A Comprehensive Introduction to Salt

System Administration ?? Comments 01/out/2018 seg

Introduction

Recent advances in IT infrastructures, namely for Cloud-based Services, where scalability, elasticity and availability are key factors, have also seen advances in the way Management and Administration of these critical infrastructures are achieved, be it on log analysis and systems monitoring or on configuration management and automation. This is a rising subject, with exponential development evolution as new tools with different approaches have emerged in the last few years. [2]

Delaet et al. [3] provide a framework for evaluating different System Configuration tools. They also define a conceptual architecture for these tools, illustrated in Figure 1 . In essence, this type of tool is composed by a master element (typically residing in a deployment node) containing a set of specialized modules that translate through some form of manifest files, the specific configuration for each component of the remote managed nodes of the infrastructure. Each remote managed node, through some form of local agent conveys to the master detailed information about the node and executes configuration actions determined by the master. The master compiles in a repository a catalog (inventory) with information on the nodes and on how they should be configured.

Figure 1

Figure 1. A conceptual architecture of system configuration tools

In their paper, John Benson et al. [1], address the challenge of building an effective multi-cloud application deployment controller as a customer add-on outside of a cloud utility service using automation tools such as Ansible (ansible.com), SaltStack (saltstack.com) and Chef (www.chef.io) and compare them to each other as well as with the case where there is no automation framework just plain shell code.

The authors compare those three tools in terms of performance in executing three sets of tasks (installing a chemistry software package, installing an analytics software package, installing both software packages), the amount of code needed to execute those tasks and different features they have or do not have, such as a GUI and the need for an agent.

Ebert, C. et al. [4] address and discuss this recent organization culture coined DevOps, namely the toolset used at different phases, from Build, to Deployment, to Operations. The authors conclude that "DevOps is impacting the entire software and IT industry. Building on lean and agile practices, DevOps means end-to-end automation in software development and delivery."[4], with which I agree on. The authors also state that "Hardly anybody will be able to approach it with a cookbook-style approach, but most developers will benefit from better connecting development and operations."[4] with which I do not totally agree with, as I will try to demonstrate, by showing that a reactive infrastructure can be created to act autonomously on certain events, without leaving the cookbook-like approach, but through a set of triggered events that can be used to automate software development and delivery. There are already tools, such as SaltStack that makes possible for certain events to intelligently trigger actions, allowing therefore IT operators to create autonomic systems and data centers instead of just static runbooks. The authors also state that a "mutual understanding from requirements onward to maintenance, service, and product evolution will improve cycle time by 10 to 30 percent and reduce costs up 20 percent."[4] pointing out that the major drivers are fewer requirements changes, focused testing, quality assurance, and a much faster delivery cycle with feature-driven teams. They also point out the large dynamics of this DevOps culture as "each company needs its own approach to achieve DevOps, from architecture to tools to culture."[4]

Saltstack took a different approach for its architecture, illustrated in Figure 2, when compared to the architecture defined by Delaet et al. [3], illustrated in Figure 1. Comparing both, there are some similarities but it can also be seen that there is a bidirectional channel of communication between the managed devices and the Salt Master which corresponds to an Event Bus that connects to different parts of the tool. The Runner is used for executing modules specific to the Salt Master but can also provide information about the managed devices. The Reactor is responsible for receiving events and corresponding them to the appropriate state or even to fire another event as a consequence of the former. As stated in their documentation, "Salt Engines are long-running, external system processes that leverage Salt"(https://docs.saltstack.com), allowing to integrate external tools into the Salt system, for example to send to the salt Event Bus messages of a channel on the Slack (slack.com) communication tool (see https://github.com/saltstack/salt/blob/develop/salt/engines/slack.py).

Figure 2

Figure 2: Saltstack architecture

Architecture

For this work we will be using Saltstack's orchestration tool Salt. Its requirements for a device to be managed by it can be as little as having a REpresentational State Transfer (REST) API through which one can interact with. Its overall architecture can be seen in Figure 2. In this section we will see the architectural, network and hardware requirements, as well as Salt's software requirements and how to use Salt as a full infrastructure management tool. The minimum configuration for Salt is one machine, which takes the role of either Master and Minion, or just a Minion, with no master see Figure 3. Although this does not achieve much, it still allows to test operations, be it configuration management, reactions to certain events, monitoring of processes or to just do some tests with Salt.

