benefits to Nomad. To fulfill the desired security properties (above) Nomad certificates are signed with their region and role such as:

  • client.global.nomad for a client node in the global region
  • server.us-west.nomad for a server node in the us-west region

To create certificates for the client and server in the cluster from the Getting Started guide with cfssl create (or download) the following configuration file as cfssl.json to increase the default certificate expiration time:

{
  "signing": {
    "default": {
      "expiry": "87600h",
      "usages": [
        "signing",
        "key encipherment",
        "server auth",
        "client auth"
      ]
    }
  }
}

$ # Generate a certificate for the Nomad server
$ echo '{}' | cfssl gencert -ca=nomad-ca.pem -ca-key=nomad-ca-key.pem -config=cfssl.json \
    -hostname="server.global.nomad,localhost,127.0.0.1" - | cfssljson -bare server

# Generate a certificate for the Nomad client
$ echo '{}' | cfssl gencert -ca=nomad-ca.pem -ca-key=nomad-ca-key.pem -config=cfssl.json \
    -hostname="client.global.nomad,localhost,127.0.0.1" - | cfssljson -bare client

# Generate a certificate for the CLI
$ echo '{}' | cfssl gencert -ca=nomad-ca.pem -ca-key=nomad-ca-key.pem -profile=client \
    - | cfssljson -bare cli

Using localhost and 127.0.0.1 as subject alternate names (SANs) allows tools like curl to be able to communicate with Nomad's HTTP API when run on the same host. Other SANs may be added including a DNS resolvable hostname to allow remote HTTP requests from third party tools.

You should now have the following files:

  • cfssl.json - cfssl configuration.
  • nomad-ca.csr - CA signing request.
  • nomad-ca-key.pem - CA private key. Keep safe!
  • nomad-ca.pem - CA public certificate.
  • cli.csr - Nomad CLI certificate signing request.
  • cli-key.pem - Nomad CLI private key.
  • cli.pem - Nomad CLI certificate.
  • client.csr - Nomad client node certificate signing request for the global region.
  • client-key.pem - Nomad client node private key for the global region.
  • client.pem - Nomad client node public certificate for the global region.
  • server.csr - Nomad server node certificate signing request for the global region.
  • server-key.pem - Nomad server node private key for the global region.
  • server.pem - Nomad server node public certificate for the global region.

Each Nomad node should have the appropriate key (-key.pem) and certificate (.pem) file for its region and role. In addition each node needs the CA's public certificate (nomad-ca.pem).

Configuring Nomad

Next Nomad must be configured to use the newly-created key and certificates for mTLS. Starting with the server configuration from the Getting Started guide add the following TLS configuration options:

# Increase log verbosity
log_level = "DEBUG"

# Setup data dir
data_dir = "/tmp/server1"

# Enable the server
server {
  enabled = true

  # Self-elect, should be 3 or 5 for production
  bootstrap_expect = 1
}

# Require TLS
tls {
  http = true
  rpc  = true

  ca_file   = "nomad-ca.pem"
  cert_file = "server.pem"
  key_file  = "server-key.pem"

  verify_server_hostname = true
  verify_https_client    = true
}

The new tls section is worth breaking down in more detail:

tls {
  http = true
  rpc  = true
  # ...
}

This enables TLS for the HTTP and RPC protocols. Unlike web servers, Nomad doesn't use separate ports for TLS and non-TLS traffic: your cluster should either use TLS or not.

tls {
  # ...

  ca_file   = "nomad-ca.pem"
  cert_file = "server.pem"
  key_file  = "server-key.pem"

  # ...
}

The file lines should point to wherever you placed the certificate files on the node. This guide assumes they are in Nomad's current directory.

tls {
  # ...

  verify_server_hostname = true
  verify_https_client    = true
}

These two settings are important for ensuring all of Nomad's mTLS security properties are met. If verify_server_hostname is set to false the node's certificate will be checked to ensure it is signed by the same CA, but its role and region will not be verified. This means any service with a certificate signed by same CA as Nomad can act as a client or server of any region.

verify_https_client requires HTTP API clients to present a certificate signed by the same CA as Nomad's certificate. It may be disabled to allow HTTP API clients (e.g. Nomad CLI, Consul, or curl) to communicate with the HTTPS API without presenting a client-side certificate. If verify_https_client is enabled only HTTP API clients presenting a certificate signed by the same CA as Nomad's certificate are allowed to access Nomad.

