Git Fileserver Backend Walkthrough

Note

This walkthrough assumes basic knowledge of Salt. To get up to speed, check out the Salt Walkthrough.

The gitfs backend allows Salt to serve files from git repositories. It can be enabled by adding git to the fileserver_backend list, and configuring one or more repositories in gitfs_remotes.

Branches and tags become Salt fileserver environments.

Note

Branching and tagging can result in a lot of potentially-conflicting top files, for this reason it may be useful to set top_file_merging_strategy to same in the minions' config files if the top files are being managed in a GitFS repo.

Installing Dependencies

Both pygit2 and GitPython are supported Python interfaces to git. If compatible versions of both are installed, pygit2 will be preferred. In these cases, GitPython can be forced using the gitfs_provider parameter in the master config file.

Note

It is recommended to always run the most recent version of any the below dependencies. Certain features of GitFS may not be available without the most recent version of the chosen library.

pygit2

The minimum supported version of pygit2 is 0.20.3. Availability for this version of pygit2 is still limited, though the SaltStack team is working to get compatible versions available for as many platforms as possible.

For the Fedora/EPEL versions which have a new enough version packaged, the following command would be used to install pygit2:

# yum install python-pygit2

Provided a valid version is packaged for Debian/Ubuntu (which is not currently the case), the package name would be the same, and the following command would be used to install it:

# apt-get install python-pygit2

If pygit2 is not packaged for the platform on which the Master is running, the pygit2 website has installation instructions here. Keep in mind however that following these instructions will install libgit2 and pygit2 without system packages. Additionally, keep in mind that SSH authentication in pygit2 requires libssh2 (not libssh) development libraries to be present before libgit2 is built. On some Debian-based distros pkg-config is also required to link libgit2 with libssh2.

Note

If you are receiving the error "Unsupported URL Protocol" in the Salt Master log when making a connection using SSH, review the libssh2 details listed above.

Additionally, version 0.21.0 of pygit2 introduced a dependency on python-cffi, which in turn depends on newer releases of libffi. Upgrading libffi is not advisable as several other applications depend on it, so on older LTS linux releases pygit2 0.20.3 and libgit2 0.20.0 is the recommended combination.

Warning

pygit2 is actively developed and frequently makes non-backwards-compatible API changes, even in minor releases. It is not uncommon for pygit2 upgrades to result in errors in Salt. Please take care when upgrading pygit2, and pay close attention to the changelog, keeping an eye out for API changes. Errors can be reported on the SaltStack issue tracker.

RedHat Pygit2 Issues

The release of RedHat/CentOS 7.3 upgraded both python-cffi and http-parser, both of which are dependencies for pygit2/libgit2. Both pygit2 and libgit2 packages (which are from the EPEL repository) should be upgraded to the most recent versions, at least to 0.24.2.

The below errors will show up in the master log if an incompatible python-pygit2 package is installed:

2017-02-10 09:07:34,892 [salt.utils.gitfs ][ERROR ][11211] Import pygit2 failed: CompileError: command 'gcc' failed with exit status 1
2017-02-10 09:07:34,907 [salt.utils.gitfs ][ERROR ][11211] gitfs is configured but could not be loaded, are pygit2 and libgit2 installed?
2017-02-10 09:07:34,907 [salt.utils.gitfs ][CRITICAL][11211] No suitable gitfs provider module is installed.
2017-02-10 09:07:34,912 [salt.master ][CRITICAL][11211] Master failed pre flight checks, exiting

The below errors will show up in the master log if an incompatible libgit2 package is installed:

2017-02-15 18:04:45,211 [salt.utils.gitfs ][ERROR   ][6211] Error occurred fetching gitfs remote 'https://foo.com/bar.git': No Content-Type header in response

A restart of the salt-master daemon and gitfs cache directory clean up may be required to allow http(s) repositories to continue to be fetched.

Debian Pygit2 Issues

The Debian repos currently have older versions of pygit2 (package python3-pygit2). These older versions may have issues using newer SSH keys (see [this issue](https://github.com/saltstack/salt/issues/61790)). Instead, pygit2 can be installed from Pypi, but you will need a version that matches the libgit2 version from Debian. This is version 1.6.1.

# apt-get install libgit2
# salt-pip install pygit2==1.6.1 --no-deps

Note that the above instructions assume a onedir installation. The need for --no-deps is to prevent the CFFI package from mismatching with Salt.

