...
Jira Legacy | ||||||||
---|---|---|---|---|---|---|---|---|
|
Managing Multiple Puppet environments (without SIMP)
Control repositories
Outside of SIMP, Gitgit-based control repositories have been are the dominant technology to manage and deploy Puppet code since 2015.
At a high-level, this a control repository workflow involves:
- A Puppetfile that defines a collection of Puppet modules and where to get them
- A (Git) Control repository, where each branch defines a complete Puppet environment (which includes a Puppetfile)
- A tool (r10k or PE Code Manager) to deploy Puppet environments from the control repository branches
- (Advanced, but preferred) Repository webhooks that automatically trigger environment deployments on the Puppet master(s) whenever a branch is updated.
The upshot is that a control repository branch defines everything needed to recreate a specific Puppet environment.
Puppet environments as code
A branch contains Defining everything required needed to recreate create a Puppet environment , so allows Puppet Environments can to be treated handled like code:
- (Using r10k/Code Manager,) Adding or removing a git branch in the control repository will add or remove that environment on the Puppet master(s)
- (using r10k/Code Manager + webhook triggers)
- The Puppet environment updates change control process can be submitted as handled as merge/pull requests
- User Developers can test, diff, and review Puppet environment changes before promoting them into important environments
- When As control repo branches are updated, CI pipelines can automatically trigger tests and master-side deployments
Multiple Puppet environments
...
Since Because 1 control repo branch = 1 Puppet environment, managing multiple Puppet environments is as simple as using git:
- To create a Fork an existing branch to create a new temporary testing (canary) environment, fork an existing branch
- Once After the canary branch is validated, merge it back into the original
- Control write Write access to important permanent environments by can be controlled by protecting their branches.
Control repos + Puppet deployment tools enable advanced environment scenarios:
...
Scaling Puppet with Control Repository-based workflows
- r10k, Code Manager :) Deploy make it easy to deploy consistent Puppet environments across Puppet architectures that involve multiple masters
- (Bolt:) Manage Bolt makes it possible to manage remote "agentless" hosts by checking out an environment's modules (using
bolt puppetfile install
) before runningbolt apply
...
SIMP + Multiple Puppet environments
...
Anchor | ||||
---|---|---|---|---|
|
Limitations and conflicts
It has been technically possible to manage use multiple Puppet environments from a and control repository repositories since SIMP 5.0 (
Status | ||||||
---|---|---|---|---|---|---|
|
However, in practice this has been cumbersome , risky, and easy to get wrong. Getting it wrong can accidentally leak or lose site data (such as passgen secrets and PKI files). Getting it right requires an in-depth knowledge of Puppet environments and SIMP's "extra" environments. Even then, there are some significant limitations—for instance, it's not possible to pool redundant (compile) masters behind a VIP unless site administrators implement their solution to keep the SIMP "extra" environment data in sync across each server.
Issue 1: SIMP RPMs and tools interfere with files under the Puppet and SIMP environment directories
The specifics of this issue have changed over the years, and were largely "solved" when SIMP 6.4+ stopped deploying RPMs into environment directories.
Starting with the 6.4.0 release (
Status | ||||||
---|---|---|---|---|---|---|
|
r10k puppetfile install
from a Puppet environment directory) . However, there are still two major sources of conflicts with multiple Puppet environments and SIMP:Conflicts with Control Repositories
SIMP automation can alter files under the into a single, permanent Puppet environment (i.e., production
).
RPM updates andStatus colour Red title SIMP 5.0-6.3 simp_rpm_helper
could modify existing files Puppet and SIMP environment directoriesStatus colour
...
Red title SIMP 5.0-6.3
...
- User-initiated tools like
simp
...
config could modify existing files Puppet and SIMP environment directories
Status colour Blue title SIMP 6.4
...
The user-initiated toolStatus colour Yellow title PARTIAL FIX simp environment new
can ensure that there is a corresponding secondary and writable environments- This is a safety improvement from earlier releases, because it only alters environment files
...
Conflicts with Multiple Puppet Environments
...