~> Enabling verify_https_client effectively protects Nomad from unauthorized network access at the cost of losing Consul HTTPS health checks for agents.

Client Configuration

The Nomad client configuration is similar to the server configuration. The biggest difference is in the certificate and key used for configuration.

# Increase log verbosity
log_level = "DEBUG"

# Setup data dir
data_dir = "/tmp/client1"

# Enable the client
client {
  enabled = true

  # For demo assume we are talking to server1. For production,
  # this should be like "nomad.service.consul:4647" and a system
  # like Consul used for service discovery.
  servers = ["127.0.0.1:4647"]
}

# Modify our port to avoid a collision with server1
ports {
  http = 5656
}

# Require TLS
tls {
  http = true
  rpc  = true

  ca_file   = "nomad-ca.pem"
  cert_file = "client.pem"
  key_file  = "client-key.pem"

  verify_server_hostname = true
  verify_https_client    = true
}

Running with TLS

Now that we have certificates generated and configuration for a client and server we can test our TLS-enabled cluster!

In separate terminals start a server and client agent:

$ # In one terminal...
$ nomad agent -config server1.hcl

$ # ...and in another
$ nomad agent -config client1.hcl

If you run nomad node status now, you'll get an error, like:

Error querying node status: Get http://127.0.0.1:4646/v1/nodes: malformed HTTP response "\x15\x03\x01\x00\x02\x02"

This is because the Nomad CLI defaults to communicating via HTTP instead of HTTPS. We can configure the local Nomad client to connect using TLS and specify our custom keys and certificates using the command line:

$ nomad node status -ca-cert=nomad-ca.pem -client-cert=cli.pem -client-key=cli-key.pem -address=https://127.0.0.1:4646

This process can be cumbersome to type each time, so the Nomad CLI also searches environment variables for default values. Set the following environment variables in your shell:

$ export NOMAD_ADDR=https://localhost:4646
$ export NOMAD_CACERT=nomad-ca.pem
$ export NOMAD_CLIENT_CERT=cli.pem
$ export NOMAD_CLIENT_KEY=cli-key.pem
  • NOMAD_ADDR is the URL of the Nomad agent and sets the default for -addr.
  • NOMAD_CACERT is the location of your CA certificate and sets the default for -ca-cert.
  • NOMAD_CLIENT_CERT is the location of your CLI certificate and sets the default for -client-cert.
  • NOMAD_CLIENT_KEY is the location of your CLI key and sets the default for -client-key.

After these environment variables are correctly configured, the CLI will respond as expected:

$ nomad node status
ID        DC   Name   Class   Drain  Eligibility  Status
237cd4c5  dc1  nomad  <none>  false  eligible     ready

$ nomad job init
Example job file written to example.nomad
vagrant@nomad:~$ nomad job run example.nomad
==> Monitoring evaluation "e9970e1d"
    Evaluation triggered by job "example"
    Allocation "a1f6c3e7" created: node "237cd4c5", group "cache"
    Evaluation within deployment: "080460ce"
    Evaluation status changed: "pending" -> "complete"
==> Evaluation "e9970e1d" finished with status "complete"

Server Gossip

At this point all of Nomad's RPC and HTTP communication is secured with mTLS. However, Nomad servers also communicate with a gossip protocol, Serf, that does not use TLS:

  • HTTP - Used to communicate between CLI and Nomad agents. Secured by mTLS.
  • RPC - Used to communicate between Nomad agents. Secured by mTLS.
  • Serf - Used to communicate between Nomad servers. Secured by a shared key.

Nomad server's gossip protocol use a shared key instead of TLS for encryption. This encryption key must be added to every server's configuration using the encrypt parameter or with the -encrypt command line option.

