Carvel Logo

Injecting Secrets

Overview

This document is work in progress.

Unlike most configuration, many organizations disallow storing of plain secret values next to other code/configuration.

This document:

  • presents several common approaches used to make secrets available to your templates
  • does not cover injection of secrets directly into an application at runtime (overall may be the best approach)
  • does not recommend one approach over another (though it does state pros and cons)
  • does not talk about where resulting templates are forwarded

One common question that’s asked is why not to extend ytt to allow it to shell out to other programs or why not include builtin library that can fetch secrets from outside (e.g. imagine having builtin vault function). We are not planning to support either of these features because we want to maintain strong sandbox boundary between templating language and ytt’s external environment. Our expectation is that if ytt user wants to provide specific data to templates, it should be injected into ytt rather than templates (or ytt itself) fetching it. We believe this is a good security posture and as a side effect makes templates more portable (some users may store secrets in Vault and some other in Lastpass).

There might be other ways to inject secrets into ytt that are not listed here.


Via command line args

ytt -f . --data-value username=user --data-value password=pass

Cons:

  • depending on a computing environment secret values are visible in a process tree (as part of full command line) which typically is considered unwanted

Via environment variables

export CFG_secrets__username=user
export CFG_secrets__password=pass
ytt -f . --data-values-env CFG

Cons:

  • depending on a computing environment secret values are visible in process metadata (e.g. typically accessible to root user under /proc/<pid>/environ)

Via secret references, before templating

Given following two files:

config.yml:

#@ load("@ytt:data", "data")

users:
- username: #@ data.values.username
  password: #@ data.values.password

secrets:

username: (( vault "secret/my/credentials/admin:username" ))
password: (( vault "secret/my/credentials/admin:password" ))

(Note that secrets file does not have yaml extension to make ytt exclude it from template result. Alternatively --file-mark flag can be used to exclude that file.)

One can compose ytt with external tooling that knows how to resolve referenced values. For example, spruce can resolve secret references and then forward that information onto ytt:

echo -e "#@data/values\n---\n$(spruce merge ./secrets)" | ytt -f . -f -

Pros:

  • able to commit mapping of data values to secrets in your preferred secret storage (in this example Vault)
  • actual secret values are available to templates, hence, can be inserted in the middle of a string or base64 encoded (as sometimes required by k8s)

Via encrypted secrets, before templating

Similar to above “Via secret references, before templating” approach, instead of storing secret references in secrets file one can store encrypted secret values in secrets next to other configuration. Various tools like sops make encrypting configuration files fairly easy:

echo -e "#@data/values\n---\n$(sops -d ./secrets)" | ytt -f . -f -

Pros:

  • provides a way to version (backup, etc) secrets next to other configuration
  • actual secret values are available to templates, hence, can be inserted in the middle of a string or base64 encoded (as sometimes required by k8s)

Cons:

  • depending on organization requirements even encrypted secrets may not be allowed to be stored next to other configuration

Via secret references, after templating

Given following two files:

config.yml:

#@ load("@ytt:data", "data")

users:
- username: #@ data.values.username
  password: #@ data.values.password

values.yml:

#@data/values
---
username: (( vault "secret/my/credentials/admin:username" ))
password: (( vault "secret/my/credentials/admin:password" ))

One can compose ytt with external tooling that knows how to resolve referenced values. Such tools typically look for a particular patterned map or array items. One such tool is spruce:

ytt -f . | spruce merge -

Pros:

  • provides a way to commit mapping of data values and their associated secret locations (e.g. username comes from vault under secret/my/credentials/admin:username)

Cons:

  • secret resolution tools may not be able to find secret references inside strings (e.g. http://(( ... )):(( ... ))@website)
  • secret resolution tools will not understand if secret reference was encoded (for example with base64 encoding as wanted sometimes by k8s)

Via encrypted resources, later decrypted inside the k8s cluster

This approach is Kubernetes specific.

Bitnami’s sealed-secrets project provides a way to encrypt secrets hence allowing to commit them next to your other configuration. With this approach, secret values are not accessible to templates directly; however, they can be referenced by k8s resources (for example via secretRef in Pod spec). At the time of the deploy of such resources, sealed secrets controller will create appropriate Secret objects with decrypted secret values.

Pros:

  • provides a way to version (backup, etc) secrets next to other configuration
  • no way to access decrypted secret values in templates directly (though, this may not be necessary in many cases)

Cons:

  • Kubernetes specific
  • depending on organization requirements even encrypted secrets may not be allowed to be stored next to other configuration

(Help improve our docs: edit this page on GitHub)