Figure 3

Figure 3: Monolithic example

Consider the two scenarios represented in Figure 4 and Figure 5:
  1. an Operator within the Control layer
  2. an Operator outside the Control layer, with Control and Infrastructure layers behind a Network Address Translation (NAT)

In the first scenario, the Operator is either within the Main Controller itself or in the network neighbourhood, meaning that it has network access within the private network of which the Main Controller is part of.

In the second scenario, the Operator and the Main Controller are in distinct networks, meaning that the Operator with have to have external access to the infrastructure, either using the Main Controller has a bastion server, or by assigning the minions with a public Internet Protocol (IP) address. We will see that as public IP addresses are not that abundant, using a bastion becomes a necessity. The Salt administration tool, by Saltstack, has several key components which make it a complete framework for managing and administering an IT infrastructure. In each of the following subsections we will see what they are used for and how to make use of them.

"Running pre-defined or arbitrary commands on remote hosts, also known as remote execution, is the core function of Salt" [5]. Remote execution in Salt is achieved through execution modules and returners. Salt also contains a configuration management framework, which complements the remote execution functions by allowing more complex and interdependent operations. The framework functions are executed on the minion's side, allowing for scalable, simultaneous configuration of a great number of minion targets.

"The Salt Event System is used to fire off events enabling third party applications or external processes to react to behavior within Salt" [6]. These events can be launched from within the Salt infrastructure or from applications residing outside of it. To monitor non-Salt processes one can use the beacon system which "allows the minion to hook into a variety of system processes and continually monitor these processes. When monitored activity occurs in a system process, an event is sent on the Salt event bus that can be used to trigger a reactor" [7].

The pinnacle of the two previous components is using that information to trigger actions in response to those kinds of events. In Salt we have the reactor system "a simple interface to watching Salt's event bus for event tags that match a given pattern and then running one or more commands in response." [8].

Figure 4

Figure 4: Operator within the Control layer

Figure 5

Figure 5: Operator outside the Control layer

Most of the information may be static, but it can belong to each individual minion. This information can be either gathered from the minion itself or attributed to it by the Salt system. The first "is called the grains interface, because it presents salt with grains of information. Grains are collected for the operating system, domain name, IP address, kernel, OS type, memory, and many other system properties." [9]. The latter is called the pillar, which "is an interface for Salt designed to offer global values that can be distributed to minions" [10], in a private way relative to every other minion, if need be.

Directory Structures

There are two essential locations for salt related files (excluding service files):

  • /etc/salt
  • /srv/salt

In /etc/salt we have the configuration files for salt master and minion, as well as the keys for known minions (if it's a salt master machine) and the key for the salt master (if it's a salt minion machine).

root@master:~# tree /etc/salt
/etc/salt
├── master
├── minion
├── minion_id
└── pki
    ├── master
    │   ├── master.pem
    │   ├── master.pub
    │   ├── minions
    │   │   ├── minion1
    │   │   ├── minion2
    │   │   └── minion3
    │   ├── minions_autosign
    │   ├── minions_denied
    │   ├── minions_pre
    │   └── minions_rejected
    │
    │
    │
    └── minion
        ├── minion_master.pub
        ├── minion.pem
        └── minion.pub

/etc/salt/master: master config file

/etc/salt/minion: minion config file

/etc/salt/minion_id: minion identifier file

/etc/salt/pki/master/master.pem: master private key

/etc/salt/pki/master/master.pub: master public key

/etc/salt/pki/minion/minion_master.pub: master public key

/etc/salt/pki/minion/minion.pub: minion public key

/etc/salt/pki/minion/minion.pem: minion private key

In /srv/salt we store the state, pillar and reactor files, we will read more about this further ahead.

Configuration Files

We will now see basic, and some optional, settings for both salt master and salt minion.

