The documentation explains the use of a map.jinja
to gather parameters values for a formula.
As pillars are rendered on the Salt master for every minion, this increases the load on the master as the pillar values and the number of minions grows.
As a good practice, you should:
- store non-secret data in YAML files distributed by the fileserver
- store secret data in:
- pillars (and look for the use of something like pillar.vault)
- SDB (and look for the use of something like sdb.vault)
Current best practice is to let map.jinja
handle parameters from all sources, to minimise the use of pillars, grains or configuration from sls
files and templates directly.
Table of Contents
We will see a quick setup to configure the rabbitmq
formula for different DNS domain names and several roles.
For this example, I'll define 2 kinds of fileserver sources:
- formulas git repositories with hard-coded version reference to avoid breaking my setup randomly at upstream update. they are the last sources where files are looked up
- parameters of the formulas in the file backend roots
I configure the fileserver backends to serve:
Create the file /etc/salt/master.d/fileserver.conf
and restart the master
:
---
##
## file server
##
fileserver_backend:
# parameters values and override
- roots
# formulas
- gitfs
# The files in this directory will take precedence over git repositories
file_roots:
base:
- /srv/salt
# List of formulas I'm using
gitfs_remotes:
- https://github.com/saltstack-formulas/rabbitmq-formula.git:
- base: v4.1.1
- https://github.com/saltstack-formulas/openssh-formula.git:
- base: v2.0.1
...
Now, we can provides the per DNS domain name configuration files for the rabbitmq
formulas under /srv/salt/rabbitmq/parameters/
.
We create the directory for dns:domain
grain and we add a symlink for the domain
grain which is extracted from the minion id
:
mkdir -p /srv/salt/rabbitmq/parameters/dns:domain/
ln -s dns:domain /srv/salt/rabbitmq/parameters/domain
We create a configuration for the DNS domain example.net
in /srv/salt/rabbitmq/parameters/dns:domain/example.net.yaml
:
---
values:
config:
name: /etc/rabbitmq-formula-example-net.conf
...
We create another configuration for the DNS domain example.com
in the Jinja YAML template /srv/salt/rabbitmq/parameters/dns:domain/example.com.yaml.jinja
:
---
values:
config:
name: /etc/rabbitmq-formula-{{ grains['os_family'] }}.conf
...
Now, we can provides the per role configuration files for the rabbitmq
formulas under /srv/salt/rabbitmq/parameters/
.
We create the directory for roles:
mkdir -p /srv/salt/rabbitmq/parameters/roles
We will define 2 roles:
rabbitmq/server
rabbitmq/client
We create a configuration for the role rabbitmq/server
in /srv/salt/rabbitmq/parameters/roles/rabbitmq/server.yaml
:
---
values:
config:
name: /etc/rabbitmq-formula-server.conf
...
We create another configuration for the role rabbitmq/client
in /srv/salt/rabbitmq/parameters/roles/rabbitmq/client.yaml
:
---
values:
config:
name: /etc/rabbitmq-formula-client.conf
...
We need to redefine the sources for map.jinja
to load values from our new configuration files, we provide a global configuration for all our minions.
We create the global parameters file /srv/salt/parameters/map_jinja.yaml
:
---
values:
sources:
# default values
- "Y:G@osarch"
- "Y:G@os_family"
- "Y:G@os"
- "Y:G@osfinger"
- "C@{{ tplroot ~ ':lookup' }}"
- "C@{{ tplroot }}"
# Roles activate/deactivate things
# then thing are configured depending on environment
# So roles comes before `dns:domain`, `domain` and `id`
- "Y:C@roles"
# DNS domain configured (DHCP or resolv.conf)
- "Y:G@dns:domain"
# Based on minion ID
- "Y:G@domain"
# default values
- "Y:G@id"
...
The syntax is explained later at Sources of configuration values.
We associate roles grains to minion using grains.append.
For the servers:
salt 'server-*' grains.append roles rabbitmq/server
For the clients:
salt 'client-*' grains.append roles rabbitmq/client
Note
Since we used Y:C@roles
, map.jinja
will do a salt['config.get']('roles')
to retrieve the roles so you could use any other method to bind roles to minions (pillars or SDB) but grains seems to be the preferred method.
If you have a minion running under windows, you can't use colon :
as a delimiter for grain path query (see bug 58726) in which case you should use an alternate delimiter:
Modify /srv/salt/parameters/map_jinja.yaml
to change the query for dns:domain
to define the alternate delimiter:
---
values:
sources:
# default values
- "Y:G@osarch"
- "Y:G@os_family"
- "Y:G@os"
- "Y:G@osfinger"
- "C@{{ tplroot ~ ':lookup' }}"
- "C@{{ tplroot }}"
# Roles activate/deactivate things
# then thing are configured depending on environment
# So roles comes before `dns:domain`, `domain` and `id`
- "Y:C@roles"
# DNS domain configured (DHCP or resolv.conf)
- "Y:G:!@dns!domain"
# Based on minion ID
- "Y:G@domain"
# default values
- "Y:G@id"
...