- when the user initiates it
- However, the
simp environment
tool is only partially implemented, and the remaining actions cannot be implemented safely. - As things stand, this means that SIMP 6.4.0:
- Safely supports SIMP extra data in the Local (Puppetfile-only) deployment scenario with a single environment (production)
- Does not safely support SIMP extra data in the Control Repository deployment scenarios (without additional conventions and limitations)
Issue 2: SIMP expects Secondary and Writable asset paths for each Puppet environment
SIMP has always assumed that each Puppet environment directory will be accompanied by two SIMP-specific "environment" data directories:
"Secondary" data/assets underStatus colour Red title SIMP (ALL) /var/simp/environments/$environment/.
These are files too sensitive and/or too large to check into git in plaintext, maintained by site admins.- "Site Files" modules, like
pki_files
(keydist) andkrb3_files
. - The FakeCA support script, including the FakeCA's private key
- RSync directories, used by the
simp::server::rsync_shares
profile to serve files for various modules.
- "Site Files" modules, like
"Writable" data/assets under underStatus colour Red title SIMP (ALL) /opt/puppetlabs/server/data/puppetserver/simp/environments/$environment/
Earlier SIMP releases
In earlier SIMP releases (5.0–6.3), it was technically possible to deploy multiple environments from a control repository—but in practice, it was cumbersome and risky:
- Whenever a new Puppet environment was deployed, (manual) follow-on actions were required to ensure the appropriate SIMP data was available for that environment
- RPM updates and
simp_rpm_helper
could modify existing files Puppet and SIMP environment directories - It assumed an in-depth knowledge of Puppet and SIMP
- It still suffered from significant limitations and risks
...
- These are secrets, read by special SIMP Puppet functions while compiling catalogs (and automatically generated if they are missing).
There are currently only two Puppet functions in SIMP that use the writeable environment directory:simplib::passgen()
- a popular function, used to generate secret passwords
- Used in 14 SIMP modules
- Usually (but not always) exposed as a parameter default
- Secrets read by this function can also be generated by the
simp passgen
CLI tool
ssh::autokey()
—this function is unused by SIMP modules. It is provided to allow local admins a way to automatically arrange SSH keys for service accounts.
SIMP running a single Puppet environment (safe)
A stock SIMP install starts with a unified SIMP Omni-Environment, production.
Issues with SIMP and multiple Puppet environments
After deploying a new Puppet environment (new_env, forked from production), there are several problems:
- (Red lines) The
modulepath
setting inenvironment.conf
still points to production.- Puppet catalogs compiled in new_env still source secondary module data (FakeCA PKI, Kerberos) from
production
! - Any pki/krb3 files under
/var/simp/environments/new_env/site_files
are never used.
- Puppet catalogs compiled in new_env still source secondary module data (FakeCA PKI, Kerberos) from
- (Blue lines) the SIMP "extra" ("secondary" and "writable") environment directories don't exist for new_env yet!
- Puppet code using SIMP's
rsync
type will fail in the new_env environment, because the source path will not exist.- (The
rsync
type is used in 14 SIMP modules)
- (The
-
simplib::passgen
() will silently create new and different secrets for each identifier in the new_env environment- This breaks authentication with passgen-configured accounts/services still in production.
- e.g., TPM/TPM2 owner authentication, kdb5 passwords, rsync servers, SIMP GitLab auth
- This is especially destructive to Canary nodes'—after successfully testing a new account/service that uses passgen-configured credentials, authentication will break after the node is returned to production.
- This breaks authentication with passgen-configured accounts/services still in production.
- Puppet code using SIMP's
Environment safety improvements in SIMP 6.4.0
SIMP 6.4.0 addressed many of SIMP's tool and RPM-related problems by making Puppetfile-based module deployments the preferred method to deploy modules.
- Puppetfile-based deployment tools like r10k or Code Manager and now preferred to deploy SIMP's Puppet modules
- To prevent conflicts with these tools, SIMP no longer interferes with any files under the Puppet or SIMP environment directories, unless directed to by the user.
- SIMP module RPMs now install tagged versions into local git repositories, for use in local Puppetfiles
- The
simp puppetfile
tool was introduced to automatically generate Puppetfiles based on the current module RPMs.