/etc/salt/master

# The address of the interface to bind to, here we bind to all
interface: 0.0.0.0

# The port used by the publisher, default 4505
publish_port: 4505

# The user to run salt as
user: root

# Number of hours to keep old job information
keep_jobs: 168

# Timeout for the salt command and api, default is 5 seconds
timeout: 15

# SLS files location
file_roots:
  base:
    - /srv/salt/states/base/

# Top file to use
state_top: top.sls

# Pillar location
pillar_roots:
  base:
    - /srv/salt/pillar/base

# Reactor Configurations:
reactor:
  - 'salt/auth':
    - /srv/salt/reactor/auth-pending.sls

# if we just want to see what actually changes, set this to False
state_verbose: False

# log file location, we leave this to the system's logger
log_file: file:///dev/log

# The level of messages to send to the log file.
# One of 'info', 'quiet', 'critical', 'error', 'debug', 'warning'.
log_level: info
log_fmt_logfile: '%(asctime)-15s salt-master[%(process)d] %(name)s: %(message)s'
log_datefmt_logfile: '%b %d %H:%M:%S'

From reading this configuration file we can see that we will be listening on all interfaces using the default publish port, running as user root, keep job information for a week, increasing the default timeout for salt commands, and logging to the system's logger with a custom-defined format:

[TIMESTAMP] salt-master[PID] SALT_FUNCTION: MESSAGE

Besides this we also set up the base directories for state files, pillar data and reactor states. We will read about this further in this post, in Configuration Management, Storing and Accessing Data and Event System, respectively.

/etc/salt/minion

# salt master address
master: salt-master.local

# this overrides the salt auto naming (from hostname -f)
minion_id: minion1

# log file location, we leave this to the system's logger
log_file: file:///dev/log

# The level of messages to send to the log file.
# One of 'info', 'quiet', 'critical', 'error', 'debug', 'warning'.
log_level: info
log_fmt_logfile: '%(asctime)-15s salt-minion[%(process)d] %(name)s: %(message)s'
log_datefmt_logfile: '%b %d %H:%M:%S'

As we can see, the most basic configuration file for a salt minion is its salt master network location, it can be configured via a fqdn, a hostname defined in /etc/hosts/ file or via IP address.

We can also see the minion_id setting which forces a minion identifier, instead of letting the salt-minion choose one based on the machine's hostname.

Subsystem Files

Like we saw before, we configure all of the salt subsystem files within /srv/salt/, here we have a tree view of its subdirectories (not ordered on purpose):

root@master:~# tree /srv/salt
/srv/salt
└── states
│   └── base
│       ├── ntp
│       │   ├── etc
│       │   │   └── ntp.conf.tmpl
│       │   └── init.sls
│       └── top.sls
├── pillar
│   └── base
│       ├── ntp
│       │   └── init.sls
│       └── top.sls
└── reactor
    └── auth-pending.sls

The Top File

An infrastructure can be seen as being an applicational stack by itself, by having groups of different machines set up in small clusters, each cluster performing a sequence of tasks. The Top file is were the attribution of configuration role(s) and Minion(s) is made. This kind of file is used both for State and Pillar systems. Top files can be structurally seen has having three levels:

Environment:(ϵ) A directory structure containing a set of SaLt State (SLS) files.
Target:(θ) A predicate used to target Minions.
State files:(σ) A set of SLS files to apply to a target. Each file describes one or more states to be executed on matched Minions.

These three levels relate in such a way that (θ ∈ ϵ) , and (σ ∈ θ). Reordering these two relations we have that:

  • Environments contain targets
  • Targets contain states

Putting these concepts together, we can describe a scenario (top file) in which the state defined in the SLS file ntp/init.sls is enforced to all minions.

top.sls

base:
  '*':
    - ntp

Remote Execution

Before delving into the more intricate orchestration-like functionalities of salt, let us use the most (probably) basic operation of salt, which you probably will use a lot.