And then, rename the directory:
mv /srv/salt/rabbitmq/parameters/dns:domain/ '/srv/salt/rabbitmq/parameters/dns!domain/'
When you write a new YAML file, note that it must conform to the following layout:
- a mandatory
values
key to store the configuration values - two optional keys to configure the use of salt.slsutil.merge
- an optional
strategy
key to configure the merging strategy, for examplestrategy: 'recurse'
, the default issmart
- an optional
merge_lists
key to configure if lists should be merged or overridden for therecurse
andoverwrite
strategy, for examplemerge_lists: 'true'
- an optional
Here is a valid example:
---
strategy: 'recurse'
merge_lists: 'false'
values:
pkg:
name: 'some-package'
config:
name: '/path/to/a/configuration/file'
...
You can provide a Jinja2 YAML template file with a name suffixed with .yaml.jinja
, it must produce a YAML file conform to the Format of configuration YAML files, for example:
---
strategy: 'overwrite'
merge_lists: 'true'
values:
{%- if grains["os"] == "Debian" %}
output_dir: /tmp/{{ grains["id"] }}
{%- endif %}
...
The map.jinja
file aggregates configuration values from several sources:
- YAML files stored in the fileserver
- pillars
- grains
- configuration gathered with salt['config.get']
For the values loaded from YAML files, map.jinja
will automatically try to load a Jinja2 template with the same name as the YAML file with the addition of the .jinja
extension, for example foo/bar/quux.yaml.jinja
.
After loading values from all sources, it will try to include the salt://parameters/post-map.jinja
Jinja file if it exists which can post-process the mapdata
variable.
The map.jinja
file uses several sources where to lookup parameter values. The list of sources can be configured in two places:
- globally
- with a plain YAML file
salt://parameters/map_jinja.yaml
- with a Jinja2 YAML template file
salt://parameters/map_jinja.yaml.jinja
- with a plain YAML file
- per formula
- with a plain YAML file
salt://{{ tplroot }}/parameters/map_jinja.yaml
- with a Jinja2 YAML template file
salt://{{ tplroot }}/parameters/map_jinja.yaml.jinja
- with a plain YAML file
Note
The map.jinja
configuration files must conform to the format of configuration YAML files.
Each source definition has the form [<TYPE>[:<OPTION>[:<DELIMITER>]]@]<KEY>
where <TYPE>
can be one of:
Y
to load values from YAML files from the fileserver, this is the default when no type is definedC
to lookup values with salt['config.get']G
to lookup values with salt['grains.get']I
to lookup values with salt['pillar.get']
The YAML type option can define the query method to lookup the key value to build the file name:
C
to query with salt['config.get'], this is the default when no query method is definedG
to query with salt['grains.get']I
to query with salt['pillar.get']
The C
, G
or I
types can define the SUB
option to store values in the sub key mapdata.<KEY>
instead of directly in mapdata
.
All types can define the <DELIMITER>
option to use an alternate delimiter of the <KEY>
, for example: on windows system you can't use colon :
for YAML file path name and you should use something else like exclamation mark !
.
Finally, the <KEY>
describes what to lookup to either build the YAML filename or gather values using one of the query methods.
Note
For the YAML type:
- if the
<KEY>
can't be looked up, then it's used a literal string path to a YAML file, for example:any/path/can/be/used/here.yaml
will result in the loading ofsalt://{{ tplroot }}/parameters/any/path/can/be/used/here.yaml
if it exists map.jinja
will automatically try to load a Jinja2 template, after the corresponding YAML file, with the same name as the YAML file extended with the.jinja
extension, for exampleany/path/can/be/used/here.yaml.jinja
The built-in map.jinja
sources are:
- "Y:G@osarch"
- "Y:G@os_family"
- "Y:G@os"
- "Y:G@osfinger"
- "C@{{ tplroot ~ ':lookup' }}"
- "C@{{ tplroot }}"
- "Y:G@id"
This is strictly equivalent to the following map_jinja.yaml.jinja
:
values:
sources:
- "parameters/osarch/{{ salt['grains.get']('osarch') }}.yaml"
- "parameters/osarch/{{ salt['grains.get']('osarch') }}.yaml.jinja"
- "parameters/os_family/{{ salt['grains.get']('os_family') }}.yaml"
- "parameters/os_family/{{ salt['grains.get']('os_family') }}.yaml.jinja"
- "parameters/os/{{ salt['grains.get']('os') }}.yaml"
- "parameters/os/{{ salt['grains.get']('os') }}.yaml.jinja"
- "parameters/osfinger/{{ salt['grains.get']('osfinger') }}.yaml"
- "parameters/osfinger/{{ salt['grains.get']('osfinger') }}.yaml.jinja"
- "C@{{ tplroot ~ ':lookup' }}"
- "C@{{ tplroot }}"
- "parameters/id/{{ salt['grains.get']('id') }}.yaml"
- "parameters/id/{{ salt['grains.get']('id') }}.yaml.jinja"
For each configuration source defined, map.jinja
will:
- load values depending on the source type:
- for YAML file sources
- if the
<KEY>
can be looked up:- load values from the YAML file named
salt://{{ tplroot }}/paramaters/<KEY>/{{ salt['<QUERY_METHOD>']('<KEY>') }}.yaml
if it exists - load values from the Jinja2 YAML template file named
salt://{{ tplroot }}/paramaters/<KEY>/{{ salt['<QUERY_METHOD>']('<KEY>') }}.yaml.jinja
if it exists
- load values from the YAML file named
- otherwise:
- load the YAML file named
salt://{{ tplroot }}/parameters/<KEY>.yaml
if it exists - load the Jinja2 YAML template file named
salt://{{ tplroot }}/parameters/<KEY>.yaml.jinja
if it exists
- load the YAML file named
- if the
- for
C
,G
orI
source type, lookup the value ofsalt['<QUERY_METHOD>']('<KEY>')
- for YAML file sources
- merge the loaded values with the previous ones using salt.slsutil.merge
There will be no error if a YAML or Jinja2 file does not exists, they are all optional.