GitPython

GitPython 0.3.0 or newer is required to use GitPython for gitfs. For RHEL-based Linux distros, a compatible version is available in EPEL, and can be easily installed on the master using yum:

# yum install GitPython

Ubuntu 14.04 LTS and Debian Wheezy (7.x) also have a compatible version packaged:

# apt-get install python-git

GitPython requires the git CLI utility to work. If installed from a system package, then git should already be installed, but if installed via pip then it may still be necessary to install git separately. For MacOS users, GitPython comes bundled in with the Salt installer, but git must still be installed for it to work properly. Git can be installed in several ways, including by installing XCode.

Warning

GitPython advises against the use of its library for long-running processes (such as a salt-master or salt-minion). Please see their warning on potential leaks of system resources: https://github.com/gitpython-developers/GitPython#leakage-of-system-resources.

Warning

Keep in mind that if GitPython has been previously installed on the master using pip (even if it was subsequently uninstalled), then it may still exist in the build cache (typically /tmp/pip-build-root/GitPython) if the cache is not cleared after installation. The package in the build cache will override any requirement specifiers, so if you try upgrading to version 0.3.2.RC1 by running pip install 'GitPython==0.3.2.RC1' then it will ignore this and simply install the version from the cache directory. Therefore, it may be necessary to delete the GitPython directory from the build cache in order to ensure that the specified version is installed.

Warning

GitPython 2.0.9 and newer is not compatible with Python 2.6. If installing GitPython using pip on a machine running Python 2.6, make sure that a version earlier than 2.0.9 is installed. This can be done on the CLI by running pip install 'GitPython<2.0.9', or in a pip.installed state using the following SLS:

GitPython:
  pip.installed:
    - name: 'GitPython < 2.0.9'

Simple Configuration

To use the gitfs backend, only two configuration changes are required on the master:

  1. Include gitfs in the fileserver_backend list in the master config file:

    fileserver_backend:
      - gitfs
    

    Note

    git also works here. Prior to the 2018.3.0 release, only git would work.

  2. Specify one or more git://, https://, file://, or ssh:// URLs in gitfs_remotes to configure which repositories to cache and search for requested files:

    gitfs_remotes:
      - https://github.com/saltstack-formulas/salt-formula.git
    

    SSH remotes can also be configured using scp-like syntax:

    gitfs_remotes:
      - git@github.com:user/repo.git
      - ssh://user@domain.tld/path/to/repo.git
    

    Information on how to authenticate to SSH remotes can be found here.

  3. Restart the master to load the new configuration.

Note

In a master/minion setup, files from a gitfs remote are cached once by the master, so minions do not need direct access to the git repository.

Multiple Remotes

The gitfs_remotes option accepts an ordered list of git remotes to cache and search, in listed order, for requested files.

A simple scenario illustrates this cascading lookup behavior:

If the gitfs_remotes option specifies three remotes:

gitfs_remotes:
  - git://github.com/example/first.git
  - https://github.com/example/second.git
  - file:///root/third

And each repository contains some files:

first.git:
    top.sls
    edit/vim.sls
    edit/vimrc
    nginx/init.sls
    shell/init.sls

second.git:
    edit/dev_vimrc
    haproxy/init.sls
    shell.sls

third:
    haproxy/haproxy.conf
    edit/dev_vimrc

Salt will attempt to lookup the requested file from each gitfs remote repository in the order in which they are defined in the configuration. The git://github.com/example/first.git remote will be searched first. If the requested file is found, then it is served and no further searching is executed. For example:

  • A request for the file salt://haproxy/init.sls will be served from the https://github.com/example/second.git git repo.

  • A request for the file salt://haproxy/haproxy.conf will be served from the file:///root/third repo.

Also a requested state file overrules a directory with an init.sls-file. For example:

  • A request for state.apply shell will be served from the https://github.com/example/second.git git repo.

Note

This example is purposefully contrived to illustrate the behavior of the gitfs backend. This example should not be read as a recommended way to lay out files and git repos.

The file:// prefix denotes a git repository in a local directory. However, it will still use the given file:// URL as a remote, rather than copying the git repo to the salt cache. This means that any refs you want accessible must exist as local refs in the specified repo.

Warning

Salt versions prior to 2014.1.0 are not tolerant of changing the order of remotes or modifying the URI of existing remotes. In those versions, when modifying remotes it is a good idea to remove the gitfs cache directory (/var/cache/salt/master/gitfs) before restarting the salt-master service.