The function is cmd.run, made available from the cmd execution module (see [11] for a list of all available execution modules). And we can use it like this, from a salt master environment:

root@master:~# salt \* cmd.run 'hostname && pwd && ls -la'

This will run the command, or sequence of commands, in every targeted minion and return its result. You can also specify the shell you want to use with the shell argument.

root@master:~# salt \* cmd.run 'hostname && ls -la' shell=/bin/bash
  minion3:
      minion3.local
      /root
      total 40
      drwx------  7 root root 4096 Sep 27 16:35 .
      drwxr-xr-x 23 root root 4096 Oct  2 07:00 ..
      -rw-------  1 root root 2336 Sep 29 14:11 .bash_history
      -rw-r--r--  1 root root 3201 Sep  3 14:19 .bashrc
      drwx------  3 root root 4096 Sep 19 15:21 .cache
      drwx------  3 root root 4096 Sep 27 16:35 .config
      drwxr-xr-x  2 root root 4096 Aug 17 17:00 .nano
      -rw-r--r--  1 root root  148 Aug 17  2015 .profile
      drwx------  2 root root 4096 Sep 19 15:21 .ssh
      drwxr-xr-x  3 root root 4096 Sep  6 16:43 .virtualenvs
  minion1:
      master.local
      /root
      total 40
      drwx------  7 root root 4096 Sep 27 16:35 .
      drwxr-xr-x 23 root root 4096 Oct  2 07:00 ..
      -rw-------  1 root root 2336 Sep 29 14:11 .bash_history
      -rw-r--r--  1 root root 3201 Sep  3 14:19 .bashrc
      drwx------  3 root root 4096 Sep 19 15:21 .cache
      drwx------  3 root root 4096 Sep 27 16:35 .config
      drwxr-xr-x  2 root root 4096 Aug 17 17:00 .nano
      -rw-r--r--  1 root root  148 Aug 17  2015 .profile
      drwx------  2 root root 4096 Sep 19 15:21 .ssh
      drwxr-xr-x  3 root root 4096 Sep  6 16:43 .virtualenvs
  minion2:
      minion2.local
      /root
      total 40
      drwx------  7 root root 4096 Sep 27 16:35 .
      drwxr-xr-x 23 root root 4096 Oct  2 07:00 ..
      -rw-------  1 root root 2336 Sep 29 14:11 .bash_history
      -rw-r--r--  1 root root 3201 Sep  3 14:19 .bashrc
      drwx------  3 root root 4096 Sep 19 15:21 .cache
      drwx------  3 root root 4096 Sep 27 16:35 .config
      drwxr-xr-x  2 root root 4096 Aug 17 17:00 .nano
      -rw-r--r--  1 root root  148 Aug 17  2015 .profile
      drwx------  2 root root 4096 Sep 19 15:21 .ssh
      drwxr-xr-x  3 root root 4096 Sep  6 16:43 .virtualenvs
root@master:~#

Another well used execution module is pkg, it will try to identify the package system present in the minion's system and use it to perform the requested operations.

root@master:~# salt minion\* pkg.latest_version postfix
  minion1:
      3.1.0-3ubuntu0.3
  minion2:
      3.1.0-3ubuntu0.3
  minion3:
      3.1.0-3ubuntu0.3
root@master:~# salt minion\* pkg.install postfix
minion1:
    ----------
  postfix:
      ----------
      new:
          3.1.0-3ubuntu0.3
      old:
  (... suppressed output ...)
minion3:
    ----------
  postfix:
      ----------
      new:
          3.1.0-3ubuntu0.3
      old:
  (... suppressed output ...)
minion2:
    ----------
  postfix:
      ----------
      new:
          3.1.0-3ubuntu0.3
      old:
  (... suppressed output ...)

Configuration Management

Remote execution is very useful, however we may want more complex scenarios to occur, then we can make use of SLS files, states, and state modules [12].

From the tree view seen in Subsystem Files and the configuration of the file_roots option, all of our available states are to be placed in /srv/salt/states/base.

SLS files are mainly written in YAML Ain't Markup Language (YAML) [13] and like most salt related files, can be templated using Jinja [14], by default.

ntp/init.sls

ntp:
  pkg.installed
  service.running:
    - enable: True
    - watch:
      - file: /etc/ntp.conf
  /etc/ntp.conf:
  file.managed:
    - source: salt://ntp/etc/ntp.conf.tmpl
    - mode: 644
    - user: root
    - require:
      - pkg: ntp
ntp-shutdown:
  service.dead:
    - name: ntp
    - onchanges:
      - file: /etc/ntp.conf

ntp/etc/ntp.conf.tmpl

# This file managed by Salt, do not edit
server ntp1.example.com
server ntp1.example.com

# Only allow read-only access from localhost
restrict default noquery nopeer
restrict 127.0.0.1
restrict ::1

# Location of drift file
driftfile /var/lib/ntp/ntp.drift

This salt state file, when applied, will:

