Developing Agents

What is an agent?

Agents are active software components which run as independent processes to fulfil one (or less often several) system functions. They are the engine that make DIRAC beat. Agents are processes that perform actions periodically. Each cycle agents typically contact a service, or look into a DB to check for pending actions, execute the required ones and report back the results. All agents are built in the same framework which organizes the main execution loop and provides a uniform way for deployment, configuration, control and logging of the agent activity.

Agents run in different environments. Those belonging to a DIRAC system, for example Workload Management or Data Distribution, are usually deployed close to the corresponding services. They watch for changes in the system state and react accordingly by initiating actions like job submission or result retrieval.

Simplest Agent

An agent essentially loops over and over executing the same function every X seconds. It has essentially two methods, initialize and execute. When the agent is started it will execute the initialize method. Typically this initialize method will define (amongst other stuff) how frequently the execute method will be run. Then the execute method is run. Once it finishes the agent will wait until the required seconds have passed and run the execute method again. This will loop over and over until the agent is killed or the specified amount of loops have passed.

Creating an Agent is best illustrated by the example below which is presenting a fully functional although simplest possible agent:

""" :mod: SimplestAgent

   Simplest Agent send a simple log message
"""

# # imports
from DIRAC import S_OK, S_ERROR
from DIRAC.Core.Base.AgentModule import AgentModule
from DIRAC.Core.Base.Client import Client


class SimplestAgent(AgentModule):
    """
    Simplest agent
    print a message on log
    """

    def initialize(self):
        """agent's initialisation

        :param self: self reference
        """
        self.message = self.am_getOption("Message", "SimplestAgent is working...")
        self.log.info("", self.message)
        return S_OK()

    def execute(self):
        """execution in one agent's cycle

        :param self: self reference
        """
        self.log.info("message is", self.message)
        simpleMessageService = HelloClient()
        result = simpleMessageService.sayHello(self.message)
        if not result["OK"]:
            self.log.error("Error while calling the service", result["Message"])
            return result
        self.log.info("Result of the request is", result["Value"])
        return S_OK()

First comes the documentation string describing the service purpose and behavior. Several import statements will be clear from the subsequent code.

The Agent name is SimplestAgent. The initialize method is called once when the Agent is created. Here one can put creation and initialization of the global variables if necessary. Please not that the __init__ method cannot be used when developing an Agent. It is used to intialize the module before it can be used

Now comes the definition of the execute method. This method is executed every time Agent runs. Place your code inside this method. Other methods can be defined in the same file and used via execute method. The result must always be returned as an S_OK or S_ERROR structure for the execute method. The previous example will execute the same example code in the Services section from within the agent.

Default Agent Configuration parameters

The Agent is written. It should be placed to the Agent directory of one of the DIRAC System directories in the code repository, for example FrameworkSystem. The default Service Configuration parameters should be added to the corresponding System ConfigTemplate.cfg file. In our case the Service section in the ConfigTemplate.cfg will look like the following:

Agents
{
  ##BEGIN SimplestAgent
  SimplestAgent
  {
    LogLevel = INFO
    LogBackends = stdout
    PollingTime = 60
    Message = still working...
  }
  ##END
}

‘PollingTime’ defines the time between cycles, ‘Message’ is this agent specific option. ##BEGIN SimplestAgent and ##END are used to automagically include the agent’s documentation into the docstring of the agents’ module, by placing this snippet there, see Component Options documentation

Installing the Agent

Once the Agent is ready it should be installed. As for the service part, we won’t do this part unless we want to mimic a full installation. Also, this part won’t work if we won’t have a ConfigurationServer running, which is often the case of a developer installation. For our development installation we can modify our local dirac.cfg in a very similar fashion to what we have done for the service part in the previous section, and run the agent using the dirac-agent command.

The DIRAC Server installation is described in documentation. If you are adding the Agent to an already existing installation it is sufficient to execute the following in this DIRAC instance:

$ dirac-install-agent Framework SimplestAgent

This command will do several things:

  • It will create the SimpleAgent Agent directory in the standard place and will set it up under the ‘’runit’’ control - the standard DIRAC way of running permanent processes.

  • The SimplestAgent Agent section will be added to the Configuration System.

The Agent can be also installed using the SystemAdministrator CLI interface:

$ install agent Framework SimplestAgent

The SystemAdministrator interface can also be used to remotely control the Agent, start or stop it, uninstall, get the Agent status, etc.

Checking the Agent output from log messages

In case you are running a SystemAdministrator service, you’ll be able to log in to the machine using (as administrator) dirac-admin-sysadmin-cli and show the log of SimplestAgent using:

$ show log Framework SimplestAgent

An info message will appear in log:

Framework/SimplestAgent  INFO: message: still working...

Note that the service is always returning the result in the form of S_OK/S_ERROR structure.