Per-remote Configuration Parameters

New in version 2014.7.0.

The following master config parameters are global (that is, they apply to all configured gitfs remotes):

Note

pygit2 only supports disabling SSL verification in versions 0.23.2 and newer.

These parameters can now be overridden on a per-remote basis. This allows for a tremendous amount of customization. Here's some example usage:

gitfs_provider: pygit2
gitfs_base: develop

gitfs_remotes:
  - https://foo.com/foo.git
  - https://foo.com/bar.git:
    - root: salt
    - mountpoint: salt://bar
    - base: salt-base
    - ssl_verify: False
    - update_interval: 120
  - https://foo.com/bar.git:
    - name: second_bar_repo
    - root: other/salt
    - mountpoint: salt://other/bar
    - base: salt-base
    - ref_types:
      - branch
  - http://foo.com/baz.git:
    - root: salt/states
    - user: joe
    - password: mysupersecretpassword
    - insecure_auth: True
    - disable_saltenv_mapping: True
    - saltenv:
      - foo:
        - ref: foo
  - http://foo.com/quux.git:
    - all_saltenvs: master

Important

There are two important distinctions which should be noted for per-remote configuration:

  1. The URL of a remote which has per-remote configuration must be suffixed with a colon.

  2. Per-remote configuration parameters are named like the global versions, with the gitfs_ removed from the beginning. The exception being the name, saltenv, and all_saltenvs parameters, which are only available to per-remote configurations.

The all_saltenvs parameter is new in the 2018.3.0 release.

In the example configuration above, the following is true:

  1. The first and fourth gitfs remotes will use the develop branch/tag as the base environment, while the second and third will use the salt-base branch/tag as the base environment.

  2. The first remote will serve all files in the repository. The second remote will only serve files from the salt directory (and its subdirectories). The third remote will only server files from the other/salt directory (and its subdirectories), while the fourth remote will only serve files from the salt/states directory (and its subdirectories).

  3. The third remote will only serve files from branches, and not from tags or SHAs.

  4. The fourth remote will only have two saltenvs available: base (pointed at develop), and foo (pointed at foo).

  5. The first and fourth remotes will have files located under the root of the Salt fileserver namespace (salt://). The files from the second remote will be located under salt://bar, while the files from the third remote will be located under salt://other/bar.

  6. The second and third remotes reference the same repository and unique names need to be declared for duplicate gitfs remotes.

  7. The fourth remote overrides the default behavior of not authenticating to insecure (non-HTTPS) remotes.

  8. Because all_saltenvs is configured for the fifth remote, files from the branch/tag master will appear in every fileserver environment.

    Note

    The use of http:// (instead of https://) is permitted here only because authentication is not being used. Otherwise, the insecure_auth parameter must be used (as in the fourth remote) to force Salt to authenticate to an http:// remote.

  9. The first remote will wait 120 seconds between updates instead of 60.

Per-Saltenv Configuration Parameters

New in version 2016.11.0.

For more granular control, Salt allows the following three things to be overridden for individual saltenvs within a given repo:

  • The mountpoint

  • The root

  • The branch/tag to be used for a given saltenv

Here is an example:

gitfs_root: salt

gitfs_saltenv:
  - dev:
    - mountpoint: salt://gitfs-dev
    - ref: develop

gitfs_remotes:
  - https://foo.com/bar.git:
    - saltenv:
      - staging:
        - ref: qa
        - mountpoint: salt://bar-staging
      - dev:
        - ref: development
  - https://foo.com/baz.git:
    - saltenv:
      - staging:
        - mountpoint: salt://baz-staging

Given the above configuration, the following is true:

  1. For all gitfs remotes, files for the dev saltenv will be located under salt://gitfs-dev.

  2. For the dev saltenv, files from the first remote will be sourced from the development branch, while files from the second remote will be sourced from the develop branch.

  3. For the staging saltenv, files from the first remote will be located under salt://bar-staging, while files from the second remote will be located under salt://baz-staging.

  4. For all gitfs remotes, and in all saltenvs, files will be served from the salt directory (and its subdirectories).

Custom Refspecs

New in version 2017.7.0.