  1. install the ntp package, using the system's package manager
  2. create/change the file /etc/ntp/ntp.conf, on the minion, according to the contents of the /srv/salt/states/base/ntp/etc/ntp.conf.tmpl, on master
  3. make sure the ntp service is stopped if there are changes to the previously managed file.
  4. make sure the ntp service is running and enabled, restarting it if there were changes to the managed file (it should not restart because of the the ntp-shutdown state.

To apply a state we use the state execution module, and its function apply. One key difference between using execution modules and state modules is the possibility of using the keyword test that, when True, will not effectively apply the state but rather it will show what it would do and, in our case, which changes it would apply.

root@master:~# salt --state-verbose=True minion2 state.apply ntp test=True
minion2:
----------
          ID: ntp-pkg
    Function: pkg.installed
        Name: ntp
      Result: True
     Comment: All specified packages are already installed
     Started: 15:26:10.712651
    Duration: 928.968 ms
     Changes:
----------
          ID: ntp-pkg
    Function: file.managed
        Name: /etc/ntp.conf
      Result: None
     Comment: The file /etc/ntp.conf is set to be changed
     Started: 15:26:11.664815
    Duration: 63.205 ms
     Changes:
              ----------
              diff:
                  ---
                  +++
                  @@ -1,53 +1,20 @@
                  -# /etc/ntp.conf, configuration for ntpd; see ntp.conf(5) for help
                  +# This file managed by Salt, do not edit
                  +#
                  +#
                  +# With the default settings below, ntpd will only synchronize your clock.

                  -driftfile /var/lib/ntp/ntp.drift
                  -
                  -
                  -# Enable this if you want statistics to be logged.
                  -#statsdir /var/log/ntpstats/
                  -
                  -statistics loopstats peerstats clockstats
                  -filegen loopstats file loopstats type day enable
                  -filegen peerstats file peerstats type day enable
                  -filegen clockstats file clockstats type day enable
                  -
                  -# Specify one or more NTP servers.

                   # Use servers from the NTP Pool Project. Approved by Ubuntu Technical Board
                   # on 2011-02-08 (LP: #104525). See http://www.pool.ntp.org/join.html for
                   # more information.
                  -server ntp1.somewhere.in.time
                  -server ntp2.somewhere.in.time
                  +server ntp1.example.com
                  +server ntp2.example.com

                  -# Use Ubuntu's ntp server as a fallback.
                  -server ntp.ubuntu.com

                  -# Access control configuration; see /usr/share/doc/ntp-doc/html/accopt.html for
                  -# details.  The web page <http://support.ntp.org/bin/view/Support/AccessRestrictions>
                  -# might also be helpful.
                  -#
                  -# Note that "restrict" applies to both servers and clients, so a configuration
                  -# that might be intended to block requests from certain clients could also end
                  -# up blocking replies from your own upstream servers.
                  -
                  -# By default, exchange time with everybody, but don't allow configuration.
                  -restrict -4 default kod notrap nomodify nopeer noquery
                  -restrict -6 default kod notrap nomodify nopeer noquery
                  -
                  -# Local users may interrogate the ntp server more closely.
                  +# Only allow read-only access from localhost
                  +restrict default noquery nopeer
                   restrict 127.0.0.1
                   restrict ::1

                  -# Clients from this (example!) subnet have unlimited access, but only if
                  -# cryptographically authenticated.
                  -#restrict 192.168.123.0 mask 255.255.255.0 notrust
                  -
                  -
                  -# If you want to provide time to your local subnet, change the next line.
                  -# (Again, the address is an example only.)
                  -#broadcast 192.168.123.255
                  -
                  -# If you want to listen to time broadcasts on your local subnet, de-comment the
                  -# next lines.  Please do this only if you trust everybody on the network!
                  -#disable auth
                  -#broadcastclient
                  +# Location of drift file
                  +driftfile /var/lib/ntp/ntp.drift
----------
          ID: ntp-pkg
    Function: service.running
        Name: ntp
      Result: None
     Comment: Service ntp is set to start
     Started: 15:26:11.729479
    Duration: 134.939 ms
     Changes:
----------
          ID: ntpdate
    Function: pkg.installed
      Result: True
     Comment: All specified packages are already installed
     Started: 15:26:11.864881
    Duration: 25.952 ms
     Changes:

