Agents

Agents are daemons that listens to Redis queues and execute tasks inside the system.

Agents

Agent types

The core cluster and node agent, and every module instance run their own agent process. The process is started by the agent.service Systemd user unit and the agent@.service Systemd system template unit provided by the core.

Systemd units for agents are:

agent@cluster.service running as root. Its actions are defined in /var/lib/nethserver/cluster/actions
agent@node.service running as root. Its actions are defined in /var/lib/nethserver/node/actions
agent.service running as non-privileged Unix user for each rootless module instance. See the “Additional modules” section below for more details

Tasks processing

Agents wait for tasks on a Redis list. Only the cluster agent and the api-server are allowed to push a new task for an arbitrary module agent. The task contain two main attributes: the action name (action), and the action input (data).

The following Redis command run the list-actions action on the cluster agent:

LPUSH cluster/tasks '{"id":"1a4a3965-d8d2-4c22-99d5-e17e6a5db36b","action":"list-actions","data":{}}'

Read the action output:

GET cluster/task/1a4a3965-d8d2-4c22-99d5-e17e6a5db36b/output

The action output is a string in JSON format.

[
  "list-actions",
  "destroy-module",
  "add-module",
  "add-user",
  "remove-module",
  "remove-repository",
  "update-routes",
  "alter-user",
  "grant-actions",
  "remove-user",
  "add-repository",
  "get-module-info",
  "join-cluster",
  "list-installed-modules",
  "list-updates",
  "revoke-actions",
  "create-module",
  "get-status",
  "add-node",
  "alter-repository",
  "create-cluster",
  "list-modules",
  "list-repositories"
]

Read the collected stderr data from the action steps:

GET cluster/task/1a4a3965-d8d2-4c22-99d5-e17e6a5db36b/error

Read the originally submitted task payload:

GET cluster/task/1a4a3965-d8d2-4c22-99d5-e17e6a5db36b/context

Read the action exit code:

GET cluster/task/1a4a3965-d8d2-4c22-99d5-e17e6a5db36b/exit_code

The above keys are transient. After a few hours they are evicted. This command return the remaining key TTL (Time To Live), in seconds.

TTL cluster/task/1a4a3965-d8d2-4c22-99d5-e17e6a5db36b/error

While the action is running, some messages are sent through the progress channel. It is possible to get them by subscribing the channel.

SUBSCRIBE progress/cluster/task/1a4a3965-d8d2-4c22-99d5-e17e6a5db36b

Tasks submission

The Python package agent.tasks implements the task processing, handling communication errors and re-connection/retrying logic at every stage, like:

Task submission
Task progress tracking
Task response parsing
Task cancellation when the TERM signal is received

Example:

import agent.tasks
print(agent.tasks.run(agent_id='cluster', action='list-actions', endpoint='redis://cluster-leader'))

Note the endpoint= argument: it uses the redis:// protocol, that requires cluster privileges to work.

As said above, only the cluster agent and api-server are allowed to LPUSH new tasks in Redis. Module agents must submit tasks for other agents as HTTP requests, through the api-server.

Basically this can be done in two steps:

obtain a JWT token with the module Redis credentials
send the task request

The Python module agent.tasks implements the two steps internally.

It has a few assumptions, that are always satisfied by an action step environment:

It obtain the module credentials from the environment from REDIS_USER and REDIS_PASSWORD
It stores the JWT token in the current working directory (the file is ./apitoken.cache)

This is an example call, note that the agent_id is obtained from the environment.

import agent.tasks, os
print(agent.tasks.run(agent_id=os.environ['AGENT_ID'], action='list-actions'))

The above call fails, because agents are not generally authorized to run list-actions. However, just for this example, it is possible to override the Redis credentials received from the environment.

import agent.tasks, os
os.environ['REDIS_USER'] = 'admin'
os.environ['REDIS_PASSWORD'] = 'Nethesis,1234'
print(agent.tasks.run(agent_id=os.environ['AGENT_ID'], action='list-actions'))

The next section explains how to authorize agents to run actions through roles assignment.

Roles and authorizations

In normal conditions module actions must be assigned to roles, and roles must be granted to agents and users.

In Redis, a SET defines the actions assigned to a role. If default assignments are not enough the set can be modified by the module during the create-module action. A set element represent an exact action name or an action name pattern, expressed in glob-pattern syntax (e.g. get-*).

The following Bash command in the create-module action allows agents with the actionsreader role to run list-actions on AGENT_ID (e.g. module/mymod1).

    redis-exec SADD "${AGENT_ID}/roles/actionsreader" "list-actions"

Same as above, but to allow also any other action with name prefix list-:

    redis-exec SADD "${AGENT_ID}/roles/actionsreader" "list-*"

Now it is possible to grant the role actionsreader on module/mymod1 to another agent (e.g. module/authmod2). This is the corresponding Redis command:

HSET roles/module/authmod2 module/mymod1 actionsreader

Roles are always granted by the cluster agent.

Users with the owner role can use the following actions to manage users and their roles:

add-user
remove-user
grant-actions
revoke-actions

Those actions are implemented with the cluster.grants Python module.

A module can require additional roles to other modules when it is instantiated. Set the label org.nethserver.authorizations of the module image to a space separated list of values. Each value describes what is the required role and what are the modules. The value syntax is {agent selector}:{role}. For instance

org.nethserver.authorizations = mymod@node:actionsreader funmod@cluster:rolex coolmod@any:roley

If the installed module is authmod2, the above label is interpreted in the following way:

grant authmod2 role actionsreader on the default instance of module mymod running on the same node (@node suffix). The first instance of mymod installed on the node of authmod2 is usually the default one.
grant authmod2 role rolex on the default instance of funmod at cluster level (@cluster suffix). That means the first instance of funmod installed in the cluster on any node.
grant authmod2 role roley on any existing instance of module coolmod.

Other possible agent selector values:

cluster - selects the cluster agent
node - selects the node agent where the module is running

Depending on the running node ID the value resolves to an agent ID. The above rules are implemented in Python by the agent.resolve_agent_id() function

import agent, os
print(agent.resolve_agent_id('traefik@node', node_id=os.getenv('NODE_ID', 1)))

The node_id argument is optional.