GitFS will by default fetch remote branches and tags. However, sometimes it can be useful to fetch custom refs (such as those created for GitHub pull requests). To change the refspecs GitFS fetches, use the gitfs_refspecs config option:

gitfs_refspecs:
  - '+refs/heads/*:refs/remotes/origin/*'
  - '+refs/tags/*:refs/tags/*'
  - '+refs/pull/*/head:refs/remotes/origin/pr/*'
  - '+refs/pull/*/merge:refs/remotes/origin/merge/*'

In the above example, in addition to fetching remote branches and tags, GitHub's custom refs for pull requests and merged pull requests will also be fetched. These special head refs represent the head of the branch which is requesting to be merged, and the merge refs represent the result of the base branch after the merge.

Important

When using custom refspecs, the destination of the fetched refs must be under refs/remotes/origin/, preferably in a subdirectory like in the example above. These custom refspecs will map as environment names using their relative path underneath refs/remotes/origin/. For example, assuming the configuration above, the head branch for pull request 12345 would map to fileserver environment pr/12345 (slash included).

Refspecs can be configured on a per-remote basis. For example, the below configuration would only alter the default refspecs for the second GitFS remote. The first remote would only fetch branches and tags (the default).

gitfs_remotes:
  - https://domain.tld/foo.git
  - https://domain.tld/bar.git:
    - refspecs:
      - '+refs/heads/*:refs/remotes/origin/*'
      - '+refs/tags/*:refs/tags/*'
      - '+refs/pull/*/head:refs/remotes/origin/pr/*'
      - '+refs/pull/*/merge:refs/remotes/origin/merge/*'

Global Remotes

New in version 2018.3.0: for all_saltenvs, 3001 for fallback

The all_saltenvs per-remote configuration parameter overrides the logic Salt uses to map branches/tags to fileserver environments (i.e. saltenvs). This allows a single branch/tag to appear in all GitFS saltenvs.

Note

all_saltenvs only works within GitFS. That is, files in a branch configured using all_saltenvs will not show up in a fileserver environment defined via some other fileserver backend (e.g. file_roots).

The fallback global or per-remote configuration can also be used.

This is very useful in particular when working with salt formulas. Prior to the addition of this feature, it was necessary to push a branch/tag to the remote repo for each saltenv in which that formula was to be used. If the formula needed to be updated, this update would need to be reflected in all of the other branches/tags. This is both inconvenient and not scalable.

With all_saltenvs, it is now possible to define your formula once, in a single branch.

gitfs_remotes:
  - http://foo.com/quux.git:
    - all_saltenvs: anything

If you want to also test working branches of the formula repository, use fallback:

gitfs_remotes:
  - http://foo.com/quux.git:
    - fallback: anything

Update Intervals

Prior to the 2018.3.0 release, GitFS would update its fileserver backends as part of a dedicated "maintenance" process, in which various routine maintenance tasks were performed. This tied the update interval to the loop_interval config option, and also forced all fileservers to update at the same interval.

Now it is possible to make GitFS update at its own interval, using gitfs_update_interval:

gitfs_update_interval: 180

gitfs_remotes:
  - https://foo.com/foo.git
  - https://foo.com/bar.git:
    - update_interval: 120

Using the above configuration, the first remote would update every three minutes, while the second remote would update every two minutes.

Configuration Order of Precedence

The order of precedence for GitFS configuration is as follows (each level overrides all levels below it):

  1. Per-saltenv configuration (defined under a per-remote saltenv param)

    gitfs_remotes:
      - https://foo.com/bar.git:
        - saltenv:
          - dev:
            - mountpoint: salt://bar
    
  2. Global per-saltenv configuration (defined in gitfs_saltenv)

    gitfs_saltenv:
      - dev:
        - mountpoint: salt://bar
    
  3. Per-remote configuration parameter

    gitfs_remotes:
      - https://foo.com/bar.git:
        - mountpoint: salt://bar
    
  4. Global configuration parameter

    gitfs_mountpoint: salt://bar
    

Note

The one exception to the above is when all_saltenvs is used. This value overrides all logic for mapping branches/tags to fileserver environments. So, even if gitfs_saltenv is used to globally override the mapping for a given saltenv, all_saltenvs would take precedence for any remote which uses it.

It's important to note however that any root and mountpoint values configured in gitfs_saltenv (or per-saltenv configuration) would be unaffected by this.

Serving from a Subdirectory

The gitfs_root parameter allows files to be served from a subdirectory within the repository. This allows for only part of a repository to be exposed to the Salt fileserver.