Summary for minion2
------------
Succeeded: 4 (unchanged=2, changed=1)
Failed:    0
------------
Total states run:     4
Total run time:   1.153 s

Now for the actual application of the state

----------
          ID: ntp-pkg
    Function: pkg.installed
        Name: ntp
      Result: True
     Comment: All specified packages are already installed
     Started: 15:30:11.488468
    Duration: 902.224 ms
     Changes:
----------
          ID: ntp-pkg
    Function: file.managed
        Name: /etc/ntp.conf
      Result: True
     Comment: File /etc/ntp.conf updated
     Started: 15:30:12.394500
    Duration: 94.999 ms
     Changes:
              ----------
              diff:
                  ---
                  +++
                  @@ -1,53 +1,20 @@
                  -# /etc/ntp.conf, configuration for ntpd; see ntp.conf(5) for help
                  +# This file managed by Salt, do not edit
                  +#
                  +#
                  +# With the default settings below, ntpd will only synchronize your clock.

                  -driftfile /var/lib/ntp/ntp.drift
                  -
                  -
                  -# Enable this if you want statistics to be logged.
                  -#statsdir /var/log/ntpstats/
                  -
                  -statistics loopstats peerstats clockstats
                  -filegen loopstats file loopstats type day enable
                  -filegen peerstats file peerstats type day enable
                  -filegen clockstats file clockstats type day enable
                  -
                  -# Specify one or more NTP servers.

                   # Use servers from the NTP Pool Project. Approved by Ubuntu Technical Board
                   # on 2011-02-08 (LP: #104525). See http://www.pool.ntp.org/join.html for
                   # more information.
                  -server ntp1.somewhere.in.time
                  -server ntp2.somewhere.in.time
                  +server ntp1.example.com
                  +server ntp2.example.com

                  -# Use Ubuntu's ntp server as a fallback.
                  -server ntp.ubuntu.com

                  -# Access control configuration; see /usr/share/doc/ntp-doc/html/accopt.html for
                  -# details.  The web page <http://support.ntp.org/bin/view/Support/AccessRestrictions>
                  -# might also be helpful.
                  -#
                  -# Note that "restrict" applies to both servers and clients, so a configuration
                  -# that might be intended to block requests from certain clients could also end
                  -# up blocking replies from your own upstream servers.
                  -
                  -# By default, exchange time with everybody, but don't allow configuration.
                  -restrict -4 default kod notrap nomodify nopeer noquery
                  -restrict -6 default kod notrap nomodify nopeer noquery
                  -
                  -# Local users may interrogate the ntp server more closely.
                  +# Only allow read-only access from localhost
                  +restrict default noquery nopeer
                   restrict 127.0.0.1
                   restrict ::1

                  -# Clients from this (example!) subnet have unlimited access, but only if
                  -# cryptographically authenticated.
                  -#restrict 192.168.123.0 mask 255.255.255.0 notrust
                  -
                  -
                  -# If you want to provide time to your local subnet, change the next line.
                  -# (Again, the address is an example only.)
                  -#broadcast 192.168.123.255
                  -
                  -# If you want to listen to time broadcasts on your local subnet, de-comment the
                  -# next lines.  Please do this only if you trust everybody on the network!
                  -#disable auth
                  -#broadcastclient
                  +# Location of drift file
                  +driftfile /var/lib/ntp/ntp.drift
----------
          ID: ntp-pkg
    Function: service.running
        Name: ntp
      Result: True
     Started: 15:30:12.587597
    Duration: 49.742 ms
     Changes:
----------
          ID: ntpdate
    Function: pkg.installed
      Result: True
     Comment: All specified packages are already installed
     Started: 15:30:12.637827
    Duration: 29.567 ms
     Changes:

Summary for minion2
------------
Succeeded: 4 (changed=1)
Failed:    0
------------
Total states run:     4
Total run time:   1.077 s

Now if we apply the ntp state again we will see no changes, and no service stop/start.