- The
- The
simp environment new
tool was introduced to help users ensure that corresponding SIMP and Puppet environments are created.
Improvements focused on making it simple to use Puppetfiles in a single environment, with multiple
...
Safely Referencing SIMP assets from multiple Puppet Environments
Anchor | ||||
---|---|---|---|---|
|
Status | ||||||
---|---|---|---|---|---|---|
|
Status | ||||
---|---|---|---|---|
|
simp environment new
Site admins must ensure that SIMP "extra" environment directories exist for every Puppet environment. There are basically three strategies for this:
- Create a new "clean" SIMP extra environment directories for the new environment
- Link to another SIMP extra environment to keep all assets the same
- pros: keeps environments'
- cons: linked directories can fool admins into thinking it is safe to alter/remove data from a
- Copy directories from another SIMP extra environment
These choices in have been automated in SIMP 6.4.0 by the simp environment new
tool.
Advantages
simp environment new
automates the error-prone process of ensuring all three members of the SIMP Omni-environment exist on a local Puppet master
Drawbacks
- Automates a mis-modeled workaround (devised long ago, when Puppet environments were new)
- Doesn't make SIMP assets safer or easier to manage over time
- Can't work across multiple Puppet masters
- Can't be implemented to safely remove extra environments
During the development of SIMP 6.4.0, it became apparent that the simp environment
workflow would have problems down the road:
- Making
Requiring a
newSIMP writable and secondary environment to exist for
everevery Puppet environment was
arbitrary The dependence on$environment
made it necessary to adopta mistake.
It requires coarse workarounds like `
simp environment new [--copy|--link
]`every time a new Puppet environment is deployed
even when environments used the same (or similar) resources
- Linked
It prevents referring to a mix of the same assets in some cases
It adds a source of truth that is independent from the control repository
Linked SIMP extra environments make it easy to assume that it is safe to alter/delete assets, because the path shows an unimportant name
The proposed
from linked environments with the proposedsimp environment rm
command would make it too easy to permanently lose datasimp environment rm
command
Puppet Environments and SIMP Assets
SIMP 6.4
Proposed SIMP 6.5+
Legacy 6.4 assets in SIMP 6.5+
...
in linked environments
Anchor se02 se02
SE02. Status colour Blue title SIMP 5.0-6.3
(Good:) Using hiera-eyaml in the control repo to replace Writeable environment data
se02 | |
se02 |
colour | Blue |
---|---|
title | SIMP 5.0-6.3 |
Site admins can prevent SIMP from using the Writable
environment directory by overriding all uses of simplib::passgen()
with the Hiera eyaml backend.
Advantages
- Secret data now scales with additional compile masters as part of the r10k/code manager deployment
- Secret data is encrypted and versioned
- PCKS7 and GPG are supported
- It is simple to configure multiple Hiera eyaml backends tiers, with separate keys
- Separate teams can encrypt their secrets with their own key
Drawbacks
- Site administrators are responsible to manage and distribute the hiera-eyaml key files (independently of SIMP)
- key files must exist at the paths
hiera.yaml
expects them on compile masters
- key files must exist at the paths
- This approach is not possible for some SIMP users
- The SIMP codebase doesn't expose every use of of
simplib::passgen()
as an overridable parameter - Depending on the way your site is configured, you may not be avoid SIMP code
- The SIMP codebase doesn't expose every use of of
Suggested improvements
Expose all uses ofStatus colour Red title FUTURE simp::passgen()