For sources with of type C
declared in map_jinja:sources
, you can configure the merge
option of salt['config.get'] by defining per formula strategy
configuration key (retrieved with salt['config.get'](tplroot ~ ':strategy')
with one of the following values:
recurse
merge recursively dictionaries. Non dictionary values replace already defined valuesoverwrite
new value completely replace old ones
By default, no merging is done, the first value found is returned.
To summarise, here is a complete example of the load order of formula configuration values for an AMD64
Ubuntu 18.04
minion named minion1.example.net
for the libvirt
formula:
parameters/defaults.yaml
parameters/defaults.yaml.jinja
parameters/osarch/amd64.yaml
parameters/osarch/amd64.yaml.jinja
parameters/os_family/Debian.yaml
parameters/os_family/Debian.yaml.jinja
parameters/os/Ubuntu.yaml
parameters/os/Ubuntu.yaml.jinja
parameters/osfinger/Ubuntu-18.04.yaml
parameters/osfinger/Ubuntu-18.04.yaml.jinja
salt['config.get']('libvirt:lookup')
salt['config.get']('libvirt')
parameters/id/minion1.example.net.yaml
parameters/id/minion1.example.net.yaml.jinja
Remember that the order is important, for example, the value of key1:subkey1
loaded from parameters/os_family/Debian.yaml
is overridden by a value loaded from parameters/id/minion1.example.net.yaml
.
map.jinja
requires:
- salt minion 2018.3.3 minimum to use the traverse jinja filter
- to be located at the root of the formula named directory (e.g.
libvirt-formula/libvirt/map.jinja
) - the
libsaltcli.jinja
library, stored in the same directory, to disable themerge
option of salt['config.get'] over salt-ssh - the
libmapstack.jinja
library to load the configuration values - the
libmatchers.jinja
library used bylibmapstack.jinja
to parse compound like matchers
The map.jinja
exports a unique mapdata
variable which could be renamed during import.
Here is the best way to use it in an sls
file:
{#- Get the `tplroot` from `tpldir` #}
{%- set tplroot = tpldir.split("/")[0] %}
{%- from tplroot ~ "/map.jinja" import mapdata as rabbitmq with context %}
test-does-nothing-but-display-rabbitmq-as-json:
test.nop:
- name: {{ rabbitmq | json }}
When you need to process salt templates, you should avoid calling salt['config.get'] (or salt['pillar.get'] and salt['grains.get']) directly from the template. All the needed values should be available within the mapdata
variable exported by map.jinja
.
Here is an example based on rabbitmq-formula/rabbitmq/config/file.sls:
# -*- coding: utf-8 -*-
# vim: ft=sls
{#- Get the `tplroot` from `tpldir` #}
{%- set tplroot = tpldir.split('/')[0] %}
{%- set sls_package_install = tplroot ~ '.package.install' %}
{%- from tplroot ~ "/map.jinja" import mapdata as rabbitmq with context %}
{%- from tplroot ~ "/libtofs.jinja" import files_switch with context %}
include:
- {{ sls_package_install }}
rabbitmq-config-file-file-managed:
file.managed:
- name: {{ rabbitmq.config }}
- source: {{ files_switch(['example.tmpl'],
lookup='rabbitmq-config-file-file-managed'
)
}}
- mode: 644
- user: root
- group: {{ rabbitmq.rootgroup }}
- makedirs: True
- template: jinja
- require:
- sls: {{ sls_package_install }}
- context:
rabbitmq: {{ rabbitmq | json }}
This sls
file expose a rabbitmq
context variable to the jinja template which could be used like this:
########################################################################
# File managed by Salt at <{{ source }}>.
# Your changes will be overwritten.
########################################################################
This is another example file from SaltStack rabbitmq-formula.
# This is here for testing purposes
{{ rabbitmq | json }}
winner of the merge: {{ rabbitmq['winner'] }}