Assume the below layout:

.gitignore
README.txt
foo/
foo/bar/
foo/bar/one.txt
foo/bar/two.txt
foo/bar/three.txt
foo/baz/
foo/baz/top.sls
foo/baz/edit/vim.sls
foo/baz/edit/vimrc
foo/baz/nginx/init.sls

The below configuration would serve only the files under foo/baz, ignoring the other files in the repository:

gitfs_remotes:
  - git://mydomain.com/stuff.git

gitfs_root: foo/baz

The root can also be configured on a per-remote basis.

Mountpoints

New in version 2014.7.0.

The gitfs_mountpoint parameter will prepend the specified path to the files served from gitfs. This allows an existing repository to be used, rather than needing to reorganize a repository or design it around the layout of the Salt fileserver.

Before the addition of this feature, if a file being served up via gitfs was deeply nested within the root directory (for example, salt://webapps/foo/files/foo.conf, it would be necessary to ensure that the file was properly located in the remote repository, and that all of the parent directories were present (for example, the directories webapps/foo/files/ would need to exist at the root of the repository).

The below example would allow for a file foo.conf at the root of the repository to be served up from the Salt fileserver path salt://webapps/foo/files/foo.conf.

gitfs_remotes:
  - https://mydomain.com/stuff.git

gitfs_mountpoint: salt://webapps/foo/files

Mountpoints can also be configured on a per-remote basis.

Using gitfs in Masterless Mode

Since 2014.7.0, gitfs can be used in masterless mode. To do so, simply add the gitfs configuration parameters (and set fileserver_backend) in the _minion_ config file instead of the master config file.

Using gitfs Alongside Other Backends

Sometimes it may make sense to use multiple backends; for instance, if sls files are stored in git but larger files are stored directly on the master.

The cascading lookup logic used for multiple remotes is also used with multiple backends. If the fileserver_backend option contains multiple backends:

fileserver_backend:
  - roots
  - git

Then the roots backend (the default backend of files in /srv/salt) will be searched first for the requested file; then, if it is not found on the master, each configured git remote will be searched.

Note

This can be used together with file_roots accepting __env__ as a catch-all environment, since 2018.3.5 and 2019.2.1:

file_roots:
  base:
    - /srv/salt
  __env__:
    - /srv/salt

Branches, Environments, and Top Files

When using the GitFS backend, branches, and tags will be mapped to environments using the branch/tag name as an identifier.

There is one exception to this rule: the master branch is implicitly mapped to the base environment.

So, for a typical base, qa, dev setup, the following branches could be used:

master
qa
dev

To map a branch other than master as the base environment, use the gitfs_base parameter.

gitfs_base: salt-base

The base can also be configured on a per-remote basis.

Use Case: Code Promotion (dev -> qa -> base)

When running a highstate, the top.sls files from all of the different branches and tags will be merged into one. This does not work well with the use case where changes are tested in development branches before being merged upstream towards production, because if the same SLS file from multiple environments is part of the highstate, it can result in non-unique state IDs, which will cause an error in the state compiler and not allow the highstate to proceed.

To accomplish this use case, you should do three things:

  1. Use {{ saltenv }} in place of your environment in your top.sls. This will let you use the same top file in all branches, because {{ saltenv }} gets replaced with the effective saltenv of the environment being processed.

  2. Set top_file_merging_strategy to same in the minion configuration. This will keep the base environment from looking at the top.sls from the dev or qa branches, etc.

  3. Explicitly define your saltenv. (More on this below.)

Consider the following example top file and SLS file:

top.sls

{{ saltenv }}:
  '*':
    - mystuff

mystuff.sls

manage_mystuff:
  pkg.installed:
    - name: mystuff
  file.managed:
    - name: /etc/mystuff.conf
    - source: salt://mystuff/files/mystuff.conf
  service.running:
    - name: mystuffd
    - enable: True
    - watch:
      - file: /etc/mystuff.conf

Imagine for a moment that you need to change your mystuff.conf. So, you go to your dev branch, edit mystuff/files/mystuff.conf, and commit and push.