The Nomad CLI includes a operator keygen command for generating a new secure gossip encryption key:

$ nomad operator keygen
cg8StVXbQJ0gPvMd9o7yrg==

Alternatively, you can use any method that base64 encodes 16 random bytes:

$ openssl rand -base64 16
raZjciP8vikXng2S5X0m9w==
$ dd if=/dev/urandom bs=16 count=1 status=none | base64
LsuYyj93KVfT3pAJPMMCgA==

Put the same generated key into every server's configuration file or command line arguments:

server {
  enabled = true

  # Self-elect, should be 3 or 5 for production
  bootstrap_expect = 1

  # Encrypt gossip communication
  encrypt = "cg8StVXbQJ0gPvMd9o7yrg=="
}

Switching an existing cluster to TLS

Since Nomad does not use different ports for TLS and non-TLS communication, the use of TLS must be consistent across the cluster. Switching an existing cluster to use TLS everywhere is operationally similar to upgrading between versions of Nomad, but requires additional steps to preventing needlessly rescheduling allocations.

  1. Add the appropriate key and certificates to all nodes.
  • Ensure the private key file is only readable by the Nomad user.
  1. Add the environment variables to all nodes where the CLI is used.
  2. Add the appropriate tls block to the configuration file on all nodes.
  3. Generate a gossip key and add it the Nomad server configuration.

~> Once a quorum of servers are TLS-enabled, clients will no longer be able to communicate with the servers until their client configuration is updated and reloaded.

At this point a rolling restart of the cluster will enable TLS everywhere. However, once servers are restarted clients will be unable to heartbeat. This means any client unable to restart with TLS enabled before their heartbeat TTL expires will have their allocations marked as lost and rescheduled.

While the default heartbeat settings may be sufficient for concurrently restarting a small number of nodes without any allocations being marked as lost, most operators should raise the heartbeat_grace configuration setting before restarting their servers:

  1. Set heartbeat_grace = "1h" or an appropriate duration on servers.
  2. Restart servers, one at a time.
  3. Restart clients, one or more at a time.
  4. Set heartbeat_grace back to its previous value (or remove to accept the default).
  5. Restart servers, one at a time.

~> In a future release Nomad will allow upgrading a cluster to use TLS by allowing servers to accept TLS and non-TLS connections from clients during the migration.

Jobs running in the cluster will not be affected and will continue running throughout the switch as long as all clients can restart within their heartbeat TTL.

Changing Nomad certificates on the fly

As of 0.7.1, Nomad supports dynamic certificate reloading via SIGHUP.

Given a prior TLS configuration as follows:

tls {
  http = true
  rpc  = true

  ca_file   = "nomad-ca.pem"
  cert_file = "server.pem"
  key_file  = "server-key.pem"

  verify_server_hostname = true
  verify_https_client    = true
}

Nomad's cert_file and key_file can be reloaded via SIGHUP simply by updating the TLS stanza to:

tls {
  http = true
  rpc  = true

  ca_file   = "nomad-ca.pem"
  cert_file = "new_server.pem"
  key_file  = "new_server_key.pem"

  verify_server_hostname = true
  verify_https_client    = true
}

Migrating a cluster to TLS

Reloading TLS configuration via SIGHUP

Nomad supports dynamically reloading both client and server TLS configuration. To reload an agent's TLS configuration, first update the TLS block in the agent's configuration file and then send the Nomad agent a SIGHUP signal. Note that this will only reload a subset of the configuration file, including the TLS configuration.

The agent reloads all its network connections when there are changes to its TLS configuration during a config reload via SIGHUP. Any new connections established will use the updated configuration, and any outstanding old connections will be closed. This process works when upgrading to TLS, downgrading from it, as well as rolling certificates. We recommend upgrading to TLS.

RPC Upgrade Mode for Nomad Servers

When migrating to TLS, the rpc_upgrade_mode option (defaults to false) in the TLS configuration for a Nomad server can be set to true. When set to true, servers will accept both TLS and non-TLS connections. By accepting non-TLS connections, operators can upgrade clients to TLS without the clients being marked as lost because the server is rejecting the client connection due to the connection not being over TLS. However, it is important to note that rpc_upgrade_mode should be used as a temporary solution in the process of migration, and this option should be re-set to false (meaning that the server will strictly accept only TLS connections) once the entire cluster has been migrated.