SLS files can include other SLS files, use externally generated data and depend on states declared on included SLS files.

Each state can require other state completion, be required in other states, watch a certain state, execute unless a given command returns True, execute only if a certain command returns True and more. [2] [15]

Storing and Accessing Data

When planning and implementing this type of orchestration, it is useful to have information that can be shared between states, and which is known only to the intended recipients (i.e: specific minions).

In salt we have the Pillar, Its configuration is similar to the one used by States, as we saw in Subsystem Files its files are located in /srv/salt/pillar/base.

The way the information is defined and assigned is similar to states, but instead of defining State Module executions we define information in the form of YAML, as shown

/srv/salt/pillar/base/ntp/init.sls

ntp:
  servers:
    - ntp1.somewhere.in.time
    - ntp2.somewhere.in.time
  ntp_conf: /etc/ntp.conf
  drift_file: /var/lib/ntp/ntp.drift

and we define the attributions of data in the corresponding pillar top file, where all minions are assigned information regarding Network Time Protocol (NTP) server domain names and their NTP service configuration file.

/srv/salt/pillar/base/top.sls

base:
  '*':
    - ntp

This information can be used within states or even template files, using the pillar keyword.

Taking the previous SLS file and its managed template, we can now dynamically fill NTP related info.

ntp/init.sls

ntp:
  pkg.installed
  service.running:
    - enable: True
    - watch:
      - file: {{ pillar['ntp']['ntp_conf'] }}
  {{ pillar['ntp']['ntp_conf'] }}:
  file.managed:
    - source: salt://ntp/etc/ntp.conf.tmpl
    - mode: 644
    - user: root
    - require:
      - pkg: ntp
ntp-shutdown:
  service.dead:
    - name: ntp
    - onchanges:
      - file: {{ pillar['ntp']['ntp_conf'] }}

ntp/etc/ntp.conf.tmpl

# This file managed by Salt, do not edit
{% for server in pillar['ntp']['servers'] %}
server {{ server }}
{% endfor %}

# Only allow read-only access from localhost
restrict default noquery nopeer
restrict 127.0.0.1
restrict ::1

# Location of drift file
driftfile {{ pillar['ntp']['drift_file'] }}

This example could be further enhanced, let us try and make this able to configure ntp servers on a windows machine. For that we need to use another data source in the salt system, the grains, which is generated by the minion itself.

root@master:~# salt win\* grains.get os
windows1:
    Windows
root@master:~#

Interesting, so now we can have selective behaviour in our state:

ntp/etc/ntp.conf.tmpl

ntp:
{% if grains['os'] == 'Windows' %}
  ntp.managed:
    - servers: {{ pillar['ntp']['servers'] }}
{% else %}
  pkg.installed
  service.running:
    - enable: True
    - watch:
      - file: {{ pillar['ntp']['ntp_conf'] }}
  {{ pillar['ntp']['ntp_conf'] }}:
  file.managed:
    - source: salt://ntp/etc/ntp.conf.tmpl
    - mode: 644
    - user: root
    - require:
      - pkg: ntp
ntp-shutdown:
  service.dead:
    - name: ntp
    - onchanges:
      - file: {{ pillar['ntp']['ntp_conf'] }}
{% endif %}

And now if we apply the state to the windows machine:

root@master:~# salt win\* state.apply ntp test

Event System

In Introduction we saw in Figure 2 a general view of the Salt System, which has a Reactor element, this element is part of a more broad system, the Event System (represented in Figure 6) which is "used to fire off events enabling third party applications or external processes to react to behavior within Salt" [6]. The Event System has two main components:

Event Socket:from where events are published
Event Library:used for listening to events and forward them to the salt system
Figure 6

Figure 6: Saltstack Event System Flow Example

"The event system is a local ZeroMQ PUB interface which fires salt events. This event bus is an open system used for sending information notifying Salt and other systems about operations" [8]. This associates SLS files to event tags on the master, which the SLS files in turn represent reactions.