If you have only done the first two steps recommended above, and you run your highstate, you will end up with conflicting IDs:

myminion:
    Data failed to compile:
----------
    Detected conflicting IDs, SLS IDs need to be globally unique.
    The conflicting ID is 'manage_mystuff' and is found in SLS 'base:mystuff' and SLS 'dev:mystuff'
----------
    Detected conflicting IDs, SLS IDs need to be globally unique.
    The conflicting ID is 'manage_mystuff' and is found in SLS 'dev:mystuff' and SLS 'qa:mystuff'

This is because, in the absence of an explicit saltenv, all environments' top files are considered. Each environment looks at only its own top.sls, but because the mystuff.sls exists in each branch, they all get pulled into the highstate, resulting in these conflicting IDs. This is why explicitly setting your saltenv is important for this use case.

There are two ways of explicitly defining the saltenv:

  1. Set the saltenv in your minion configuration file. This allows you to isolate which states are run to a specific branch/tag on a given minion. This also works nicely if you have different salt deployments for dev, qa, and prod. Boxes in dev can have saltenv set to dev, boxes in qa can have the saltenv set to qa, and boxes in prod can have the saltenv set to base.

  2. At runtime, you can set the saltenv like so:

    salt myminion state.apply saltenv=dev
    

    A couple notes about setting the saltenv at runtime:

    • It will take precedence over the saltenv setting from the minion config file, and pairs nicely with cases where you do not have separate salt deployments for dev/qa/prod. You can have a box with saltenv set to base, which you can test your dev changes on by running your state.apply with saltenv=dev.

    • If you don't set saltenv in the minion config file, you _must_ specify it at runtime to avoid conflicting IDs.

If you branched qa off of master, and dev off of qa, you can merge changes from dev into qa, and then merge qa into master to promote your changes to from dev to qa to prod.

Environment Whitelist/Blacklist

New in version 2014.7.0.

The gitfs_saltenv_whitelist and gitfs_saltenv_blacklist parameters allow for greater control over which branches/tags are exposed as fileserver environments. Exact matches, globs, and regular expressions are supported, and are evaluated in that order. If using a regular expression, ^ and $ must be omitted, and the expression must match the entire branch/tag.

gitfs_saltenv_whitelist:
  - base
  - v1.*
  - 'mybranch\d+'

Note

v1.*, in this example, will match as both a glob and a regular expression (though it will have been matched as a glob, since globs are evaluated before regular expressions).

The behavior of the blacklist/whitelist will differ depending on which combination of the two options is used:

  • If only gitfs_saltenv_whitelist is used, then only branches/tags which match the whitelist will be available as environments

  • If only gitfs_saltenv_blacklist is used, then the branches/tags which match the blacklist will not be available as environments

  • If both are used, then the branches/tags which match the whitelist, but do not match the blacklist, will be available as environments.

Authentication

pygit2

New in version 2014.7.0.

Both HTTPS and SSH authentication are supported as of version 0.20.3, which is the earliest version of pygit2 supported by Salt for gitfs.

Note

The examples below make use of per-remote configuration parameters, a feature new to Salt 2014.7.0. More information on these can be found here.

HTTPS

For HTTPS repositories which require authentication, the username and password can be provided like so:

gitfs_remotes:
  - https://domain.tld/myrepo.git:
    - user: git
    - password: mypassword

If the repository is served over HTTP instead of HTTPS, then Salt will by default refuse to authenticate to it. This behavior can be overridden by adding an insecure_auth parameter:

gitfs_remotes:
  - http://domain.tld/insecure_repo.git:
    - user: git
    - password: mypassword
    - insecure_auth: True

SSH

SSH repositories can be configured using the ssh:// protocol designation, or using scp-like syntax. So, the following two configurations are equivalent:

  • ssh://git@github.com/user/repo.git

  • git@github.com:user/repo.git

Both gitfs_pubkey and gitfs_privkey (or their per-remote counterparts) must be configured in order to authenticate to SSH-based repos. If the private key is protected with a passphrase, it can be configured using gitfs_passphrase (or simply passphrase if being configured per-remote). For example:

gitfs_remotes:
  - git@github.com:user/repo.git:
    - pubkey: /root/.ssh/id_rsa.pub
    - privkey: /root/.ssh/id_rsa
    - passphrase: myawesomepassphrase

Finally, the SSH host key must be added to the known_hosts file.

Note

There is a known issue with public-key SSH authentication to Microsoft Visual Studio (VSTS) with pygit2. This is due to a bug or lack of support for VSTS in older libssh2 releases. Known working releases include libssh2 1.7.0 and later, and known incompatible releases include 1.5.0 and older. At the time of this writing, 1.6.0 has not been tested.

