Using Salt Cloud with the Event Reactor

One of the most powerful features of the Salt framework is the Event Reactor. As the Reactor was in development, Salt Cloud was regularly updated to take advantage of the Reactor upon completion. As such, various aspects of both the creation and destruction of instances with Salt Cloud fire events to the Salt Master, which can be used by the Event Reactor.

Event Structure

As of this writing, all events in Salt Cloud have a tag, which includes the ID of the instance being managed, and a payload which describes the task that is currently being handled. A Salt Cloud tag looks like:

salt/cloud/<minion_id>/<task>

For instance, the first event fired when creating an instance named web1 would look like:

salt/cloud/web1/creating

Assuming this instance is using the ec2-centos profile, which is in turn using the ec2-config provider, the payload for this tag would look like:

{'name': 'web1',
 'profile': 'ec2-centos',
 'provider': 'ec2-config:ec2'}

Available Events

When an instance is created in Salt Cloud, whether by map, profile, or directly through an API, a minimum of five events are normally fired. More may be available, depending upon the cloud provider being used. Some of the common events are described below.

salt/cloud/<minion_id>/creating

This event states simply that the process to create an instance has begun. At this point in time, no actual work has begun. The payload for this event includes:

name profile provider

salt/cloud/<minion_id>/requesting

Salt Cloud is about to make a request to the cloud provider to create an instance. At this point, all of the variables required to make the request have been gathered, and the payload of the event will reflect those variables which do not normally pose a security risk. What is returned here is dependent upon the cloud provider. Some common variables are:

name image size location

salt/cloud/<minion_id>/querying

The instance has been successfully requested, but the necessary information to log into the instance (such as IP address) is not yet available. This event marks the beginning of the process to wait for this information.

The payload for this event normally only includes the instance_id.

salt/cloud/<minion_id>/waiting_for_ssh

The information required to log into the instance has been retrieved, but the instance is not necessarily ready to be accessed. Following this event, Salt Cloud will wait for the IP address to respond to a ping, then wait for the specified port (usually 22) to respond to a connection, and on Linux systems, for SSH to become available. Salt Cloud will attempt to issue the date command on the remote system, as a means to check for availability. If no ssh_username has been specified, a list of usernames (starting with root) will be attempted. If one or more usernames was configured for ssh_username, they will be added to the beginning of the list, in order.

The payload for this event normally only includes the ip_address.

salt/cloud/<minion_id>/deploying

The necessary port has been detected as available, and now Salt Cloud can log into the instance, upload any files used for deployment, and run the deploy script. Once the script has completed, Salt Cloud will log back into the instance and remove any remaining files.

A number of variables are used to deploy instances, and the majority of these will be available in the payload. Any keys, passwords or other sensitive data will be scraped from the payload. Most of the variables returned will be related to the profile or provider config, and any default values that could have been changed in the profile or provider, but weren't.

salt/cloud/<minion_id>/created

The deploy sequence has completed, and the instance is now available, Salted, and ready for use. This event is the final task for Salt Cloud, before returning instance information to the user and exiting.

The payload for this event contains little more than the initial creating event. This event is required in all cloud providers.

Configuring the Event Reactor

The Event Reactor is built into the Salt Master process, and as such is configured via the master configuration file. Normally this will be a YAML file located at /etc/salt/master. Additionally, master configuration items can be stored, in YAML format, inside the /etc/salt/master.d/ directory.

These configuration items may be stored in either location; however, they may only be stored in one location. For organizational and security purposes, it may be best to create a single configuration file, which contains only Event Reactor configuration, at /etc/salt/master.d/reactor.

The Event Reactor uses a top-level configuration item called reactor. This block contains a list of tags to be watched for, each of which also includes a list of sls files. For instance:

reactor:
  - 'salt/minion/*/start':
    - '/srv/reactor/custom-reactor.sls'
  - 'salt/cloud/*/created':
    - '/srv/reactor/cloud-alert.sls'
  - 'salt/cloud/*/destroyed':
    - '/srv/reactor/cloud-destroy-alert.sls'

The above configuration configures reactors for three different tags: one which is fired when a minion process has started and is available to receive commands, one which is fired when a cloud instance has been created, and one which is fired when a cloud instance is destroyed.

Note that each tag contains a wildcard (*) in it. For each of these tags, this will normally refer to a minion_id. This is not required of event tags, but is very common.

Reactor SLS Files

Reactor sls files should be placed in the /srv/reactor/ directory for consistency between environments, but this is not currently enforced by Salt.

Reactor sls files follow a similar format to other sls files in Salt. By default they are written in YAML and can be templated using Jinja, but since they are processed through Salt's rendering system, any available renderer (JSON, Mako, Cheetah, etc.) can be used.

As with other sls files, each stanza will start with a declaration ID, followed by the function to run, and then any arguments for that function. For example:

# /srv/reactor/cloud-alert.sls
new_instance_alert:
  cmd.pagerduty.create_event:
    - tgt: alertserver
    - kwarg:
        description: "New instance: {{ data['name'] }}"
        details: "New cloud instance created on {{ data['provider'] }}"
        service_key: 1626dead5ecafe46231e968eb1be29c4
        profile: my-pagerduty-account

When the Event Reactor receives an event notifying it that a new instance has been created, this sls will create a new incident in PagerDuty, using the configured PagerDuty account.

The declaration ID in this example is new_instance_alert. The function called is cmd.pagerduty.create_event. The cmd portion of this function specifies that an execution module and function will be called, in this case, the pagerduty.create_event function.

Because an execution module is specified, a target (tgt) must be specified on which to call the function. In this case, a minion called alertserver has been used. Any arguments passed through to the function are declared in the kwarg block.

Example: Reactor-Based Highstate

When Salt Cloud creates an instance, by default it will install the Salt Minion onto the instance, along with any specified minion configuration, and automatically accept that minion's keys on the master. One of the configuration options that can be specified is startup_states, which is commonly set to highstate. This will tell the minion to immediately apply a highstate, as soon as it is able to do so.

This can present a problem with some system images on some cloud hosts. For instance, Salt Cloud can be configured to log in as either the root user, or a user with sudo access. While some hosts commonly use images that lock out remote root access and require a user with sudo privileges to log in (notably EC2, with their ec2-user login), most cloud hosts fall back to root as the default login on all images, including for operating systems (such as Ubuntu) which normally disallow remote root login.

For users of these operating systems, it is understandable that a highstate would include configuration to block remote root logins again. However, Salt Cloud may not have finished cleaning up its deployment files by the time the minion process has started, and kicked off a highstate run. Users have reported errors from Salt Cloud getting locked out while trying to clean up after itself.

The goal of a startup state may be achieved using the Event Reactor. Because a minion fires an event when it is able to receive commands, this event can effectively be used inside the reactor system instead. The following will point the reactor system to the right sls file:

reactor:
  - 'salt/cloud/*/created':
    - '/srv/reactor/startup_highstate.sls'

And the following sls file will start a highstate run on the target minion:

# /srv/reactor/startup_highstate.sls
reactor_highstate:
  cmd.state.apply:
    - tgt: {{ data['name'] }}

Because this event will not be fired until Salt Cloud has cleaned up after itself, the highstate run will not step on salt-cloud's toes. And because every file on the minion is configurable, including /etc/salt/minion, the startup_states can still be configured for future minion restarts, if desired.