This means that the reactor system has two steps to work properly. First, the reactor key needs to be configured in the master configuration file, this associates event tags SLS reaction files. Second, these reaction files use a YAML data structure similar to the state system to define which reactions are to be executed.

/etc/salt/master

(... suppressed ...)

reactor:
- 'salt/auth':
  - /srv/salt/reactor/auth-pending.sls

(... suppressed ...)

this entry means that for every 'salt/auth' event, the reactor state /srv/salt/reactor/auth-pending.sls should be executed. Let us see what that reactor state actually does. [8]

auth-pending.sls

{# minion failed to authenticate -- remove accepted key #}
{% if not data['result'] and data['id'].endswith('mydomain.com') %}
minion_remove:
  wheel.key.delete:
    - match: {{ data['id'] }}

minion_rejoin:
  cmd.cmd.run:
    - tgt: master.local
    - arg:
      - ssh -o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no "{{ data['id'] }}" 'sleep 10 && systemctl restart salt-minion'
{% endif %}

{# minion is sending new key -- accept this key #}
{% if 'act' in data and data['act'] == 'pend' and data['id'].endswith('mydomain.com') %}
minion_add:
  wheel.key.accept:
    - match: {{ data['id'] }}
{% endif %}

"In this SLS file, we say that if the key was rejected we will delete the key on the master and then also tell the master to ssh in to the minion and tell it to restart the minion, since a minion process will die if the key is rejected. We also say that if the key is pending and the id starts with ink we will accept the key. A minion that is waiting on a pending key will retry authentication every ten seconds by default." [8]

Conclusion

In this post we saw from where some of the system configuration tools methods come from, and the similarities between then and now. We briefly explored the content management and orchestration tool Salt, saw how to execute commands remotely via the cmd.run execution module function, how to orchestrate more complex scenarios with the state.apply state module function, how to generate, share and fetch static and dynamic information via the grains and the pillar subsystems, and we also saw how to catch certain events and react to them via the Event System. Hope this was useful, and that now you have some grasp of what is salt and how it can be powerful to plan and manage an IT infrastructure. Feel free to use the comment box below to ask any questions.

References

[1]
    1. Benson, J. J. Prevost, and P. Rad, Survey of automated software deployment for computational and engineering research, *10th Annual International Systems Conference, SysCon 2016 - Proceedings, 2016.
[2](1, 2)
  1. Torrinha, Adaptive Management and Administration of IT Infrastructures, MSc thesis, Instituto Superior Técnico, Lisbon, Portugal, 2017.
[3](1, 2)
  1. Delaet, W. Joosen, and B. Vanbrabant, A survey of system configuration tools, International Conference on Large Installation System Adminstration, pp. 1 - 8, 2010.
[4](1, 2, 3, 4, 5)
  1. Ebert, G. Gallardo, J. Hernantes, and N. Serrano, DevOps, vol. 33, no. 3, pp. 94 - 100, 2016.
[5]--, Remote Execution - Saltstack Documentation - v2018.3.2, 2018. [link]
[6](1, 2) --, Event System - Saltstack Documentation - v2018.3.2, 2018. [link]
[7]--, Beacons - Saltstack Documentation - v2018.3.2, 2018. [link]
[8](1, 2, 3, 4) --, Reactor System - Saltstack Documentation - v2018.3.2, 2018. [link]
[9]--, Grains - Saltstack Documentation - v2018.3.2, 2018. [link]
[10]--, Pillar - Saltstack Documentation - v2018.3.2, 2018. [link]
[11]--, List of execution modules - Saltstack Documentation - v2018.3.2, 2018 [link]
[12]--, List of state modules - Saltstack Documentation - v2018.3.2, 2018 [link]
[13]--, YAML website [link]
[14]--, Jinja2 website [link]
[15]--, Requisites and Other Global State Arguments - Saltstack Documentation - v2018.3.2, 2018 [link]

Manuel Torrinha is an information systems engineer, with more than 10 years of experience in managing GNU/Linux environments. Has an MSc in Information Systems and Computer Engineering. Work interests include High Performance Computing, Data Analysis, and IT management and Administration. Knows diverse programming, scripting and markup languages. Speaks Portuguese and English.

Related content