Since upgrading libssh2 would require rebuilding many other packages (curl, etc.), followed by a rebuild of libgit2 and a reinstall of pygit2, an easier workaround for systems with older libssh2 is to use GitPython with a passphraseless key for authentication.

GitPython

HTTPS

For HTTPS repositories which require authentication, the username and password can be configured in one of two ways. The first way is to include them in the URL using the format https://<user>:<password>@<url>, like so:

gitfs_remotes:
  - https://git:mypassword@domain.tld/myrepo.git

The other way would be to configure the authentication in /var/lib/salt/.netrc:

machine domain.tld
login git
password mypassword

If the repository is served over HTTP instead of HTTPS, then Salt will by default refuse to authenticate to it. This behavior can be overridden by adding an insecure_auth parameter:

gitfs_remotes:
  - http://git:mypassword@domain.tld/insecure_repo.git:
    - insecure_auth: True

SSH

Only passphrase-less SSH public key authentication is supported using GitPython. The auth parameters (pubkey, privkey, etc.) shown in the pygit2 authentication examples above do not work with GitPython.

gitfs_remotes:
  - ssh://git@github.com/example/salt-states.git

Since GitPython wraps the git CLI, the private key must be located in ~/.ssh/id_rsa for the user under which the Master is running, and should have permissions of 0600. Also, in the absence of a user in the repo URL, GitPython will (just as SSH does) attempt to login as the current user (in other words, the user under which the Master is running, usually root).

If a key needs to be used, then ~/.ssh/config can be configured to use the desired key. Information on how to do this can be found by viewing the manpage for ssh_config. Here's an example entry which can be added to the ~/.ssh/config to use an alternate key for gitfs:

Host github.com
    IdentityFile /root/.ssh/id_rsa_gitfs

The Host parameter should be a hostname (or hostname glob) that matches the domain name of the git repository.

It is also necessary to add the SSH host key to the known_hosts file. The exception to this would be if strict host key checking is disabled, which can be done by adding StrictHostKeyChecking no to the entry in ~/.ssh/config

Host github.com
    IdentityFile /root/.ssh/id_rsa_gitfs
    StrictHostKeyChecking no

However, this is generally regarded as insecure, and is not recommended.

Adding the SSH Host Key to the known_hosts File

To use SSH authentication, it is necessary to have the remote repository's SSH host key in the ~/.ssh/known_hosts file. If the master is also a minion, this can be done using the ssh.set_known_host function:

# salt mymaster ssh.set_known_host user=root hostname=github.com
mymaster:
    ----------
    new:
        ----------
        enc:
            ssh-rsa
        fingerprint:
            16:27:ac:a5:76:28:2d:36:63:1b:56:4d:eb:df:a6:48
        hostname:
            |1|OiefWWqOD4kwO3BhoIGa0loR5AA=|BIXVtmcTbPER+68HvXmceodDcfI=
        key:
            AAAAB3NzaC1yc2EAAAABIwAAAQEAq2A7hRGmdnm9tUDbO9IDSwBK6TbQa+PXYPCPy6rbTrTtw7PHkccKrpp0yVhp5HdEIcKr6pLlVDBfOLX9QUsyCOV0wzfjIJNlGEYsdlLJizHhbn2mUjvSAHQqZETYP81eFzLQNnPHt4EVVUh7VfDESU84KezmD5QlWpXLmvU31/yMf+Se8xhHTvKSCZIFImWwoG6mbUoWf9nzpIoaSjB+weqqUUmpaaasXVal72J+UX2B+2RPW3RcT0eOzQgqlJL3RKrTJvdsjE3JEAvGq3lGHSZXy28G3skua2SmVi/w4yCE6gbODqnTWlg7+wC604ydGXA8VJiS5ap43JXiUFFAaQ==
    old:
        None
    status:
        updated

If not, then the easiest way to add the key is to su to the user (usually root) under which the salt-master runs and attempt to login to the server via SSH:

$ su -
Password:
# ssh github.com
The authenticity of host 'github.com (192.30.252.128)' can't be established.
RSA key fingerprint is 16:27:ac:a5:76:28:2d:36:63:1b:56:4d:eb:df:a6:48.
Are you sure you want to continue connecting (yes/no)? yes
Warning: Permanently added 'github.com,192.30.252.128' (RSA) to the list of known hosts.
Permission denied (publickey).