in SIMP classes as parameter defaults, so users can override them viahiera-eyaml
Examples
Example tiers from control repository's
hiera.yaml
:No Format --- version: 5 hierarchy: # [...] # ---------------------------------------------------------------------------- # NOTE: This tier determines which secrets to use via the top-scope variable # `$::hostgroup` (set by ENC or manifests/site.pp, prior to any lookups) # ---------------------------------------------------------------------------- - name: "Per-hostgroup data (encrypted)" lookup_key: eyaml_lookup_key path: "secrets/hostgroups/%{::hostgroup}.eyaml" options: pkcs7_private_key: /etc/puppetlabs/puppet/eyaml/hostgroups/%{::hostgroup}__private_key.pkcs7.pem pkcs7_public_key: /etc/puppetlabs/puppet/eyaml/hostgroups/%{::hostgroup}__public_key.pkcs7.pem # ---------------------------------------------------------------------------- # WARNING: The interpolations in this tier rely on an agent-determined fact # (`%{facts.datacenter}`) to determine which secrets will be returned # during lookups. # # When possible, it is STRONGLY recommended to use *trusted* facts # instead of agent-determined facts for this purpose. # (i.e., `%{trusted.datacenter}` instead of `%{facts.datacenter}`). # # However, this requires either: # # 1. Baking the information into the Puppet certificate at the time # it is signed, with CSR attributes/certificate extensions and # setting short names in the `config_file_oid_map.yaml` file: # # * https://puppet.com/docs/puppet/5.5/config_file_oid_map.html # * https://puppet.com/docs/puppet/5.5/ssl_attributes_extensions.html # # 2. Use the (experimental as of Puppet 6.11) `trusted_external_command` # setting to add trusted facts from an external source during # catalog compilation. # # * https://tickets.puppetlabs.com/browse/PUP-9994 # * https://puppet.com/docs/puppet/latest/release_notes_puppet.html#experimental-feature:-catalog-compilation-with-external-trusted-data-from-third-parties # # ---------------------------------------------------------------------------- - name: "Per-datacenter secret data (encrypted)" lookup_key: eyaml_lookup_key path: "secrets/datacenters/%{facts.datacenter}.eyaml" options: pkcs7_private_key: /etc/puppetlabs/puppet/eyaml/datacenters/%{facts.datacenter}__private_key.pkcs7.pem pkcs7_public_key: /etc/puppetlabs/puppet/eyaml/datacenters/%{facts.datacenter}__public_key.pkcs7.pem - name: "Site-wide secret data (encrypted)" lookup_key: eyaml_lookup_key path: "secrets/site.eyaml" options: pkcs7_private_key: /etc/puppetlabs/puppet/eyaml/site__private_key.pkcs7.pem pkcs7_public_key: /etc/puppetlabs/puppet/eyaml/site__public_key.pkcs7.pem # [...]
Example encrypted data in `data/secrets/site.eyaml`:
No Format --- simp::puppetdb::database_password : > ENC[PKCS7,Y22exl+OvjDe+drmik2XEeD3VQtl1uZJXFFF2NnrMXDWx0csyqLB/2NOWefv NBTZfOlPvMlAesyr4bUY4I5XeVbVk38XKxeriH69EFAD4CahIZlC8lkE/uDh jJGQfh052eonkungHIcuGKY/5sEbbZl/qufjAtp/ufor15VBJtsXt17tXP4y l5ZP119Fwq8xiREGOL0lVvFYJz2hZc1ppPCNG5lwuLnTekXN/OazNYpf4CMd /HjZFXwcXRtTlzewJLc+/gox2IfByQRhsI/AgogRfYQKocZgFb/DOZoXR7wm IZGeunzwhqfmEtGiqpvJJQ5wVRdzJVpTnANBA5qxeA==] krb5::kdc::config::kdb5_password: > ENC[PKCS7,Y22exl+OvjDe+drmik2XEeD3VQtl1uZJXFFF2NnrMXDWx0csyqLB/2NOWefv NBTZfOlPvMlAesyr4bUY4I5XeVbVk38XKxeriH69EFAD4CahIZlC8lkE/uDh jJGQfh052eonkungHIcuGKY/5sEbbZl/qufjAtp/ufor15VBJtsXt17tXP4y l5ZP119Fwq8xiREGOL0lVvFYJz2hZc1ppPCNG5lwuLnTekXN/OazNYpf4CMd /HjZFXwcXRtTlzewJLc+/gox2IfByQRhsI/AgogRfYQKocZgFb/DOZoXR7wm IZGeunzwhqfmEtGiqpvJJQ5wVRdzJVpTnANBA5qxeA==] # ...
- For more details, see:
- The Puppet Hiera hierarchy level documentation at https://puppet.com/docs/puppet/5.5/hiera_config_yaml_5.html
- The hiera-eyaml project documentation at https://github.com/voxpupuli/hiera-eyaml