It doesn't matter if the login was successful, as answering yes will write the fingerprint to the known_hosts file.

Verifying the Fingerprint

To verify that the correct fingerprint was added, it is a good idea to look it up. One way to do this is to use nmap:

$ nmap -p 22 github.com --script ssh-hostkey

Starting Nmap 5.51 ( http://nmap.org ) at 2014-08-18 17:47 CDT
Nmap scan report for github.com (192.30.252.129)
Host is up (0.17s latency).
Not shown: 996 filtered ports
PORT     STATE SERVICE
22/tcp   open  ssh
| ssh-hostkey: 1024 ad:1c:08:a4:40:e3:6f:9c:f5:66:26:5d:4b:33:5d:8c (DSA)
|_2048 16:27:ac:a5:76:28:2d:36:63:1b:56:4d:eb:df:a6:48 (RSA)
80/tcp   open  http
443/tcp  open  https
9418/tcp open  git

Nmap done: 1 IP address (1 host up) scanned in 28.78 seconds

Another way is to check one's own known_hosts file, using this one-liner:

$ ssh-keygen -l -f /dev/stdin <<<`ssh-keyscan github.com 2>/dev/null` | awk '{print $2}'
16:27:ac:a5:76:28:2d:36:63:1b:56:4d:eb:df:a6:48

Warning

AWS tracks usage of nmap and may flag it as abuse. On AWS hosts, the ssh-keygen method is recommended for host key verification.

Note

As of OpenSSH 6.8 the SSH fingerprint is now shown as a base64-encoded SHA256 checksum of the host key. So, instead of the fingerprint looking like 16:27:ac:a5:76:28:2d:36:63:1b:56:4d:eb:df:a6:48, it would look like SHA256:nThbg6kXUpJWGl7E1IGOCspRomTxdCARLviKw6E5SY8.

Refreshing gitfs Upon Push

By default, Salt updates the remote fileserver backends every 60 seconds. However, if it is desirable to refresh quicker than that, the Reactor System can be used to signal the master to update the fileserver on each push, provided that the git server is also a Salt minion. There are three steps to this process:

  1. On the master, create a file /srv/reactor/update_fileserver.sls, with the following contents:

    update_fileserver:
      runner.fileserver.update
    
  2. Add the following reactor configuration to the master config file:

    reactor:
      - 'salt/fileserver/gitfs/update':
        - /srv/reactor/update_fileserver.sls
    
  3. On the git server, add a post-receive hook

    1. If the user executing git push is the same as the minion user, use the following hook:

    #!/usr/bin/env sh
    salt-call event.fire_master update salt/fileserver/gitfs/update
    
    1. To enable other git users to run the hook after a push, use sudo in the hook script:

    #!/usr/bin/env sh
    sudo -u root salt-call event.fire_master update salt/fileserver/gitfs/update
    
  4. If using sudo in the git hook (above), the policy must be changed to permit all users to fire the event. Add the following policy to the sudoers file on the git server.

    Cmnd_Alias SALT_GIT_HOOK = /bin/salt-call event.fire_master update salt/fileserver/gitfs/update
    Defaults!SALT_GIT_HOOK !requiretty
    ALL ALL=(root) NOPASSWD: SALT_GIT_HOOK
    

The update argument right after event.fire_master in this example can really be anything, as it represents the data being passed in the event, and the passed data is ignored by this reactor.

Similarly, the tag name salt/fileserver/gitfs/update can be replaced by anything, so long as the usage is consistent.

The root user name in the hook script and sudo policy should be changed to match the user under which the minion is running.

Using Git as an External Pillar Source

The git external pillar (a.k.a. git_pillar) has been rewritten for the 2015.8.0 release. This rewrite brings with it pygit2 support (allowing for access to authenticated repositories), as well as more granular support for per-remote configuration. This configuration schema is detailed here.

Why aren't my custom modules/states/etc. syncing to my Minions?

In versions 0.16.3 and older, when using the git fileserver backend, certain versions of GitPython may generate errors when fetching, which Salt fails to catch. While not fatal to the fetch process, these interrupt the fileserver update that takes place before custom types are synced, and thus interrupt the sync itself. Try disabling the git fileserver backend in the master config, restarting the master, and attempting the sync again.

This issue is worked around in Salt 0.16.4 and newer.